Radeon rx 480 8gb benchmarks in games. Video cards. New display controller

AMD Radeon RX 480 8GB Review | Meet Polaris 10

Eight months ago, AMD began to unleash the power of next-generation GPUs, starting with an updated display controller that supports HDMI 2.0b and DisplayPort 1.3 HBR3, FreeSync over HDMI, and an HDR-compatible pipeline. Later, additional information began to appear, which talked about the release of two different GPUs, one of which was designed specifically for the mainstream desktop market, and the other for mobile solutions that offer console-level performance in thin and light form factors.

The second product includes 16 Compute Units (CU), 128-bit memory bus and accelerated 4K video encoding / decoding. It is not available yet. Video card AMD Radeon RX 480 uses the larger Polaris 10 processor design. It is no larger than an Nvidia GP100 processor with 15.3 billion transistors in terms of physical size, but it is capable of driving the best virtual reality headsets. In terms of performance, the card is on a par with the AMD Radeon R9 290 and Nvidia GeForce GTX 970.

The card's mid-range performance is hardly eye-popping, especially when compared to the new Nvidia GP104 GPU. but AMD Radeon RX 480 it costs significantly less than solutions similar in speed, and power consumption is limited to 150 watts. Thus, AMD expects to make virtual reality available to a wider audience of gamers (it would be good if companies that sell HMDs for $ 800 and $ 600 would play along).

Two versions available AMD Radeon RX 480: $ 200 (MSRP) model with 4GB GDDR5 video memory at 7Gbps and $ 240 (MSRP) version with 8GB GDDR5 at 8Gbps. Today we are testing an 8GB model.

Features of Polaris 10

Polaris 10 consists of 5.7 billion transistors on a 230 mm2 chip. In comparison, a Hawaii crystal has 6.2 billion transistors and an area of ​​438 mm2. Despite the fewer transistors and about 55% lower power consumption, the RX 480 sits between the R9 290 and 390 in most tests. This is largely due to the 14nm FinFET process from GlobalFoundries, which gives AMD significant performance and power advantages over AMD. planar transistors manufactured using the 28 nm process technology. FinFET provides a higher frequency at any level of power consumption, and vice versa, at any clock frequency, a chip with 14 nm consumes less power. In the case of Polaris, AMD took advantage of both by increasing clock speeds and lowering power consumption. This way it managed to outperform the more powerful GPU Hawaii while maintaining a power ceiling of 150 watts (although our measurements show that this figure is a little underestimated).

Despite the new codename, the Polaris 10 is based on the 4th generation AMD Graphics Core Next architecture. Therefore, the building blocks of Polaris processor design will seem familiar to many enthusiasts, and it will be easier for us to describe it.

Specifications

	AMD Radeon RX 480	AMD Radeon R9 390	AMD Radeon R9 290
	Polaris 10	Grenada pro	Hawaii pro
Computing Units (CU)	36	40	40
Stream Processors	2304	2560	2560
Clock frequency (base / Boost), MHz	1120/1266	1000	947
Peak computation rate, GFLOPs (at base frequency)	5161	5120	4849
Number of texture blocks	144	160	160
Speed ​​up texture fill Gtex / s	182,3	160	160
Number of ROP units	32	64	64
L2 cache size, MB	2	1	1
Memory data transfer rate, Gb / s	8 (8GB) / 7 (4GB)	6	5
Memory bandwidth, GB / s	256	384	320
Memory bus, bit	256	512	512
Thermal package, W	150	275	250
Number of transistors, billion	5,7	6,2	6,2
Crystal area, mm2	230	438	438
Process technology, nm	14	28	28
starting price	$ 240 (8GB) / $ 200 (4GB)	$ 330 (8 GB)	$ 400 (4 GB)

One GCP-Graphics Command Processor is still responsible for dispatching the sequence of graphics instructions to the Shader Engine. Asynchronous Compute Engine (ACE) takes care of the sequence of computational instructions. Only instead of eight ACEs, the command execution logic now consists of four ACEs and two Hardware Schedulers that perform the tasks of prioritizing queues, managing temporal / spatial resources, and offloading scheduling tasks for the CPU kernel-mode driver. In essence, these are not separate or new blocks, but rather an additional mode in which existing pipelines can operate. Dave Nalasco, AMD's senior manager of graphics workflows, made the following comment:

"Hardware Workgroup / Wavefront Schedulers (HWS) are essentially ACE pipelines without dispatch controllers. Their job is to offload the CPU by controlling the scheduling process for user / driver-defined queues on the available hardware queue slots. These are programmable processors with microcode that can different scheduling policies applied. We used them to implement Quick Response Queue and CU Reservation. We've also been able to port these changes to 3rd generation GCN graphics cards through driver updates. "

The Quick Response Queues feature allows developers to prioritize certain tasks that run asynchronously without completely preempting other processes. A more detailed explanation can be found on Dave's blog(English). In short, AMD wants flexibility. Its architecture allows for different approaches to optimize resource utilization and minimize rendering latency, both of which are extremely important for VR applications.

The well-known CU computational units consist of 64 shader units, compliant with the IEEE 754-2008 standard, divided into four vector units, a scalar unit, and 16 texture sample loading / storage units. In addition, each CU includes four texturing units, 16KB L1 cache, 64KB of local space for data exchange, and register space for vector and scalar units. AMD claims to have made many adjustments to improve CU efficiency, including adding support for FP16 (and Int16), optimizing cache access, and improving instruction lookahead. Taken together, these changes deliver up to 15% better CU performance over Hawaii GPUs (2nd Gen GCN).

Nine CUs form a large shader unit (SE - Shader Engine). The Polaris 10 video chip has four such SEs and we know that this is the maximum for this architecture. In total, we get 2304 stream processors and 144 texture units (64 shaders x 9 CU x 4 SE).

Each shader unit is associated with a geometry unit (GE - Geometry Engine). According to AMD, a primitive discard accelerator has been added to the geometry block, it filters out the simplest geometric elements that are not rasterized into a pixel before scanning transformation, thus increasing the throughput. This is an automatic feature of the pre-rasterization stage of the graphics pipeline and is new to Polaris. In addition, an index cache for cloned geometry has appeared, although we do not know its size and the degree of influence during cloning.

By analogy with the Hawaii video chip, the Polaris 10 processor is capable of rendering four simple elements per cycle. However, compared to Hawaii / Grenada GPUs up to 1050 MHz (in the case of the R9 390X), AMD has raised the base clock AMD Radeon RX 480 up to 1120 MHz, and the frequency in Boost mode is up to 1266 MHz. It turns out that the company compensates for the loss of resources on the crystal with an increased frequency. The single precision floating point performance of the Radeon R9 290X is 5.6 TFLOPS, while the RX 480 reaches 5.8 TFLOPS in Boost mode.

How realistic is the 1266 MHz clock speed? The Hawaii GPU had a hard time keeping up with the specs as it got very hot, and we wanted to make sure that didn't happen with Polaris. Using GPU-Z, we took the clock speed readings in the integrated benchmark of the Metro: Last Light Redux game, repeated 10 times in a row, and received the following graph:

Stress Test Clock - Built-in Metro: Last Light Redux Benchmark, 10 passes, MHz

The difference between the upper (1265 MHz) and lower (1118 MHz) points on the graph is 148 MHz. We can say that AMD fits well into the indicated limits, although the frequency is constantly being adjusted during the test. But at least the 1208 MHz average is closer to the top.

The Hawaii and Fiji SE GPUs each have four rendering backends capable of outputting 16 pixels per clock (64 pixels per clock in total). The Polaris 10 has cut this component in half. Each SE has two render backends, each with four ROPs, which collectively render 32 pixels per clock. The difference with the Hawaii-based Radeon R9 290 is quite significant. The situation is aggravated by the 256-bit memory bus Polaris 10, which is twice as narrow as the memory bus of the Hawaii video chip (512-bit). Version AMD Radeon RX 480 The 4GB uses 7Gbps GDDR5 memory and has 224GB / s bandwidth, while the 8GB model we are testing today uses 8Gbps memory and bandwidth increased to 256 GB / s. But in any case, this is much less than the 320 GB / in the R9 290.

The reduction in hardware resources is partially offset by improved delta color compression, which reduces the amount of information transmitted over the bus. AMD also supports lossless 2/4/8: 1 compression, just like Nvidia's Pascal architecture. In addition, the Polaris 10 uses a 2MB L2 cache, the same size used by Fiji. This will reduce the number of GDDR5 memory accesses and further reduce the GPU's dependence on a wide bus and high data transfer rate.

However, the depletion of the GPU backend should affect performance as the resolution and intensity of anti-aliasing increases. We wondered how the Polaris would look against the Hawaii as the intensity increased. To test this, we ran the Grand Theft Auto V benchmark at a modest 1920x1080 resolution with "Very High" graphics detail settings and gradually increased the anti-aliasing quality.

The graph clearly shows that when the MSAA anti-aliasing is changed from 2x to 4x AMD Radeon RX 480 loses average frame rate noticeably faster than R9 390. With anti-aliasing disabled, RX 480 reaches 97.3 FPS, and R9 390 - 90.4 FPS. But towards the end of the graph AMD Radeon RX 480 showed only 57.5 frames per second, while the 390th averaged 62.9 frames per second.

AMD Radeon RX 480 8GB Review | Display Controller, UVD, VCE & WattMan

New display controller

We've already covered some of the Polaris display controller improvements in this article. "Plans for the functional development of GPU AMD in 2016"... But it was published almost seven months ago.

At the time, we knew that Polaris would support DisplayPort 1.3 with High Bit Rate 3, using existing cables and connectors to deliver up to 32.4 Gbps over four lanes. The controller specification now includes the DisplayPort 1.4-HDR standard. It does not increase the bit rate, but it includes Display Stream Compression 1.2 technology to deliver 10-bit 4K content at 96Hz refresh rate. Also, the DisplayPort 1.4 standard supports the color space.

In the short term, AMD is still looking at DP 1.3 as a tool for implementing FreeSync in 4K. According to the company, panels with a refresh rate of 120 Hz will be available by the end of 2016, but in order to achieve good performance with high graphics settings in this configuration, the possibilities AMD Radeon RX 480 will not be enough. At the same time, the Vega processor design with HBM2 support will not officially appear until 2017.

We discussed HDR support in Polaris late last year, but AMD reiterates that the display pipeline is ready for the first generation of 10-bit HDR displays, and 12-bit HDR displays in the future. An easy-to-program color processing block includes gamma remapping, gamma control, floating point processing, and 1: 1 projection with any display.

Video encoding / decoding acceleration

During its heyday, ATI was known for its performance and quality video decode acceleration systems that shifted video playback tasks from the central processor to a combination of programmable shaders and fixed function blocks installed in the GPU.

We do not have details on where the Polaris decoder completes its tasks, but it is known that it is based on the UVD decoder and appears to have fixed functionality. AMD specifies in the specs to have HEVC decoding up to 4K60 using the Main 10 profile, which supports 10-bit 4: 2: 0 (all of which are required for HDR to work). There is hardware support for VP9 decoding, although AMD drivers have not yet implemented it, we only know that the feature is planned in a future update. If AMD wants to implement HEVC 10-bit / 4: 2: 0 color downsampling with HDR, at least Profile 2 compatibility is required. Hardware acceleration of the M-JPEG format up to 4K30 is also provided.

The evolution of the AMD Video Encoder (VCE - Video Coding Engine) is also not well documented. Polaris is known to be able to encode HEVC 8-bit video up to 4K60, but GPUs based on GCN 1.2 architecture have the same equipment. It looks like AMD is working on expanding the list of VCE-compatible applications. Of course, the proprietary Gaming Evolved client is supported. But apart from that, the lists include Open Broadcaster Software, which previously only supported QuickSync and NVEnc. There is also Plays.tv, a social network from the company in charge of the Gaming Evolved client.

At the end of June this year, AMD announced a line of new Polaris 10 and Polaris 11 GPUs, which are based on the most advanced 14nm FinFET process technology. At the moment, among the video cards released on the new graphics processors, there are three models: AMD Radeon RX 480, AMD Radeon RX 470 and AMD Radeon RX 460. In today's article we will conduct a brief overview of the reference model of the older video card and its comprehensive testing.

Since all the theoretical calculations about the architecture of the new Polaris GPUs have long been published by other resources, today we will not touch on this topic. Just briefly note that the main innovations in the updated architecture of the fourth generation Graphics Core Next concerned improved processing of geometry and blocks of encoding and decoding of video data, support for asynchronous computations in DirectX 12, support for the Vulkan API, more efficient data compression methods, increased energy efficiency, support for DisplayPort 1.4 video outputs. -HDR and HDMI 2.0b and much more.

1. Review of the AMD Radeon RX 480 8 GB video card

technical characteristics and recommended cost

The technical characteristics and cost of the AMD Radeon RX 480 video card are shown in the table in comparison with the reference AMD Radeon R9 390, Radeon R9 380X and NVIDIA GeForce GTX 1060.

* - according to Yandex.Market data as of 09/15/2016.

PCB design and features

The AMD Radeon RX 480 reference design is virtually indistinguishable from the design of the Radeon R9 Fury X and Nano, presenting a simple yet stylish 244 x 102 x 38mm graphics card. The entire front side is covered with a plastic casing with a structure of small round cells, and a large RADEON inscription is stamped on the left.

It is also visible in the upper part of the casing.

Combined with a similarly styled fan rotor, the reference Radeon RX 480's design looks complete and austere.

The video output panel has three DisplayPort version 1.4 and one HDMI version 2.0b.

As you can see, most of this panel is occupied by a grille for the unobstructed passage of heated air outside the case of the system unit. And the video card heats up very much, it should be noted.

To power the Radeon RX 480, there is one six-pin connector located at the top of the casing. The declared power consumption level of the video card is 150 watts, and for a system with one such video card, a 500-watt power supply is recommended. The power section of the printed circuit board is made according to a seven-phase scheme, where six phases are allocated to power the graphics processor and one to the video memory.

The new 14nm Polaris 10 XT GPU contains around 5.7 billion transistors and contains 2,304 unified shaders, 144 texture units and 32 raster operations (ROPs).

The frequency of the GPU in 3D mode should vary in the range from 1120 to 1266 MHz, but in practice it was far from always the case, which we will discuss below.

The Radeon RX 480 can be equipped with either 4 GB ($ 199) or 8 GB ($ 229) of video memory. Our copy of the video card had 8 GB of DDR5 memory with Samsung chips (according to GPU-Z).

The effective frequency of the video memory is 8000 MHz, which with a 256-bit bus can provide a bandwidth of 256 GB / s. This is immediately 33% higher than the main competitor NVIDIA GeForce GTX 1060 with its 192-bit bus (192.2 GB / s).

cooling system

From a practical point of view, it makes no sense to evaluate the efficiency of the standard cooler of the reference version of AMD Radeon RX 480, since by now original versions with branded coolers have appeared on the market. But since they have not reached us yet, there is no choice, so today we will test the standard cooler, which is a combination of an aluminum radiator with a copper base for the GPU, a metal heat-distributor plate for power circuits and a turbine that pumps air through the radiator.

The entire system is covered with a plastic casing that directs the air heated by the video card to the panel with video outputs and a grill. The turbine speed is regulated by PWM in the range from 1200 to 4960 rpm.

We used nineteen cycles of the 3DMark stress test to test the temperature conditions of the video card as a load.

Since we did not yet have an updated version of MSI Afterburner at the time of writing this article, the GPU-Z utility version 1.9.0 was used to monitor temperatures. All tests were carried out in a closed case of the system unit, the configuration of which you can see in the next section of the article, at an average room temperature 25 degrees Celcius.

First, we checked the temperature mode of the video card with fully automatic fan speed control.

Automatic mode (1200 ~ 2450 rpm)

The temperatures are very high, it is obvious that the stock cooler, even after overclocking to 2450 rpm, is simply not suitable for ensuring the operation of the Radeon RX 480 at the standard maximum frequency of 1266 MHz, since during testing it dropped down to 1000 MHz, and on average “floated” "At the 1050-1070 MHz mark.

At the maximum possible fan speed of the cooler, the peak temperature of the processor is 12 degrees Celsius lower, due to which the GPU frequency does not jump as much as with automatic adjustment.

Maximum speed (~ 4960 rpm)

Remarkably, the percentage of stability of the video card in the 3DMark stress test also increased from 87.6% to 97.8%.

Therefore, we can conclude that in order to ensure stable operation of the Radeon RX 480 and maintain the frequency of its GPU at a high level (and hence performance), it needs effective cooling as never before, despite the new 14nm process technology.

As for overclocking, for obvious reasons we did not study it on the reference video card. Let's hope that the original Radeon RX 480 models will allow us to fully disclose this issue and get acquainted with AMD's proprietary WattMan technology, implemented simultaneously with the appearance of the Radeon RX 480.

2. Test configuration, tools and testing methodology

Performance testing of video cards was carried out in a closed case on a system with the following configuration:

motherboard: ASUS Sabertooth X79 (Intel X79 Express, LGA2011, BIOS 4801 dated 07/28/2014);
CPU: Intel Core i7-3970X Extreme Edition 3.5 / 4.0 GHz(Sandy Bridge-E, C2, 1.1 V, 6 x 256 KB L2, 15 MB L3);
CPU cooling system: ARCTIC Liquid Freezer 240 (4 x 1100 rpm);
thermal interface: ARCTIC MX-4;
video cards:

Inno3D iChill GF GTX 980 Ultra HerculeZ X4 Air Boss 4 GB 1266-1367 / 7200 MHz;
Sapphire NITRO R9 390 OC Tri-X 8 GB 1040/6000 MHz;
NVIDIA GeForce GTX 1060 Founders Edition 6 GB 1506-1708 (1886) / 8008 MHz;
AMD Radeon RX 480 8 GB 1120-1266 / 8000 MHz;
ASUS GeForce GTX 970 DC Mini 4 GB 1050-1178 / 7012 MHz (GTX970-DCMOC-4GD5);
ASUS STRIX R9 380X Gaming 4 GB 1030/5700 MHz;

RAM: DDR3 4 x 8 GB G.SKILL TridentX F3-2133C9Q-32GTX(X.M.P. 2133 MHz, 9-11-11-31, 1.6 V);
system and game disk: Intel SSD 730 480GB (SATA-III, BIOS vL2010400);
disk for storing programs and games: Western Digital VelociRaptor (SATA-II, 300 GB, 10,000 rpm, 16 MB, NCQ);
archive disk: Samsung Ecogreen F4 HD204UI (SATA-II, 2 TB, 5400 rpm, 32 MB, NCQ);
sound card: Auzen X-Fi HomeTheater HD;
case: Thermaltake Core X71 (four be quiet! Silent Wings 2 (BL063) at 900 rpm);
control and monitoring panel: Zalman ZM-MFC3;
PSU: Corsair AX1500i Digital ATX (1500W, 80 Plus Titanium), 140mm fan.
monitor: 27-inch Samsung S27A850D (DVI, 2560 x 1440, 60 Hz).

As top benchmarks for AMD Radeon RX 480 performance, we included NVIDIA's original Inno3D iChill GF GTX 980 Ultra HerculeZ X4 Air Boss and AMD's original Sapphire NITRO R9 390 OC Tri-X in our benchmarks.

The direct competitor to the heroine of today's testing will be NVIDIA GeForce GTX 1060, which is represented by the reference version of the Founders Edition. Next to it in the photo is the ASUS STRIX R9 380X Gaming, which, following the marking of video cards in the AMD line, is being replaced by the new Radeon RX 480.

And finally, the fifth video card in testing is the ASUS GeForce GTX 970 DC Mini, which, oddly enough, today is barely cheaper than the Radeon RX 480, which means it can also hypothetically compete with it.

Let's add that the Power Limit value on all video cards was set to the maximum.

To reduce the dependence of the performance of video cards on the platform speed, a 32-nm six-core processor with a multiplier of 48, a reference frequency of 100 MHz and the Load-Line Calibration function activated at the Ultra High level was overclocked to 4.8 GHz when the voltage in the BIOS of the motherboard rises to 1.385 V.

Hyper-Threading Technology is activated. At the same time, 32 gigabytes of RAM operated at a frequency of 2.133 GHz with timings 9-11-11-20_CR1 at a voltage of 1.6125 V.

Testing, which began on August 8, 2016, was conducted under the operating system Microsoft Windows 10 Professional with all updates as of the specified date and with the following drivers installed:

motherboard chipset Intel Chipset Drivers - 10.1.1.27 WHQL from 06.07.2016;
Intel Management Engine Interface (MEI) - 11.5.0.1019 WHQL from 08/09/2016;
drivers for video cards on graphics processors NVIDIA - GeForce 369.05 WHQL from 04/08/2016;
video card drivers on AMD GPUs - AMD Radeon Software Crimson 16.8.1 WHQL from 08/07/2016.

The performance of the graphics cards was tested at resolutions of 1920 x 1080 and 2560 x 1440 pixels. Two graphics quality modes were used for the tests: Quality + AF16x - texture quality in the drivers by default with anisotropic filtering at 16x and Quality + AF16x + MSAA 4x (8x) with anisotropic filtering at 16x and full-screen anti-aliasing at 4x or 8x, in cases when average frames per second remained high enough for comfortable gaming. In some games, due to the specifics of their game engines, other anti-aliasing algorithms were used, which will be indicated later in the methodology and in the diagrams. Anisotropic filtering and full-screen anti-aliasing was enabled directly in the game settings. If these settings were absent in games, then the parameters were changed in the control panel of the Crimson or GeForce drivers. V-Sync was also forcibly disabled there. Apart from the above, no additional changes were made to the driver settings.

The video cards were tested in one semi-synthetic graphics test and in fifteen games, updated to the latest versions as of the date of the preparation of this material. The list of test applications is as follows (games and further test results in them are arranged in the order of their official release):

3DMark(DirectX 9/11/12) - version 2.1.2852, tested in Fire Strike, Fire Strike Extreme, Fire Strike Ultra and Time Spy scenes;
Crysis 3(DirectX 11) - version 1.3.0.0, all graphics quality settings to maximum, the degree of blur is medium, glare is on, modes with FXAA and with MSAA 4x, double sequential pass of a scripted scene from the beginning of the Swamp mission lasting 105 seconds;
Metro: Last Light(DirectX 11) - version 1.0.0.15, the built-in game test, graphics quality settings and tessellation at the Very High level, Advanced PhysX technology in two test modes, tests with SSAA and without anti-aliasing, double sequential D6 scene run were used;
Company of Heroes 2(DirectX 11) - version 4.0.0.21543, double sequential run of the test built into the game with the maximum settings for graphics quality and physical effects;
Battlefield 4(DirectX 11) - version 1.2.0.1, all graphics quality settings on Ultra, double sequential run of the scripted scene from the beginning of the TASHGAR mission lasting 110 seconds (for video cards based on AMD GPUs, the Mantle API was used);
Thief(DirectX 11) - version 1.7 build 4158.21, graphics quality settings to the maximum level, Paralax Occlusion Mapping and Tessellation technologies are activated, a double sequential run of the benchmark built into the game (API Mantle was used for video cards based on AMD GPUs);
Sniper elite iii(DirectX 11) - version 1.15a, quality settings at Ultra, V-Sync disabled, tessellation and all effects enabled, tests with SSAA 4x and without anti-aliasing, double sequential run of the benchmark built into the game (for video cards based on AMD GPUs API Mantle was used );
(DirectX 11) - build 1951.27, all quality settings are manually set to the maximum and Ultra levels, tessellation and depth of field are activated, at least two consecutive runs of the benchmark built into the game;
Grand theft auto v(DirectX 11) - build 757.4, quality settings at the Very High level, ignoring the proposed restrictions enabled, V-Sync disabled, FXAA enabled, NVIDIA TXAA disabled, MSAA for reflections disabled, NVIDIA / AMD soft shadows;
DiRT Rally(DirectX 11) - version 1.2, we used the built-in test on the Okutama track, graphics quality settings to the maximum level for all points, Advanced Blending - On; tests with MSAA 8x and without anti-aliasing;
Batman: arkham knight(DirectX 11) - version 1.6.2.0, quality settings at High, Texture Resolutioin normal, Anti-Aliasing on, V-Sync disabled, tests in two modes - with and without activation of the last two NVIDIA GameWorks options, dual sequential run of the built-in into a dough game;
(DirectX 11) - version 3.1, texture quality settings at the Very High level, Texture Filtering - Anisotropic 16X, and other maximum quality settings, tests with MSAA 4x and without anti-aliasing, double sequential run of the test built into the game.
Rise of the Tomb Raider(DirectX 12) - version 1.0 build 668.1_64, all parameters at the Very High level, Dynamic Foliage - High, Ambient Occlusion - HBAO +, tessellation and other quality improvement techniques are activated, two test cycles of the built-in benchmark (Geothermal Valley scene) without anti-aliasing and with SSAA 4.0 activation;
Far cry primal(DirectX 11) - version 1.3.3, maximum quality level, high-resolution textures, volumetric fog and shadows to maximum, built-in performance test without anti-aliasing and with SMAA activation;
Tom clancy's the division(DirectX 11) - version 1.3, maximum quality level, all picture enhancement parameters are activated, Temporal AA - Supersampling, testing modes without anti-aliasing and with activation of SMAA 1X Ultra, built-in performance test, but fixing FRAPS results;
Hitman(DirectX 12) - version 1.2.2, built-in test with graphics quality settings at the Ultra level, SSAO enabled, shadow quality Ultra, memory protection disabled.

If the games implemented the ability to fix the minimum number of frames per second, then it was also reflected in the diagrams. Each test was carried out twice, the best of the two obtained values ​​was taken as the final result, but only if the difference between them did not exceed 1%. If the deviations of the test runs exceeded 1%, then the testing was repeated at least one more time in order to obtain a reliable result.

3. Results of performance tests and their analysis

On the diagrams, the test results for video cards on NVIDIA GPUs are highlighted in green, and on AMD GPUs they are reflected in the usual red color scheme for this manufacturer. To highlight the performance of the Radeon RX 480, we chose a dark red fill color. Let's add that on the diagrams in each quality mode, the test results are sorted from top to bottom in descending order of the cost of video cards.

3DMark

In almost all 3DMark test scenes, the performance of video cards confirms their cost, clearly placing the products from top to bottom. Only in the Time Spy test the density of the results is higher. AMD Radeon RX 480 is at the level of ASUS GeForce GTX 970, slightly lagging behind its direct competitor NVIDIA GeForce GTX 1060 and noticeably ahead of ASUS STRIX R9 380X Gaming. Obviously, the performance of the Sapphire NITRO R9 390 OC Tri-X is beyond the reach of the heroine of today's article even when overclocked.

Crysis 3

Crysis 3 showed us a different picture.

Here AMD Radeon RX 480 does not look so confident anymore, yielding even to ASUS GeForce GTX 970 from the past NVIDIA lineup. The advantage of the new item over the ASUS STRIX R9 380X Gaming is not at all impressive, and the difference with the Sapphire NITRO R9 390 OC Tri-X is too great. Unfortunately, there can be no question of any struggle with NVIDIA GeForce GTX 1060.

Metro: Last Light

Recall that we tested the Metro: Last Light game both with Advanced PhysX activation and without it.

However, disabling Advanced PhysX did not help at all for AMD video cards today - the competitors turned out to be much stronger. The advantage of AMD Radeon RX 480 over ASUS STRIX R9 380X Gaming here ranges from 16 to 28%, and the backlog from Sapphire NITRO R9 390 OC Tri-X is from 2 to 24%.

Company of Heroes 2

In Company of Heroes 2, the alignment of forces does not differ much from Metro: Last Light - video cards based on AMD GPUs are inferior to their competitors based on NVIDIA chips.

AMD Radeon RX 480 is losing even ASUS GeForce GTX 970 here, what can we say about NVIDIA GeForce GTX 1060, which, in turn, successfully fights with Sapphire NITRO R9 390 OC Tri-X?

Battlefield 4

The situation in Battlefield 4 is even worse for video cards with AMD GPUs.

AMD Radeon RX 480 was only able to demonstrate a slight advantage over the ASUS STRIX R9 380X Gaming, but even the ASUS GeForce GTX 970 was too tough for her, not to mention the GeForce GTX 1060.

Thief

Things are much better for AMD in the game Thief, which uses the Mantle API.

Despite the lack of obvious performance gaps, the AMD Radeon RX 480 competes only with the ASUS GeForce GTX 970, slightly outperforming the ASUS STRIX R9 380X Gaming. In turn, the NVIDIA GeForce GTX 1060 not only convincingly beats the AMD Radeon RX 480, but is also able to withstand the more expensive Sapphire NITRO R9 390 OC Tri-X.

Sniper elite iii

The results of testing video cards in the game Sniper Elite III are very dependent on the quality mode, namely the activation of SSAA 4.0.

Nevertheless, even here we cannot call the performance of AMD Radeon RX 480 convincing, since the advantage over ASUS STRIX R9 380X Gaming is absolutely insignificant, and there is no need to talk about rivalry with NVIDIA GeForce GTX 1060.

Middle-earth: Shadow of Mordor

Here the performance of the AMD Radeon RX 480 is higher than that of the ASUS STRIX R9 380X Gaming by 4-26%, although this only applies to the resolution of 1920 x 1080 pixels, since in the large 2560 x 1440 pixels the novelty is ahead of its predecessor by only a couple of average frames per second, and the minimum FPS for the AMD Radeon RX 480 is even slightly lower. The NVIDIA GeForce GTX 1060 is way ahead of both of them, as is the Sapphire NITRO R9 390 OC Tri-X.

Grand theft auto v

In the game Grand Theft Auto V, we can see the picture that has already become familiar to today's testing.

And yet, unlike previous benchmarks, here AMD Radeon RX 480 manages to outperform ASUS GeForce GTX 970 in modes without anti-aliasing and not even lag far behind NVIDIA GeForce GTX 1060 with Sapphire NITRO R9 390 OC Tri-X. When MSAA4x is turned on, we are only talking about the fight against the ASUS GeForce GTX 970 and the advantage over the ASUS STRIX R9 380X Gaming. No more, unfortunately.

DiRT Rally

In the dirt road racing simulator, the AMD Radeon RX 480 is on par with the ASUS GeForce GTX 970 and lags quite a bit behind the NVIDIA GeForce GTX 1060. As for the huge difference with the ASUS STRIX R9 380X Gaming, it is most likely not due to driver optimization for this video card, or the peculiarity of the latest game patches with Radeon R9 3xx video cards.

Batman: arkham knight

Batman: Arkham Knight was created with the support of NVIDIA, and it actively uses the graphics technologies of this company, but this fact did not prevent video cards based on AMD GPUs from confidently performing in these tests.

Yes, AMD Radeon RX 480 has once again lost out to NVIDIA GeForce GTX 1060, but this time not as much as in previous games. And the difference with ASUS STRIX R9 380X Gaming here is quite good 24-33%.

Tom Clancy's Rainbow Six: Siege

Rainbow Six: Siege was the first game where AMD Radeon RX 480 beat Sapphire NITRO R9 390 OC Tri-X to finally compete with NVIDIA GeForce GTX 1060.

Its difference with ASUS STRIX R9 380X Gaming is also impressive, reaching 48% in one of the quality modes. In addition, the ASUS GeForce GTX 970 was finally defeated with a good margin. In general, the first game justifying the release of AMD Radeon RX 480. Unfortunately, the holiday did not last long - already in Rise of the Tomb Raider everything returned to square one.

Rise of the Tomb Raider

Support for the game Rise of the Tomb Raider API DirectX 12, it would seem, should help AMD Radeon RX 480, but the results indicate the opposite - the new product is still losing to its main competitor.

But in modes without anti-aliasing, AMD Radeon RX 480 is quite confidently ahead of ASUS STRIX R9 380X Gaming, and when AA is activated, the frame rate is so low, it makes no difference which of these two video cards to choose.

Far cry primal

Far Cry Primal very clearly places video cards in terms of performance, based on their cost, especially in the most resource-intensive quality mode.

AMD Radeon RX 480 is 14-23% faster than ASUS STRIX R9 380X Gaming in this game, and 8-11% slower than NVIDIA GeForce GTX 1060.

Tom clancy's the division

Except for the abnormally high results of the Sapphire NITRO R9 390 OC Tri-X, the rest of the graphics cards' ranking in performance in Tom Clancy's The Division is not out of the ordinary.

Nevertheless, we note that in this game AMD Radeon RX 480 lags behind NVIDIA GeForce GTX 1060 by a few percent.

Hitman

The latest version of Hitman is a celebration on AMD's red streets, as it was in this game that GPUs powered by Polaris and Grenada managed to outpace their competitors on GPUs Pascal and Maxwell 2.0.

We add that on ASUS GeForce GTX 970 at 2560 x 1440 pixels using the maximum anti-aliasing mode, the test ended with an error, so there is no result for this video card in this quality mode.

Let's supplement the constructed diagrams with a summary table with test results with the displayed average and minimum value of the number of frames per second for each video card.

In addition to the gaming tests, today we will present the results of testing two competing video cards in the CompuBench CL 1.5 benchmark.

AMD Radeon RX 480 4 GB NVIDIA GeForce GTX 1060 6 GB

4. Pivot charts

On the first pair of summary diagrams, we will evaluate the difference in performance between AMD Radeon RX 480 and its predecessor Radeon R9 380X represented by ASUS STRIX R9 380X Gaming, the results of which in each game are taken as the starting point of reference, and the average FPS of the heroine of today's testing is postponed as a percentage of them.

Basically, the AMD Radeon RX 480 demonstrates a good performance gain over the Radeon R9 380X in almost all games. Except for the abnormally low results of the Radeon R9 380X in the DiRT Rally game, Hitman is especially indicative in this regard, where the Radeon RX 480, thanks to a doubled memory size and a faster graphics processor, is 62 to 83% ahead of its predecessor. On average across all games, the Radeon RX 480 is 27-31% faster.

Next, let's check how the Radeon RX 480 looks against the background of the Sapphire NITRO R9 390 OC Tri-X with the same amount of video memory, but the old Hawaii GPU. By the way, now the cost of the original versions of the Radeon R9 390 has dropped to the level of the new Radeon RX 480, so such a comparison will be quite appropriate and relevant.

Well, we can see how the Radeon RX 480 was unable to defeat the Radeon R9 390. The only exceptions were Rainbow Six: Siege and the anti-aliasing mode in Hitman. On average, in all tests, the novelty lags behind by 10-11% at a resolution of 1920 x 1080 pixels and by 14-15% at a resolution of 2560 x 1440 pixels.

Finally, the most important and interesting pair of pivot charts: comparing the performance of AMD Radeon RX 480 and NVIDIA GeForce GTX 1060 - two graphics cards that were released to face off against each other two weeks apart.

The advantage of a video card with an NVIDIA GPU is obvious, except, again, Hitman. On the whole, one cannot fail to note the tendency that when moving from older to newer games (from top to bottom), the performance of video cards is leveled, and the Radeon RX 480 does not at all look like a "whipping boy", as it seemed at first. Nevertheless, in our test suite of games it turned out that, on average, the Radeon RX 480 lags behind the GeForce GTX 1060 by 13.7-14.7% at a resolution of 1920 x 1080 pixels and by 14.1-15.0% at a resolution of 2560 x 1440 pixels.

5. Power consumption

The energy consumption was measured using a Corsair AX1500i power supply via the Corsair Link interface and the HWiNFO64 monitoring software version 5.35-2950. The energy consumption of the entire system as a whole was measured without taking into account the monitor. The measurement was carried out in 2D mode during normal work in Microsoft Word or Internet surfing, as well as in 3D mode. In the latter case, the load was created using four consecutive cycles of the Swamp-level intro scene from Crysis 3 at 2560 x 1440 pixels at maximum graphics quality settings and using MSAA 4X. Let us add that the diagram shows both the peak level of power consumption in 3D mode and the average consumption value for the entire testing cycle.

Let's compare the level of power consumption of systems with the video cards tested today according to the diagram.

The power consumption of the system with the AMD Radeon RX 480 video card did not exceed the consumption of the configuration with the Radeon R9 380X and turned out to be significantly lower than with the Radeon R9 390 video card. However, in comparison with the system in which the GeForce GTX 1060 is installed, the new product loses quite a lot for one class of video cards. opposed to each other. So, if at the peak of the load the configuration with the GeForce GTX 1060 consumes only 461 watts, then with the Radeon RX 480 it already 518 watts, which is 12.3% more. In terms of average power consumption, the picture is almost the same, and in 2D NVIDIA is even more economical than AMD. Of course, the level of power consumption of video cards is not a determining factor when choosing them, but we cannot fail to note that in this indicator AMD is inferior to its eternal competitor.

Conclusion

Summing up today's material, we can briefly summarize that at the moment AMD Radeon RX 480 is inferior in performance to NVIDIA GeForce GTX 1060 by about 14-15%, but in the most modern games that support DirectX 12, the difference between these video cards is reduced. Therefore, we can assume that Polaris still has prospects. In terms of power consumption, AMD also lost this round to NVIDIA - at the moment the reference GeForce GTX 1060 is more economical than the Radeon RX 480. As for the overclocking potential of both video cards, we will draw conclusions after checking the original models with reinforced printed circuit boards and effective cooling systems. In addition, in the near future, the test suite will include two more new games with DirectX 12 support, which may also affect the balance of power between AMD and NVIDIA in this class of video cards. In terms of retail price, these video cards are now almost the same, so the choice, as always, is yours.

Thanks to AMD for
the video card provided for testing.

Video card AMD Radeon RX 480 became a hit among a huge number of users after a big PR campaign in which the manufacturer promised a fairly high performance, which is close to the GTX 970 and R9390, for a relatively low cost of $ 229 for 8Gb and $ 199 for 4Gb.

Such characteristics did not go unnoticed and many potential buyers were impatiently waiting for the X-hour to get acquainted with the new product. The expectations were confirmed. The developers, as promised, created a truly interesting product that gained popularity, and the first batches were sold out very quickly.

The situation is a little worse with "non-references", which, even weeks later, have just begun to be delivered to stores.

But now we will not talk about them, but about the ancestor of the Polaris 10 line in the reference performance. The AMD Radeon RX 480 8Gb video card turned out to be more interesting than its predecessors due to the new 14-nm process technology, low power consumption, increased frequency potential and updated Crimson drivers, which introduced the Wattman overclocking utility.

Specification

• Manufacturer: AMD
• Model: Radeon RX 480
• GPU: Polaris 10;
• Process technology: 14 nm;
• GPU frequency: 1266 MHz;
• Number of shader processors: 2304;
• Video memory: 8 GB;
• Video memory type: GDDR5;
• Video memory bus width: 256bit;
• Video memory frequency: 2000 MHz (8.0 GHz QDR);
• CrossFire support: yes;
• Ports: HDMI, 3 xDisplayPort;
• Additional power connector: 6-pin;
• Length: 241mm;
• Price: 18500 rub.

Appearance and design

The AMD Radeon RX 480 8Gb video card is presented in a natural variation in the form of a "reference" with a familiar shape - a radial fan that drives air through the entire board and throws hot air through the rear grille, and a radiator hidden by an updated decorative casing that came to us from the Radeon Fury X model ...

The novelty boasts small dimensions: the length of the video card is 241 mm, the width is 112 mm. In the system unit, it will cover only two expansion slots. The developers have moved on to a new design that includes a lot of black. As the company said: even a red tint in the logo of the series frightened off potential buyers who associate it with high heating. Reystyling was good for AMD.

The downside of the graphics accelerator reveals a couple of interesting points. First, as we can see, the length of the PCB is much smaller than the entire video card and is 170 mm, which fits perfectly into the dimensions of ITX systems, allowing AMD Radeon RX 480 8Gb to be installed in Mini-ITX form factor cases. Second, a protective retention spider in the center of the GPU redistributes the load on the heatsink, protecting the chip from kinking damage. As for a novelty with a not very large cooling system, this is an extra reinsurance.

The side elements are hidden by dense casing walls that completely redirect hot air to the rear interface panel, where it exits and does not heat up the internal components of the PC.

Additional power is realized with just one six-pin connector, as if hinting to the user about low power consumption. However, this has become a stumbling block for overclockers and news that it can damage the PCI-Express connector on the motherboard. As the AMD developers themselves said, the installation of the 6-pin connector is justified by the fact that many budget "machines", and our video card is from the Middle-end segment, that is, for a wide and affordable class, are equipped with low-power power supplies with only a six-pin power connector for the video card.

The rear interface panel has also undergone significant changes. The usual DVI-D video output is not installed on the reference design Radeon RX 480 8Gb models, although there is a contact pad. This is done so that the hot air meets the least resistance: the errors in the Radeon R9290 (X) model have been fixed. It now boasts one HDMI 2.0b output and three DisplayPort 1.4 (HDR) outputs.

Cooling system

The decorative cover for the cooling system is attached to the sides with screws. Inside, it has a simple shape with a guide wall in the front for a radial fan or, more simply, a "turbine".

The cooling system for the AMD Radeon RX 480 8Gb graphics card, despite its low cost, was not as simple as expected. The engineers were faced with a difficult task - to install a cooler on the video card for $ 229, which can boast of efficiency and low noise level.

In the center there is an aluminum radiator with high fins and a complete absence of heat pipes.

Having removed the crosspiece and dismantled the cooler, you understand that the cooler and the plate are not soldered to each other as it was before, but are separate components. In practice, in the case of installing third-party cooling systems, this allows you to leave the black plate and cool the elements on the board, as intended in the "reference".

The radiator is presented in the form of a simple design with a copper insert. A similar aluminum bar with reservations copes with the heating of the Polaris 10 chip, however, the fan and smart power management algorithms do a good job.

The VRM area is cooled by a single black plate, which is ribbed in this area. This design is ubiquitous on inexpensive video cards.

In general, the developers have tried not to ignore any element on the printed circuit board, be it transistors or memory chips. Not all graphics accelerators are awarded with such a cooling arrangement.

Printed circuit board

The novelty is made on a black PCB board, the length of which is only 170 mm. This length was achieved through a dense arrangement of elements and a GPU that does not require complex PCB layouts. The AMD Radeon RX 480 8Gb graphics card is built using a high-quality element base.

The power subsystem is located on the left, it is made according to the "6 + 1" scheme, where six phases are allocated to the graphics processor and one to the video memory. For a video card with a TDP of 150 W, there is a large power reserve. The IR3567B microcircuit is used as a PWM controller, which is also installed on the models of previous series. It supports voltage regulation and protection OVP, UVP, OCP and OTP.

The Polaris 10 chip is located in the center of the PCB, equipped with a protective frame and rotated 45 ° C. A similar implementation was found on Pitcairn GPUs. It includes 2304 shaders, 32 raster units, 144 texture units, produced on week 18 of 2016.

Eight video memory chips with a total volume of 8192 MB operate at a frequency of 2000 MHz (effective frequency - 8000 MHz). These are Samsung chips, marking K4G80325FB-HC25. They are among the most productive solutions in the line, however, they can boast of high overclocking potential, which, unfortunately, is still limited at around 2250 MHz.

Test bench configuration

• Processor: Intel Core i7-4770K (4000 MHz);
• Motherboard: MSI Z97 Gaming 5, BIOS version 1.11;
• Cooler:;
• Thermal interface: Arctic Cooling MX-2;
• Memory: 2 x 4 GB DDR3 2133, Kingston HyperX Genesis (KHX18C10 / 4);
• Video card: AMDRadeon RX 4808Gb;
• SSD storage: SanDisk X110 256GB;
• Power supply unit: ChieftecAPS-1000C 1000W;
• Case: Cooler Master HAF 922;
• Monitor: BenQ GW2460HM;
• Operating system: Windows 7 64-bit Service Pack 1;
• Drivers: AMD Catalyst 16.7.3.

An Intel Core i7-4770K was used as a central processor, the frequency of which was increased to 4000 MHz. The memory frequency was fixed at around 1600 MHz at timings of 9-9-9-27. The role of the platform was performed by the MSI Z97 Gaming 5 motherboard.

The AMD Radeon RX 480 8Gb graphics card has an increased frequency potential. The base frequency is 1120 MHz, which dynamically rises to 1266 MHz. In idle, the fan only runs at 800 rpm, the temperature of the GPU is kept at around 41 ° C.

In games, the cooling system operates at 2150 rpm, and despite the abnormal heat, it does not allow the chip to warm up above 84 ° C.

Synthetic tests

To assess performance in synthetics, we used the Valley Benchmark, Heaven Benchmark and 3DMark 2013 tests.

Game tests

Let's move on to gaming applications and focus on the testing methodology. FPS was measured using the FRAPS and MSI AfterBurner utilities, the resolution in all games was set to 1920x1080. The following options are manually disabled:

• VSync (vertical sync)

All other settings in the games were set to the maximum possible.

* the list of games will expand.

Temperature and overclocking

The AMD Radeon RX 480 8Gb graphics card is built on the Graphics Core Next architecture version 1.4, which brought new features and technologies, however, let's talk about the basic settings regarding such a parameter as Power Limit. Directly Power Limit adjusts the threshold value of power consumption, upon passing which the graphics accelerator begins to drop frequencies. The developers tried to provide the public that the Radeon RX 480 8Gb is energy efficient, and this is true, however, the new product is very squeezed in the TDP parameter, and the increase in Power Limit, like no other video card from AMD, gives such a performance boost.

WattMan is AMD's new graphics card overclocking utility built into the Crimson driver. It is possible to manually set the core and video memory frequencies, as well as the voltage for the GPU and Memory. Fan speed control is implemented in interesting ways, where we can now set both the rpm directly and indirect indicators such as the critical temperature and the target temperature.

With the help of WattMan, the AMD Radeon RX 480 8Gb video card was overclocked, you can see all the values ​​set for this instance in the screenshot.

We managed to raise the core frequency from 1266 MHz to 1350 MHz at a voltage of 1.15 V - it is impossible to continue using standard means, third-party utilities allow increasing the voltage to 1.3 V, which allows overclocking the video card to 1500 MHz. The memory frequency, as mentioned earlier, is limited to 2250 MHz and no bypass tools have been developed yet.

Overclocking was 7% and 12%, respectively.

These operations increased productivity by 14%.

Abnormal heat reigned in the room at the time of testing - about 30 ° C. Despite this factor, the cooling system worked quite quietly, and the GPU temperature did not exceed 83-84 ° C in nominal mode and 89 ° C in manual overclocking.

Conclusion

Our shelf has a replenishment in the person of AMD Radeon RX 480 8Gb, which is built on a new 14nm process technology, and its performance is comparable to the more expensive GeForce GTX 970 and Radeon R9390 models. The novelty may not yet outperform competitors in terms of the average FPS value, however, this first sign on FinFET and fine tuning and optimization of drivers is just beginning. The developers have already released two versions of the software that improve gaming performance.

The AMD Radeon RX 480 8Gb graphics card boasts improved energy efficiency, frequency and overclocking potential, improved GCN microarchitecture, new revisions of video outputs and a quiet cooling system.

Summing up, I would like to note that optimists will see AMD Radeon RX 480 8Gb as a step forward, realists - an excellent video card, and pessimists - an analogue of the GeForce GTX 970, released two years later.

Advantages:

• High performance;
• By modern standards - 8 GB of video memory;
• Low power consumption;
• Quiet cooling system;
• High-quality element base;
• Recommended cost.

Disadvantages:

• Not found.

New midrange, catching up with the top accelerators of the previous generation

• Part 2 - Practical introduction

Introducing a basic in-depth study of AMD Radeon RX 480.

Object of study: 3D Graphics Accelerator (Graphics Card) AMD Radeon RX 480 8GB 256-bit GDDR5 PCI-E

Developer Details: ATI Technologies (ATI trademark) was founded in 1985 in Canada as Array Technology Inc. In the same year it was renamed ATI Technologies. Headquarters in Markham, Toronto. Since 1987, the company has focused on providing graphics solutions for the PC. Since 2000, Radeon has become the main brand of ATI graphics solutions, under which GPUs are produced for both desktop PCs and laptops. In 2006, ATI Technologies was acquired by AMD to form the AMD Graphics Products Group (AMD GPG). Since 2010, AMD has abandoned the ATI brand, leaving only Radeon. AMD is headquartered in Sunnyvale, California, while AMD GPG remains headquartered in the former AMD office in Markham, Canada. There is no production of its own. The total number of AMD GPG employees (including regional offices) is about 2000 people.

Part 1: Theory and Architecture

In our previous articles, we have repeatedly complained about the stagnation in the field of graphics processors, associated with delays in the production of GPUs for new technological processes and the actual omission of one of them - the 20 nm process technology, which turned out to be unsuitable for the mass production of complex video chips. Over the course of five long (!) Years, both companies that are GPU manufacturers have been releasing solutions based on the already very old 28 nm technical process.

Manufacturers of microelectronic chips were able to establish mass production using new FinFET technological processes (14 and 16 nm, depending on the manufacturer) of such complex and large chips only closer to the middle of the year. Not so long ago, Nvidia, which released rather expensive video cards intended for the upper part of their lineup, "shot out", and now it is time for AMD, which went its own way, first releasing not the most expensive video cards, roughly similar to the Radeon HD 4850 and HD 4870 models. which became quite popular in their time.

To better understand the line of thought of AMD representatives, which differs from that of their competitors, let's look at their ideas about the most demanded video cards on the market. According to AMD, a fairly small percentage of PC gamers buy expensive graphics cards that provide comfort at high resolutions and maximum settings, and most of them use very outdated GPUs. 84% of gamers buy graphics cards for between $ 100 and $ 300 according to AMD, and only the rest of the players choose the one that is more expensive.

It is clear that the majority will not be able to even try the topic of virtual reality, which is so popular now, with such a desire, because VR requires very decent computing power. In addition, according to AMD, not all users are willing to invest in equipment that will become obsolete in a couple of years. True, it is unlikely that all of them will rush to buy VR helmets ... On the other hand, with outdated video cards, they will not even have the opportunity to try out virtual reality. Only 13 million PCs around the world are configured enough to run VR applications - that's just 1% of the nearly 1.5 billion PCs on hand.

According to surveys cited by AMD, two-thirds of users do not plan to purchase equipment for VR precisely because of the high cost of such a configuration. This is in addition to quite reasonable arguments, such as those that helmets are still too bulky and with interfering wires, and virtual reality, in principle, is applicable only to a small part of gaming applications. Yet the biggest barrier to VR adoption is the cost of the hardware. And AMD sees a promising opportunity to provide millions of PCs with the GPU power they need over the next few years. True, it remains unclear why AMD considers a video card to be an unavailable component, if a VR headset and controllers themselves are more expensive? However, they can really lower the threshold for entering VR by offering solutions with sufficient performance for relatively little money.

And AMD is promoting its new solutions in many ways precisely as productive and energy-efficient video cards designed to "democratize" the rather expensive virtual reality, providing those who wish with sufficient GPU power. And yet another goal of the company's new graphics solutions is both compact PCs with ultra-low power consumption and gaming laptops, for which it is now possible to easily provide the same power or even surpass that of gaming consoles. For example, the junior Polaris chip has not only low power consumption, but is also specifically designed for compact laptops - the total packaging height of this GPU is only 1.5 mm compared to 1.9 mm for Bonaire, which will help AMD win tenders for the supply of solutions for mobile PCs.

To meet these demands clearly, AMD decided to design two GPU models, the Polaris 10 and Polaris 11, to meet specific levels of capability and performance. The senior chip in the Polaris series will provide PC gamers with enough power for VR applications and all modern games, while the lower-performing low-end GPU is designed for thin and light laptops, but offers features and performance that surpass those of game consoles.

Accordingly, at the time of the announcement, AMD offers the following desktop solutions:
Radeon RX 460- an energy efficient video card with low power consumption for undemanding games and future mobile solutions, with a capacity of more than 2 teraflops, with 2 GB of video memory connected via a 128-bit bus;
Radeon RX 470- a very profitable mid-range video card for an affordable price, with enough power for games in Full HD-resolution, with a capacity of more than 4 teraflops, 4 GB of video memory and a 256-bit bus;
Radeon RX 480- so far the most productive solution of the new family, designed for VR and modern games with a performance of more than 5 teraflops, 4 or 8 gigabytes of memory with a 256-bit bus, consuming less than 150 watts.

Today we're looking at the Radeon RX 480, which offers premium gaming experiences - Premium HD Gaming. What is this term in AMD's understanding? This includes both the capabilities of new graphics APIs, such as asynchronous execution in DirectX 12, as well as FreeSync and CrossFire technologies. But the main thing is the advantage over similarly priced competitor solutions in modern games with DirectX 12 support:

In most games of the current year with support for DirectX 12 (Ashes of the Singularity, Hitman, Total War: Warhammer, Quantum Break, Gears of War and Forza APEX), even the previous generation AMD Radeon video cards often outperform their counterparts at the price of Nvidia: we noted the advantage of Fury X versus 980 Ti, R9 390 versus GTX 970 and R9 380 versus GTX 960, and the latest Polaris 10 model is bound to do even better.

In addition to DirectX 12, one more API can be noted - Vulkan. In the corresponding version of the Doom game, AMD claims an increase of up to 45% on the Radeon RX 480 compared to the OpenGL version of the game, although the difference is expected to be slightly smaller on older video cards - about 20-25%.

And what about virtual reality, is the new AMD product capable of sufficient performance for VR applications? Thanks to the high power of the GPU and support for features such as Asynchronos Time Warp, it provides a comfortable viewing of relevant VR applications, and even with low power consumption. So, the generally accepted test for evaluating the performance of the SteamVR Performance Test shows a clear superiority over the solutions of the previous generation (it is not clear, however, why was it compared with the Radeon R9 380?):

Since the basis of the Radeon RX 480 model is the Polaris 10 graphics processor, which has the fourth generation GCN architecture, which is similar in many details to the previously released solutions from AMD, it will be useful to familiarize yourself with our previous materials on the company's past video cards before reading the theoretical part of the article. based on the GCN architecture of previous generations:

• AMD Radeon R9 Fury X: New AMD flagship with HBM support
• AMD Radeon R9 285: Tahiti got a 256-bit bus and turned into Tonga
• AMD Radeon R9 290X: Reach Hawaii! You will get new heights of speed and functionality
• AMD Radeon HD 7970: New Single-Processor 3D Leader

Let's take a look at the detailed specifications of the Radeon RX 480 video card based on the full version of the next generation Polaris 10 GPU.

Graphics Accelerator Radeon RX 480
Parameter	Meaning
Chip codename	Polaris 10 XT (Ellesmere)
Production technology	14 nm FinFET
Number of transistors	5.7 billion
Core area	232 mm²
Architecture	Unified, with an array of common processors for streaming processing of numerous types of data: vertices, pixels, etc.
DirectX hardware support	DirectX 12, with support for Feature Level 12_0
Memory bus	256-bit: Eight independent 32-bit memory controllers supporting GDDR5 memory
GPU frequency	1120 (1266) MHz
Computing units	36 GCN computing units, including 144 SIMD cores, consisting of 2304 ALUs for floating point calculations (integer and floating point formats are supported, with FP16, FP32 and FP64 precision)
Texture units	144 texture units, with support for trilinear and anisotropic filtering for all texture formats
ROP units	32 ROPs with support for anti-aliasing modes with the possibility of programmable sampling of more than 16 samples per pixel, including FP16 or FP32 framebuffer format. Peak performance up to 32 samples per cycle, and in Z only mode - 128 samples per cycle
Monitor support	Integrated support for up to six monitors connected via DVI, HDMI 2.0b and DisplayPort 1.3 / 1.4 Ready

Radeon RX 480 Reference Graphics Specifications
Parameter	Meaning
Core frequency	1120 (1266) MHz
Number of universal processors	2304
Number of texture units	144
Number of blending blocks	32
Effective memory frequency	7000-8000 (4 × 1750-2000) MHz
Memory type	GDDR5
Memory bus	256-bit
Memory	4/8 GB
Memory bandwidth	224-256 GB / s
Computational Performance (FP32)	up to 5.8 teraflops
Theoretical maximum fill rate	41 gigapixels / s
Theoretical texture sampling rate	182 gigatexels / s
Tire	PCI Express 3.0
Connectors	One HDMI and three DisplayPort
Energy consumption	up to 150 W
Additional food	One 6-pin connector
The number of slots occupied in the system chassis	2
Recommended price	$ 199 / $ 229 (for the US market)

The name of the AMD video card model released today is quite consistent with their current naming system. Its name differs from its predecessors by the changed symbol in the first part of the index and the generation digit - RX 480. If everything is clear with the second change, because the generation is really new, then replacing R9 with RX is not entirely logical, in our opinion, because this figure used to show the level of a video card : The R7 were slower than the R9, but they were all produced in the same generation. And now it is not clear, firstly, why the RX 480 has this figure higher than the R9 390X, for example, and what numbers after the R in the name will be in the junior solutions based on new GPUs.

The first model in the new Radeon 400 family is replacing previous solutions similar in positioning in the company's current lineup, replacing them in the market. Since the released video card belongs rather to the average level in terms of price and speed, taking into account the new generation, they decided to leave the 490 index for future solutions on GPUs of even greater power.

The reference Radeon RX 480 will be offered for a suggested price of $ 199 for the 4GB model and $ 229 for the 8GB model, and these prices are very attractive! Compared to the top-end video cards of the previous generation, this is a very good price tag, since the Radeon RX 480 should not be inferior in speed to such models as the Radeon R9 390 and GeForce GTX 970. It is with them that the new product will compete, at least at the beginning of its life path, until the release of the upcoming release of the GeForce GTX 1060. But at the time of its release, today's new product is definitely the best performance offering in its class.

Reference Radeon RX 480 graphics cards will come in versions with 4GB of GDDR5 memory at 7GHz effective frequency and 8GB of memory clocked at 8GHz. But as AMD partners' own video cards go on sale, other options will appear, but all of them will be equipped with GDDR5 memory with a frequency of at least 7 GHz - such is the will of AMD.

The decision to install 4 and 8 GB of memory is very wise. The younger version will allow you to save a little, because 4 GB at the moment can be considered the "golden mean", and the advantage from 8 GB of memory in the second version of the Radeon RX 480 will be revealed in the future. Although a 4 GB version of a video card will provide acceptable performance in modern games, 8 GB of memory will allow you to have a decent headroom for the future, since the requirements for video memory for games are constantly growing. An example, the advantage of which is already noticeable, is the game Rise of the Tomb Raider in DirectX 12, at very high settings and a resolution of 2560x1440 pixels:

The larger amount of video memory in the Radeon RX 480 8 GB and Radeon R9 390 helps to avoid extremely unpleasant drops in performance and jerks in FPS, compared to 4 GB variants, including solutions from competitors GeForce GTX 970 and GTX 960. It is the Radeon RX 480 8 GB that makes it possible to obtain smooth gameplay with no lags associated with loading data that does not fit into the local video memory. And since current generation gaming consoles each have 8GB of shared memory, the benefit of more memory will only grow over time, and the 8GB variant of the Radeon RX 480 will be great for gaming over the next few years.

The board uses a single 6-pin connector for additional power, and the Radeon RX 480 has a typical power consumption of 150W on the Polaris 10 GPU. In reality, without overclocking, the board consumes even less, about 120 watts of energy, but a small power reserve will improve the overclocking potential. By the way, AMD partners are planning a quick release of factory-overclocked versions of this video card, which differ in cooling and power systems.

Architectural features

The Polaris 10 GPU is the fourth generation of the Graphics Core Next architecture, the most advanced yet. The basic building block of the architecture is the Compute Unit (CU), from which all AMD GPUs are assembled. The computational unit CU has a dedicated local data storage for exchanging data or expanding the local register stack, as well as a first-level cache memory with read / write capabilities and a full-fledged texture pipeline with fetch and filter units, it is divided into subsections, each of which works on its own flow of commands. Each of these blocks is responsible for the planning and distribution of work independently.

Basically, the architecture of Polaris has not changed too much, although not the main units of the video chip have changed more noticeably - the units for encoding and decoding video data and outputting information to display devices have been seriously improved. Otherwise, this is the next generation of the famous Graphics Core Next (GCN) architecture, already the fourth in a row. So far, the family has included two chips: Polaris 10 (formerly known as Ellesmere) and Polaris 11 (formerly known as Baffin).

And yet, some hardware changes to the GPU have been made. The list of improvements and changes includes: improved geometry handling, support for multiple projections when rendering VR at different resolutions, an updated memory controller with improved data compression, modified instruction prefetching and improved buffering, scheduling and prioritization of computational tasks in asynchronous mode, support for operations on data in FP16 / Int16 format. Consider the diagram of the new GPU (an enlarged version of the illustration is available by clicking on the image):

The full-fledged Polaris 10 GPU includes one Graphics Command Processor, four Asynchronous Compute Engines (ACE), two Hardware Schedulers (HWS), 36 Compute Units (CU), four geometry processors, 144 texture TMUs (with four LSUs per TMU) and 32 ROPs. The memory subsystem of AMD's new GPU includes eight 32-bit GDDR5 memory controllers, providing a shared 256-bit memory bus, and a 2MB L2 cache.

Improvement of geometry engines in Polaris is announced - in particular, the so-called accelerator for discarding geometric primitives Primitive Discard Accelerator has appeared, which works at the very beginning of the graphics pipeline, discarding invisible triangles (for example, with zero area). Also in the new GPU, a new index cache for instanced geometry was introduced, which optimizes data movement and frees up the resources of the internal data transfer buses and increases the efficiency of using the memory bandwidth when duplicating the geometry (instancing).

The geometry drop accelerator helps to increase the speed of geometry processing, especially in tasks like tessellation with multisampling. The diagram shows that under different conditions, the new block can increase productivity up to three times. However, these are synthetic data of the interested party, it is better to look at the game results of independent tests.

Also in the fourth generation GCN, the efficiency of shader execution was improved - instruction prefetching was introduced, which improves instruction caching, reduces pipeline downtime and increases overall computational efficiency. The instruction buffer size for the instruction array (wavefront) was also increased, increasing single-threaded performance, support for operations on data in FP16 and Int16 formats was introduced, which helps to reduce memory load, increase computation speed and improve energy efficiency. The latter can be applied to a wide range of graphics, machine vision, and learning tasks.

The hardware scheduler (HWS), which is used for asynchronous computations, has been improved once again. Its tasks include: unloading the CPU from scheduling tasks, prioritizing real-time tasks (virtual reality or sound processing), parallel execution of tasks and processes, resource management, coordination and balancing of the execution unit load. The functionality of these blocks can be updated using microcode.

In addition to doubling the L2 cache size to 2 MB, the L2 cache processing and caching has been changed and the overall efficiency of the cache and local video memory subsystem has been increased. The memory controller received support for GDDR5 memory with an effective clock frequency of up to 8 GHz, which in the case of Polaris means a memory bus bandwidth of up to 256 GB / s. But AMD did not stop there either, further improving the lossless data compression algorithms (Delta Color Compression - DCC), which support compression modes with a ratio of 2: 1, 4: 1 and 8: 1.

Intrachip data compression increases overall operational efficiency, provides better utilization of the data bus, and impacts on energy efficiency. In particular, if there was no internal data compression in the Radeon R9 290X and the effective memory bandwidth is equal to its physical memory bandwidth, then in the case of a solution on the Fiji chip, compression allowed saving almost 20% of the memory bandwidth, and in the case of Polaris, up to 35-40%.

If we compare the Radeon RX 480 with the Radeon R9 290, then the new solution consumes significantly less power to provide the same effective bandwidth as compared to the previous generation video card. As a result, the new product has noticeably higher performance per bit - although the Radeon R9 290 has a higher peak memory bandwidth, it is much more energy efficient in Polaris 10 - the total energy consumption of the memory interface is 58% of that of the old GPU.

Overall, the fourth-generation GCN changes in the Polaris GPU are related to the advanced 14nm FinFET workflow, microarchitectural changes, physical design optimizations and power management techniques. All of this has paid off in the form of significant performance and efficiency gains over previous solutions. At the lowest level, the CUs in Polaris 10 (Radeon RX 480) are about 15% more productive than the Hawaii chip units (Radeon R9 290).

It is difficult to judge how great is the contribution of one or another optimization to the overall speed gain, but if we take all the optimizations together, the difference in energy efficiency between the Radeon RX 470 and Radeon R9 270X, according to AMD specialists, reaches 2.8 times. Moreover, they estimate the contribution of the FinFET process to be less than the contribution of their optimizations. Probably, the most favorable comparison was chosen, and for other models the increase in energy efficiency is slightly less. For example, if you compare the performance of RX 480 and R9 290, the difference in energy efficiency will be closer to twofold. In any case, such huge increases occur once every few years, and for this reason we have no doubt that the sales of the Radeon RX 480 will be successful.

Technological process and its optimization

As we already said, the main thing in Polaris is not changes in hardware units, but a big step forward due to the use in the production of this GPU of a new 14 nm process technology using vertical gate transistors (FinFETs - Fin Field Effect Transistors), also known like transistors with a three-dimensional gate structure or 3D transistors.

Dynamic power consumption grows linearly with an increase in the number of computing units, and cubic with an increase in frequency by increasing the voltage (for example, a 15% increase in frequency and voltage increases consumption by more than half!), And as a result, GPUs often run at lower clock frequencies. instead, they use chips of higher density to accommodate more computing devices that work in parallel.

For the past five years, GPUs have been produced using 28 nm technical processes, and the intermediate 20 nm did not give the required parameters. It took a long time to master even more advanced technical processes, and now, for the production of graphics processors of the Polaris family, AMD chose the production of Samsung Electronics and GlobalFoundries with their 14 nm FinFET process, which ensures the production of some of the densest microprocessors. The use of FinFETs is critical to lower power consumption and lower GPU voltage by about 150mV over the previous generation, cutting power by a third.

The illustration schematically shows the conditional resizing of the same GPU, produced using different technical processes. Samsung Electronics and GlobalFoundries share orders for the production of 14 nm central and graphic processors from AMD, since they have the same technical process and it is not difficult to establish simultaneous production, dividing orders between them based on the yield of suitable chips and other parameters, which should solve potential problems with insufficient production volumes.

The Polaris architecture was originally designed for the capabilities of FinFET processes, and should use all their capabilities. In short, a FinFET is a transistor with a channel surrounded by a gate through an insulator layer on three sides - compared to a planar one, where the interface is one plane. FinFET transistors have a more complex device, and there were plenty of difficulties in implementing the new technology; it took five years to master the corresponding technical processes.

On the other hand, the new form of transistors provides a higher yield, less leakage and noticeably better energy efficiency, which is the main task of modern microelectronics. The number of transistors in GPUs per square millimeter of area has doubled roughly every two years, and static leakage has also doubled. To solve some of these problems, special tools were used, such as islands of transistors with different supply voltages and clock gating circuits, which helped to reduce leakage currents in idle or sleep modes. But these techniques do not help with active work states and can reduce peak performance.

FinFET processes solve many of the problems, resulting in revolutionary improvements in performance and power consumption over previous chips made with traditional technologies. New technical processes allow not only increasing performance, but also reducing the variability of characteristics (the difference in the characteristics of all manufactured chips of the same model) - compare the spread of parameters for the FinFET 14 nm process and the usual 28 nm at TSMC:

This chart shows both the higher average performance for FinFET products and lower leaks on average, and less variation in performance and leak rates for different samples. Improving the variability of these characteristics for the GPU in the case of FinFET means that you can increase the final frequency for all products, while for planar transistors, you had to pay more attention to the worst performance and reduce the reference characteristics for all final products.

As a result, GPUs manufactured using FinFET process technology provide a fundamental increase in performance and energy efficiency, compared to counterparts in the production of which were used traditional planar transistors. According to AMD experts, the use of FinFET-technological processes allows to provide either 50-60% less energy consumption, or 20-35% more performance, all other things being equal.

New FinFET processes are helping not only reduce power consumption and dramatically improve energy efficiency, but also open up new form factors and formats for future GPUs. So, in the future, relatively thin and light gaming laptops may appear that will not require a significant reduction in the quality of 3D graphics, sufficiently powerful desktop PCs of an ultra-compact size, but the usual gaming video cards will be able to do with fewer power connectors.

But in order to achieve greater energy efficiency, it is not enough just to transfer the chip to a "thinner" technical process, numerous changes are required in its design. For example, Polaris uses adaptive GPU clocking. GPUs operate at low voltages and high amperages, and it is difficult to supply quality voltage from power circuits. The variation in voltage can be as high as 10-15% of the nominal value, and the average voltage has to be increased in order to cover this difference, and this wastes a lot of energy.

AMD's adaptive clocking recovers these losses with a quarter of the power savings. For this, in addition to the existing energy consumption and temperature sensors, a frequency sensor is also added. As a result, the algorithm achieves maximum energy efficiency for the entire chip.

It also calibrates the power supply when the system boots. When testing the processor, a special code is run to analyze the voltage, and the voltage value is recorded by the integrated power monitors. Then, when the PC boots up, the same code is run and the resulting voltage is measured, and the voltage regulators on the board set the same voltage as during testing. This eliminates the waste of energy that is wasted due to differences in systems.

There is also adaptive compensation for aging transistors in Polaris - usually GPUs require a clock frequency reserve of about 2-3% to adapt to aging of chip transistors, and other components also show aging (for example, the GPU receives a lower voltage from the system). Modern AMD solutions are able to self-calibrate and adapt to changing conditions over time, which ensures reliable operation of the video card for a long time and slightly increased performance.

Radeon WattMan - New Overclocking and Monitoring Capabilities

An important component of any modern video driver is overclocking settings that allow you to squeeze all its capabilities out of the GPU. Previously, this was in charge of the AMD Overdrive section in the solution drivers of this company, and along with the release of new solutions, AMD decided to radically update this driver section, calling it Radeon WattMan.

Radeon WattMan is AMD's new overclocking utility that allows you to alter GPU voltage, GPU and VRAM frequencies, cooling fan speeds, and target temperatures. Radeon WattMan builds on capabilities previously seen in Radeon Software, but offers several new subtle overclocking features - with different GPU voltage and frequency control capabilities. Also in WattMan there is a convenient monitoring of GPU activity, clock frequencies, temperatures and fan speed.

Conveniently, as with other Radeon Software Crimson Edition settings, you can set your own overclocking profile for each application or game that will be applied when it starts. And after the application is closed, the settings will return to the global default. Radeon WattMan can be found in Radeon Settings, it replaces the current AMD OverDrive panel, and is compatible with the AMD Radeon RX 400 series.

Both simple control of the GPU frequency and fine tuning of the frequency curve are possible. Simple Frequency Tuning works by default and allows you to change AMD-specific values ​​that are optimal for each GPU state. Changing the frequency curve is possible with an accuracy of 0.5%. There is also a dynamic change in the frequency curve, when the clock frequency of the GPU core and video memory can change for each state along with a change in voltage for each of them. GPU and memory voltages are set independently of each other.

WattMan also has advanced fan speed control in the cooling system, when the minimum speed, target speed and minimum acoustic limit are set. In this case, the target rotation speed is the maximum at which the fan will rotate at a temperature not higher than the target one. Improved temperature management allows you to set the maximum and target temperature values. Together with the energy consumption limit, this allows for more fine tuning.

The maximum temperature is the absolute maximum at which the frequency of the graphics chip does not decrease, but after reaching it, the frequency will begin to decrease. And the target temperature is the value upon reaching which the fan speed will increase. The GPU power limit can be increased or decreased by up to 50% (in the case of the Radeon RX 480 model).

It seems that somewhere we have already seen the possibility of a subtle change in the frequency and voltage curve, and more recently, right? But what we haven't seen for sure yet is a convenient interface for monitoring and setting in the drivers themselves, and not in third-party utilities, and AMD can only be praised for such concern for users.

A new monitoring interface allows you to record and view GPU activity, temperature, fan speed and frequencies. Moreover, there is both global monitoring (Global WattMan) and separate monitoring for user profiles, which monitors peak and average data only when the application is open. Data is also collected in the background, the Radeon Settings utility does not need to be running, the data is collected up to a maximum of 20 minutes of application operation.

In general, AMD still has something to work on to improve the usability of the WattMan interface, since it is not intended for keyboard control, for example, but the initiative itself can only be welcomed - convenient configuration and monitoring tools right in the drivers can be an additional advantage of new solutions of the Radeon RX 400 family.

New options for displaying images

We have already talked about the fact that new solutions from AMD will feature support for the latest DisplayPort and HDMI standards. The new Radeon RX family of graphics cards were among the first to support DisplayPort 1.3 HBR3 and DisplayPort 1.4-HDR. Newer versions of this standard use existing cables and connectors, but additional length restrictions may apply.

The main benefit of DisplayPort 1.3 HBR3 is the increase in bandwidth to 32.4 Gbps (80% more than HDMI 2.0b), pushing the bandwidth limit of the previous generation DisplayPort 1.2. The new standard allows 5K monitors in RGB format at 60 Hz using a single cable (now you have to connect a couple of connectors and cables), as well as UHDTV TVs with a resolution of 8K (7680 × 4320) using color subsampling 4: 2: 0 at 60 Hz. Also, via DisplayPort 1.3, you can connect stereo displays with 120 Hz and 4K resolution. 5K single-cable displays and 4K HDR displays are expected towards the end of this year.

Polaris is also ready to roll out the DisplayPort 1.4-HDR standard, which supports up to 10-bit 4K output with refresh rates up to 96Hz. The company's new product supports ITU Rec.2020 Color Space Recommendations for UHDTV, as well as CTA-861.3 and SMPTE 2084 EOTF standards for HDR data transmission.

The new DisplayPort 1.3 standard will also help advance FreeSync technology for 4K monitors. AMD expects the first such devices to be available with dynamic refresh technology up to 120Hz by the end of 2016. Such monitors will be capable of operating in 4K resolution using FreeSync technologies at 30-120 FPS and will support Low Framerate Compensation.

Here is a list of the next-generation monitor specifications that are made possible by the new Extended Bandwidth DisplayPort 1.3 standard: 1920 x 1080 monitors: 240 Hz SDR and 240 Hz HDR, 2560 x 1440 monitors: 240 Hz SDR, and 170Hz HDR, 4K monitors: 120Hz SDR and 60Hz HDR, 5K monitors: 60Hz SDR.

If we have already started talking about FreeSync, then it should be mentioned that in solutions of the Polaris architecture this technology will work with monitors with HDMI 2.0b connectors. The company is currently working with partners including Acer, LG, Mstar, Novatek, Realtek and Samsung to enable dynamic refresh rate technology, including via HDMI. The list of monitors planned for release includes products with screen sizes from 20 to 34 inches and various resolutions.

One of the most exciting and promising display capabilities of Polaris is HDR support for high dynamic range displays. To obtain a high-quality picture, you need to display images in a wide color gamut with increased contrast and maximum brightness, and on current displays, a person sees only a small part of what he can observe with his own eyes in the world around him. The range of brightness and color we perceive is much greater than what current output devices can give us.

Many image quality enthusiasts are waiting for the implementation of High Dynamic Range in all stages of the image processing pipeline. In order to even get closer to the possibilities of human vision, a new industry standard for TVs was introduced - HDR UHDTV, providing a brightness range from 0.005 to 10,000 nits. The first HDR devices have a brightness of up to 600-1200 cd / m2, and LCD monitors with support for High Dynamic Range (HDR) and local backlight will be able to provide up to 2000 nits in the future, and OLED displays up to 1000 nits, but with ideal black and more contrast.

When using HDR, users will also be shown an extended color range, since the currently widespread sRGB color space lags far behind human vision. Almost all of the current content is created within the BT.709, sRGB, SMPTE 1886 (Gamma 2.4) standards, and the new HDR-10, Rec.2020 (BT.2020) standard, SMPTE 2084 is capable of displaying more than a billion colors at 10-bit on component that brings the quality of color reproduction closer to natural for humans.

Don't confuse the topic of display devices with HDR capabilities with what has long appeared in games called HDR rendering. Indeed, many modern game engines use high dynamic range rendering to preserve data in shadows and highlights, but this is done exclusively before displaying information on the display. And then the image is still reduced to the usual dynamic range in order to display it on the SDR monitor.

For this, special tone mapping algorithms are used ( tone mapping) - converts tonal values ​​from a wide range to a narrow one. With the advent of HDR devices, both improved tone mapping algorithms and their orientation to HDR displays are needed. Polaris' hardware color processing engine has programmable gamma control and gamut remapping capabilities, all calculations are performed with high accuracy, and the result will be fully consistent with the capabilities of the display.

While even current Radeon graphics cards are ready to handle HDR monitors to some extent, these new models offer noticeably higher refresh rates and color depth. Polaris GPUs are ready for HDR monitors with 10-bit and 12-bit per component color depth, although the first such displays will only support 10-bit, but more advanced ones will follow that surpass the capabilities of human vision.

In order to get a high-quality HDR image in gaming applications, it is necessary to redo not only the graphics part of the game engine, but also some of the content: the same textures must also be stored in formats that allow using a wide color and brightness gamut. AMD is working with game developers to ensure that future games can take full advantage of HDR displays and have released a dedicated Radeon Photon SDK.

And there is a lot to work on. Tone mapping in games must be done by the graphics engine, as this process by the display adds significant delays. AMD suggests doing this: the monitor is polled for its capabilities in color, contrast and brightness, then, taking into account this information, the game engine makes a tone mapping and displays it on the display ready-made. Since game engines already do SDR tone mapping, they just need to add HDR output capability.

The Photon SDK is already available for developers, HDR support for video data and rendering in DirectX 11 applications in the driver is ready, and support for DirectX 12 is planned with a future update. It remains to add that Polaris supports HDR displays connected using HDMI 2.0b (with HDCP 2.2) at 1920 x 1080 @ 192 Hz, 2560 x 1440 @ 96 Hz and 3840 x 2160 @ 60 Hz and 4 color coding. : 2: 2. In the case of connection via DisplayPort 1.4-HDR (also with HDCP 2.2), the possibilities are wider: 1920 × 1080 at 240 Hz, 2560 × 1440 at 192 Hz and 3840 × 2160 at 96 Hz. It remains to wait for such monitors with a price lower than that of a cast-iron bridge.

Improved encoding and decoding of video data

As often happens, in new generations of graphics processors, hardware video processing units are also improved. After all, time does not stand still, there are more and more new formats and conditions for their use (frame rate, color depth, etc.) Therefore, it is not surprising that Polaris has made some improvements in decoding and encoding video data.

If the previous solutions were able to encode video in H.264 up to 4K resolution at 30 or even 60 FPS, then Polaris learned to encode video in HEVC (H.265) format for the first time. The hardware video encoding unit in the new GPU supports the following resolutions and frame rates: 1080p @ 240 FPS, 1440p @ 120 FPS, and 4K @ 60 FPS.

Moreover, support for high-quality encoding of streaming video from games has been added on the Radeon RX series video cards. After all, the quality of encoding has always been a weak point of streaming video, and with a rapidly changing image, its quality suffers greatly. High image quality can be achieved with two-pass encoding with image analysis in the first pass, which was implemented in Polaris. Hardware two-pass encoding works with both H.264 and HEVC formats, and this approach produces a noticeably higher quality video stream.

Unleashing the hardware capabilities of the Polaris architecture also requires software support. A high-quality hardware encoder for games is supported by the following utilities: Plays.TV, AMD Gaming Evolved, Open Broadcaster Software.

Also Polaris is equipped with the most advanced hardware unit for decoding video data. AMD video decoder can work with HEVC format and Main-10 coding profile in resolutions up to 4K at 60 FPS, MJPEG in 4K resolution at 30 FPS, H.264 in 4K resolution up to 120 FPS, MP4-P2 up to 1080p at 60 FPS and VC1 up to 1080p @ 60 FPS.

Support for virtual reality systems

Over the past few years, the current reincarnation of virtual reality helmets has come a long way, constantly improving its consumer characteristics (although it is still very far from ideal). If it all started with less than Full HD-resolution for both eyes in 2014 at no more than 30 FPS, now it has come to a resolution of 1080 × 1200 pixels per eye at 90 FPS and 10 ms latency. And now the VR experience is much more comfortable and realistic.

AMD is also dedicated to improving VR performance. So, the LiquidVR technology assumes the implementation of some features that improve VR on the company's solutions. The latest changes include support for TrueAudio Next audio technology, reservation of computational units for specific tasks, asynchronous Quick Response Queue computing technology, variable resolution and rendering quality for VR, support for DirectX 12 and Vulkan.

So, the technology of advanced sound processing TrueAudio Next includes all work with sounds on the GPU in real time - in compliance with the physical laws of sound wave propagation and the use of ray tracing (ray tracing) for multiple sound sources. This allows you to get high-quality sound with low latencies and, using the settings (the number of processed sources and the number of reflections of sound waves), you get a well-scalable solution.

Another possibility for working with VR, which has recently appeared, is the allocation of several Compute Units for various tasks, such as sound processing - in this case, these CUs will exclusively deal with these tasks in order to avoid problems associated with the simultaneous execution of various tasks on Real-time GPU - This solution provides immediate execution of critical code and works with any type of shader, computational or graphics.

And Polaris has improved the command processor with a new quality of service (QoS) technique called Quick Response Queue. This technique allows developers to assign high priority to certain computational tasks via APIs. Both types of tasks (normal and priority) share the same GPU resources, but higher priority ensures that such tasks will use more resources and finish first, without switching the shell to low priority tasks.

Specifically in LiquidVR, this technique is used with Asynchronous Time Warp, which is used in VR systems to avoid dropped frames that impair the smoothness of the process - in VR this is a very latency task, and prioritizing tasks will help to ensure that the distortion time happens exactly when you need it. The Quick Response Queue (QRQ) technique gives precise control over timings, minimizing them.

Without using the asynchronous time warp technique in virtual reality systems, it turns out that the GPU drops about 5% of frames during operation, and with Asynchronous Time Warp these frames are not dropped, which reduces jitter (different rendering times of adjacent frames) tenfold. At the moment, the feature is already part of the library available on the GPUOpen website.

We already know about another optimization related to VR - the use of multiple projections when rendering a virtual reality scene at different resolutions. We've talked more than once about this feature, which optimizes VR rendering by using independent resolution and resolution quality settings for multiple projections, which simulate the funnel-shaped rendering used in VR headsets. In this case, the center of the frame is rendered at high resolution, and at the periphery it is reduced to optimize performance.

LiquidVR includes support for DirectX 12, an ideal graphics API for a virtual environment, as it allows you to increase the number of draw call functions in the scene, helps to reduce the load on the CPU, has native support for asynchronous computation execution and multi-chip rendering, and also provides some features for low-level GPU access. Examples of using DirectX 12 with LiquidVR, as well as related documentation, are available at GPUOpen.com.

Radeon Software Technologies

AMD continues to improve not only the hardware of its products, but also the software components. Once again, they decided to optimize the frequency of new versions of video drivers, as some users were unhappy with what happened last year. For years, they released updated WHQL drivers on a monthly basis, but some users felt it was too often. After they reduced the frequency of driver releases, other users were unhappy with the already rare releases.

So, in 2015, three WHQL drivers and 9 beta versions were released, and the plan for 2016 is as follows: six full-fledged drivers with WHQL certification per year + as many special versions with optimizations for games as you need (ideally, also WHQL) ... So far, they almost always succeed, since the release of the games, Radeon Software Crimson Edition drivers were available for the games The Division, Far Cry Primal, Hitman, Quantum Break and others. There was a slight hitch with the Doom game and video cards based on chips of previous generations of GCN, but who doesn't happen to?

AMD continues to pay attention to driver optimizations for smooth frame rates, especially in multi-chip configurations. For example, the CrossFire API for DirectX 11 was included in GPUOpen, and for some DirectX 12 applications it is planned to support multi-GPU rendering with smooth frame changes and a small difference in the rendering time of adjacent frames, and not only with high FPS.

Future Radeon Software drivers for DX12 games are planning special support for AFR frame pacing, a technology that specifically adds latency before displaying an image to the screen, which improves smoothness and eliminates jerking in multi-chip rendering.

It is very important that more and more attention is paid to operating systems other than Windows. So, Polaris support is provided for open source Linux distributions - these drivers already have support for the Vulkan version of the Dota 2 game, for example.

Of the curious, we note a special program for beta testing the Radeon Software Beta Program. This program is managed by the Quality Assurance (QA) department and anyone can join by writing to [email protected] for more information.

The most important change comes with the Radeon Settings included with the new driver. There, there was global support for Crossfire and energy efficiency, HDMI scaling and scaling depending on a specific application, changing the color temperature, choosing the language of the user interface and much more - we already talked about the overclocking and monitoring capabilities above.

All of this applies to end users, but there are ongoing changes in software support for developers. The GPUOpen open source initiative has long been known as a convenient method for providing SDK developers with open source libraries and examples. In the last month alone, 14 major updates have appeared on the portal, 41 blogs have been written by developers in four months, and more than 60 examples of code, SDKs, libraries and utilities have been posted since the initiative was launched at the end of January.

Recent examples include ShadowFX with DirectX 12 support, GeometryFX improvements for DirectX 11, updated TressFX 3.1 (DirectX 11). There are new libraries, SDKs and examples for multichip rendering in DirectX 12, an example of out of order rasterization for Vulkan, FireRays for Vulkan and OpenCL, CrossFire API support for DirectX 11. AMD also became the first hardware manufacturer to release an extension for SPIR-V - shader language in the Vulkan graphics API with support for GCN instructions). There is also Radeon support for OpenVX, an open cross-platform standard for accelerating machine vision applications.

AMD recently introduced the Shader Intrinsic Functions extension for the GPUOpen library, which will make it easier to optimize PC versions of games by making it easier to develop multi-platform applications and port games from consoles. Using Shader Intrinsic Functions, the developer can directly access low-level instructions, just like on consoles, by inserting low-level code into high-level sources. This feature can be used in applications that support DirectX 11, DirectX 12, and Vulkan.

Conclusions on the theoretical part

The Radeon RX 480 graphics card is the first in the Polaris family, the first to market in a new AMD lineup based on GPUs designed and manufactured using the 14nm FinFET process. Together with architectural optimizations, this has significantly increased the energy efficiency of the new solution, and as a result, the new product is twice or three times better than previous AMD video cards in terms of this indicator.

Although the Polaris 10 GPU is architecturally very similar to previous chips and is largely the same as their solutions, and the graphics architectures of different GCN generations do not differ too much from each other, many improvements have been made to the new GPU for more efficient computing of various types, including with asynchronous code execution, the display capabilities and the functionality of the video encoding and decoding units have been seriously improved.

Polaris 10 is the best graphics core from AMD, bringing new functionality, but most importantly, it is significantly more efficient. Thus, improvements in computational cores led to a 15% increase in the performance of mathematical calculations, compared with the GCN architecture of previous generations. Together with the use of the new 14 nm FinFET process technology and other optimizations, this has significantly improved energy efficiency - up to 2.8 times, according to the company. And this, in turn, means better user characteristics in terms of heat dissipation and noise from the cooling system.

The list of functional changes and improvements - support for encoding and decoding of modern video formats with new features: support for higher bitrates and advanced formats, readiness to decode streaming HDR video from online services, recording gameplay on the fly without the participation of CPU power, high-quality video encoding mode with two passes, etc. Also noteworthy is the emergence of support for image output standards that will become very important in the future: 10- and 12-bit output formats for HDR TVs and monitors, as well as support for displays with high resolution and refresh rates.

But the main thing about the Radeon RX 480 product presented today is its price. Let some people think that there are not so many functional innovations and optimizations in Polaris, but this new product, using a modern technological process, has significantly reduced the price of a video card, which is quite sufficient both for the latest games with high quality settings and for use in systems virtual reality, quite demanding on GPU power.

The combination of a relatively low price and fairly high performance makes the Radeon RX 480 one of the most successful video cards in terms of price / performance at the time of its release, if not the most profitable. It is important that it is focused on the middle price segment, which attracts a much larger number of potential buyers than top-end solutions, and the release of just such a model in the first place can have a positive effect on AMD's market share in the segment of gaming video cards.

In the following parts of our article, we will evaluate the performance of the new AMD Radeon RX 480 graphics card in practice, comparing its speed with those of similarly priced accelerators from Nvidia and AMD. First, we will look at the data obtained in our set of synthetic tests, and then move on to the most interesting - gaming tests.

Thermaltake DPS G 1050W power supply for the test bench provided by the company Thermaltake	Corsair Obsidian 800D Full Tower Testbed Case Courtesy of Corsair	G.Skill Ripjaws4 F4-2800C16Q-16GRK memory modules for the test bench were provided by the company G.Skill	Corsair Hydro SeriesT H100i CPU Cooler for testbed provided by the company. Corsair
Dell UltraSharp U3011 Testbed Monitor Courtesy of Yulmart	ASRock Fatal1ty X99X Killer motherboard for testbed provided by ASRock	Seagate Barracuda 7200.14 3TB hard drive for testbed provided by the company Seagate	2x Corsair Neutron SeriesT SSD 120GB for testbed courtesy of the company Corsair

The confrontation in the segment of top-end video cards always attracts the attention of users. But besides the information hype, there is also a real demand. Not every player is ready to shell out those large sums that are now being demanded for flagship products. And if NVIDIA continues to successfully storm the graphics Olympus, then AMD this time went a different way, opening a new generation of Radeon with a mid-level model, which should bypass all competitors in its price category.

According to statistics cited by AMD, up to 84% of gamers use discrete graphics that cost $ 100-300, and 95% of gamers use 1920x1080 resolution. This large audience is aimed at the Radeon RX 480 video adapter, which will offer the optimal combination of performance and cost thanks to a new architecture, a new process technology, increased frequencies and a large amount of memory.

AMD Polaris architecture

The next generation Radeon is based on the Polaris architecture, which is an evolution of the GCN architecture. This is the fourth generation in this line. The novelty under consideration is codenamed Polaris 10. The graphics processor has 36 Compute Units (CU), which are organized into four Shader Engine arrays with their own geometry processing unit and rasterization units. Each CU operates 64 stream processors and four texture units, similar to the units in older GPUs. The result is 2304 stream processors, 144 texture units and 32 ROP units.

The general structure of the GPU resembles other AMD processors, or rather, a cross between Grenada (Hawaii) and Antigua, i.e. it is an intermediate option between the Radeon R9 390X and the Radeon R9 380X. At the same time, the efficiency of shader execution has been increased, the L2 cache size has been increased to 2 MB and work with it has been improved, the memory controller has been updated, geometry processing units and support for Async Compute have been improved, support for FP16 and Int 16 instructions has been added. frequencies provide additional acceleration.

According to AMD, the efficiency of one CU increased by 15% compared to the Radeon R9 290. When processing tessellation in conjunction with heavy AA modes, the efficiency increase can be double or even triple. Data compression is supported, which improves memory bandwidth. In particular, the Delta Color Compression algorithm is supported, which allows encoding the color difference. We talked about this technique in the description of the NVIDIA Pascal architecture. AMD also supports such compression on Radeon Fury X, but the efficiency of algorithms in Polaris 10 is higher. With such an increase in efficiency in data transmission, the chip is content with a 256-bit bus. The Radeon RX 480 uses GDDR5 memory chips with an effective data rate of 8 GHz.

Asynchronous shaders allow you to optimize the execution of a blended workload that combines graphics and non-graphics computing. Efficient load balancing is realized thanks to new hardware schedulers and the familiar Asynchronous Compute Engines (ACE) blocks.

The Polaris 10 graphics chip is made using the 14nm FinFET process technology, while the NVIDIA Pascal chips are manufactured at 16nm. This is a major breakthrough for an industry where all graphics were produced using a 28nm process technology for several years. Such a delicate technical process can significantly reduce energy consumption. And this task was initially one of the key in the development of the new generation. The engineers focused on the features of the new 3D transistors, optimizing the structure of the new crystal and implementing improved voltage control mechanisms. Among other things, crystals based on the new technical process differ less in their characteristics. If we again start from the Radeon R9 290 card, with which AMD compares the new product, then the increase in performance per watt is almost twofold.

For the Radeon RX 480, a TDP of 150 W is declared, which is close to the performance of the GeForce GTX 970. At the same time, the new product should be more productive. And if we talk about temperature and noise characteristics, then according to AMD measurements, the reference version of the Radeon RX 480 has a slightly lower acoustic noise.

The new technical process allowed to increase the GPU frequency to 1266 MHz, which is the maximum Boost value. In case of exceeding the power or temperature limit, the frequency can be gradually reduced. The guaranteed baseline is 1120 MHz. You can compare the characteristics with their predecessors in the table.

Video adapter	Radeon RX 480	Radeon R9 390	Radeon R9 290	Radeon R9 380X	Radeon R9 280X
Core	Polaris 10	Grenada	Hawaii	Antigua	Tahiti
	n / a	6020	6020	5000	4313
Process technology, nm	14	28	28	28	28
Core area, sq. mm	232	438	438	366	352
	2304	2560	2560	2048	2048
Number of texture units	144	160	160	128	128
Number of rendering units	32	64	64	32	32
Core frequency, MHz	1120-1266	Up to 1000	Up to 947	up to 970	1000
Memory bus, bit	256	512	512	256	384
Memory type	GDDR5	GDDR5	GDDR5	GDDR5	GDDR5
Memory frequency, MHz	8000	6000	5000	5700	6000
Memory size, MB	8192/4096	8192	4096	4096	3072
	12	12	12	12	12
Interface	PCI-E 3.0	PCI-E 3.0	PCI-E 3.0	PCI-E 3.0	PCI-E 3.0
TDP level, W	150	275	275	190	250

Among the features of the Radeon RX 480, it should be noted that there are two versions with different memory sizes. The base model is equipped with 8 GB, while the cheaper modification will receive 4 GB.

Video cards will receive support for AFR frame pacing technology for DirectX 12. This technique smooths out unevenness in frame output in CrossFire.

Along with DirectX 12 support, the video card is also compatible with the new Vulkan API. And besides simple gaming, the Radeon RX 480 may well cope with virtual reality VR. Optimal performance will be provided by support for AMD LiquidVR capabilities, which means the best distribution of computing resources for mixed tasks, support for Asynchronous Time Warp technology on the Oculus Rift for correct and fast image refresh when moving. This also includes AMD TrueAudio Next technology for correct rendering of sound wave propagation using ray tracing technology. Moreover, these calculations are also included in the scope of Async Compute. NVIDIA is developing a similar initiative. But AMD's option provides an open source toolkit for developers through the GPUOpen program.

Variable Rate Shading technology allows you to adjust the image quality of individual segments of the image during VR rendering, maintaining the maximum resolution for the central zone and decreasing it at the periphery. This saves resources and speeds up VR performance.

The Radeon RX 480 video adapter supports DisplayPort 1.3 HBR and is DisplayPort 1.4 ready with support for the new HDR standard. That is, in the future, you will be able to connect new HDR displays and watch the corresponding content. DisplayPort supports up to 5K @ 60Hz, 4K @ 120Hz or 4K @ 96Hz HDR.

Polaris also received a new H.264 and HEVC video encoding / decoding unit with support for resolutions up to 4K. Now you can record video from games in high quality or stream it right away. A good bonus for gamers, because previously, even on top-end Radeon, through the AMD Gaming Evolved client, it was possible to capture only Full HD video.

The Radeon RX 480 works with the new AMD Radeon Settings Software Center, which provides extensive functionality for adjusting color gamut settings or graphics card performance. There are currently no third-party overclocking utilities for Polaris, but all of these features are available in AMD's new WattMan app. To access the program in AMD Radeon Settings, go to the Games tab and then to Global Settings. Here you can fine tune the Boost or overclock the card by simply increasing the frequency scale. Available fan algorithm control, changing power and temperature limits.

After a brief overview of the architectural features, let's take a look at a real copy of the Polaris 10 graphics card.

Before us is a reference video card. It is made in the already recognizable style. The design is without frills, the cooler is of the "turbine" type, looks like a brick.

The length of the Radeon RX 480 reaches 24 centimeters. There are large Radeon logos on the case and fan.

The fee turns out to be very short. The fan hangs over the textolite from the side, in this place there are specially made holes for air flow.

The Radeon RX 480 is no longer equipped with DVI connectors, but there are three DisplayPort and one HDMI on the rear panel.

The case cover can be easily unscrewed without completely disassembling the device. This allows you to evaluate the complete cooling system. We see a large base and a separate aluminum heatsink on the GPU.

The metal base plate is ribbed to increase the heat dissipation area, including in the area of ​​the power unit. So the radiator of the power elements and memory chips is made very soundly.

On the other side, a radial fan is mounted on the base, which drives air through the fins of the main radiator.

The graphics chip cooler is simpler. No copper tubing, just a copper insert in the contact area. And the dimensions of the radiator, frankly, are too small. However, we are talking about a chip with a low TDP, so this design may be quite justified.

The printed circuit board is less than 18 centimeters. The assembly of the elements is very tight. The power supply system has six phases. There is one six-pin power connector in the corner.

The Polaris processor has no surface markings, all markings are located on the backing.

Eight gigabytes of memory are collected with Samsung K4G80325FB-HC25 microcircuits.

The GPU-Z utility correctly detects all the characteristics. The frequencies, as you can see from the lower screenshot, correspond to the recommended ones. GPU runs at Boost 1266 MHz, memory at 2000 MHz (8000 MHz effective value).

Testing was carried out on an open bench at 27 ° C indoor. Under these conditions, the card's temperature easily exceeded 80 ° C in all gaming tests. The Division peaked at 84 ° C at maximum graphics quality. The screenshot below shows the maximum parameters and the value of the core frequency at a particular moment in time (by aiming at a point on the graph).

The Metro: Last Light benchmark easily warmed up the core to 85 ° C. In both tests, the frequency varied, there were drops to 1180 MHz or less. However, 1200 MHz can be taken as average in heavy tests.

The noise is moderate, the fan spins up to 2200 rpm.

How to overclock a Radeon RX 480? Go to AMD Settings, "Global Settings".

In the settings, you will have to immediately set a high speed for the fan, because a standard cooler does not have much room for cooling during overclocking. Then we experiment with frequencies. It is also beneficial to increase the target temperature, after which a phased decrease in frequency begins. But with this you need to be careful and avoid overheating. At maximum fan rpm, we raised this limit by 4 ° C, which helped to increase the average Boost in high operating temperatures.

The final overclocking was only + 4.5% to the initial core frequency. But taking into account the increase in the temperature limit, the real difference in Boost may turn out to be slightly higher. The memory worked stably at 8720 MHz. With a frequency configuration of 1235/8720 MHz, we managed to pass all tests, higher frequencies could lead to failures.

The increase is small, but the noise increases seriously. Cooling works at its limit and at peak moments howls at all 5000 rpm. In a number of tests, the frequency strove to a maximum of 1325 MHz, but in Metro: Last Light there were drops below 1300 MHz. Such a moment is reflected in the lower screenshot.

As a supplement, we present a screenshot of the mining program on the Radeon RX 480 at nominal frequencies.

Characteristics of the tested video cards

The reviewed video card will be compared with the main competitor in the face of the GeForce GTX 970. The usual version of the rival will be replaced by the MSI GTX 970 Gaming 4G. The powerful cooling gives the MSI card the advantage of constant maximum boost. To bring the performance closer to that of the reference GeForce GTX 970 with floating boost, MSI's clocks are calibrated so that the maximum Boost does not exceed 1200 MHz in gaming tests and 1220 MHz in 3DMark tests.

In some applications, there will be additional modes, which are compared with the top models from AMD and NVIDIA. Therefore, we present the characteristics of all participants in the table.

Video adapter	Radeon RX 480	Radeon R9 Fury X	GeForce GTX 1070	GeForce GTX 980 Ti	GeForce GTX 970
Core	Polaris 10	Fiji	GP104	GM200	GM204
Number of transistors, mln.pcs	n / a	8900	7200	8000	5200
Process technology, nm	14	28	16	28	28
Core area, sq. mm	232	596	314	601	398
Number of stream processors	2304	4096	1920	2816	1664
Number of texture units	144	256	120	176	104
Number of rendering units	32	64	64	96	56
Core frequency, MHz	1120-1266	Up to 1050	1506-1683	1024-1100	1051-1178
Memory bus, bit	256	4096	256	386	256
Memory type	GDDR5	HBM	GDDR5	GDDR5	GDDR5
Memory frequency, MHz	8000	1000	8000	7010	7010
Memory size, MB	8192	4096	8192	6144	3584 + 512
Supported DirectX Version	12	12	12.1	12.1	12
Interface	PCI-E 3.0	PCI-E 3.0	PCI-E 3.0	PCI-E 3.0	PCI-E 3.0
Power, W	150	275	150	250	145

Test stand

The test bench configuration is as follows:

• processor: Intel Core i7-6950X (3, [email protected], 1 GHz);
• cooler: Noctua NH-D15 (two NF-A15 PWM fans, 140 mm, 1300 rpm);
• motherboard: Gigabyte GA-X99P-SLI;
• memory: G.Skill F4-3200C14Q-32GTZ (4x8 GB, DDR4-3200, CL14-14-14-35);
• system drive: Intel SSD 520 Series 240GB (240 GB, SATA 6Gb / s);
• additional disk: Hitachi HDS721010CLA332 (1 TB, SATA 3Gb / s, 7200 rpm);
• power supply unit: Seasonic SS-750KM (750 W);
• monitor: ASUS PB278Q (2560x1440, 27 ″);

• operating system: Windows 10 Pro x64;
• Radeon RX 480 driver: AMD Crimson 16.6.2.
• Radeon R9 Fury driver: AMD Crimson 16.5.3.
• GeForce GTX 1070 driver: NVIDIA GeForce 368.39;
• GeForce GTX 1080 driver: NVIDIA GeForce 368.25;
• driver GeForce GTX 980 Ti: NVIDIA GeForce 368.22.

The test methodology described in one of the previous articles is taken as a basis. But since the test configuration was used there for top-end video cards, not all modes and applications are involved in this comparison. In some cases, when the graphics quality is forced to decrease, only Radeon RX 480 and GeForce GTX 970 are compared. In other cases, where changes were not made to the settings of test applications, their results were supplemented by the results of the flagship video cards.

Test results

Batman: arkham knight

The Radeon RX 480 beats the GeForce GTX 970 in Arkham Knight. The newcomer AMD demonstrates the performance level of its overclocked competitor in nominal terms. Increasing the frequencies allows you to gain a few more percentages.

Battlefield 4

The situation is different in Battlefield 4. There is already an advantage for the GeForce GTX 970, and the Radeon RX 480 already needs to be overclocked to get it closer to the rival.

DiRT Rally

We can talk about parity between the newcomer AMD and the GeForce GTX 970 at the initial frequencies. In overclocking, the second gets the advantage. Both are significantly behind the top solutions.

DOOM

In the new DOOM, the difference between senior and junior video cards is not so critical, but it will still not be possible to catch up with them. The strange result of the GeForce GTX 1070 can be written without optimization problems. Anything before the Radeon RX 480, then it overtakes the GeForce GTX 970 only if its frequencies are increased.

Fallout 4

In Fallout 4, we re-ran the tests in the usual Ultra-mode, so the older video cards from the previous reviews were not included in the comparison. At initial frequencies up to 5%, Radeon wins over its rival, but after overclocking the balance changes in favor of GeForce.

Far cry primal

The hero of the review wins more than 11% of the GeForce GTX 970 in Far Cry Primal when compared in nominal modes. Opponents are equal in overclocking. Acceleration itself gives an acceleration of about 9%.

Gears of War: Ultimate Edition

First surprise from a beginner. At maximum texture quality, the Radeon RX 480 demonstrates a slight lag behind the Radeon R9 Fury. With such textures, the game needs more than 4 GB, which limits the potential of the AMD flagship. For the same reason, at the end of the rating, the GeForce GTX 970 with its combined memory, where only 3.5 GB is effectively used. It is logical to assume that in the case of lowering the quality of textures to the usual level, the difference between the rivals will decrease.

Grand theft auto 5

A slight advantage over its rival in the Radeon in GTA 5 at the initial frequencies. After overclocking, the situation is the opposite, but the difference is not dramatic.

Just Cause 3

The Radeon RX 480 is 5-11% faster than the competitor in Just Cause 3, and even after overclocking it retains a tiny advantage. It is noteworthy that the accelerated Radeon RX 480 lags behind the Radeon R9 Fury X by only 10% - a good result!

Metro: Last Light

In Last Light, we ran two tests. With simpler settings, they compared our competitors in the mode that they can handle. Additionally, we compared them with the tops of the SSAA.

A slight lag behind the opponent in terms of value and more significant after overclocking. At the same time, it's still nice that you can play comfortably even in 2K.

There is no talk of competition with the tops. The gap between the Radeon RX 480 and the Radeon R9 Fury X reaches 51%. The gain from overclocking is 9%.

Quantum Break

Since the first tests, the results of the GeForce GTX 970 in Quantum Break have improved. But even after overclocking, this rival is weaker than the Radeon RX 480 in the nominal value. The lag of our hero from Fury X at the level of 25%. The merit of this is both in the updated architecture and in the large amount of memory (the game is demanding for this).

Rise of the Tomb Raider

First, let's compare the main rivals in Full HD with a very high quality profile.

Rise of the Tomb Raider is known for its heavy memory requirements. Therefore, the slight lag between the GeForce GTX 970 and the Radeon RX 480 can be considered surprising. In overclocking, the opponent even takes the lead.

If you put the fighters with older video adapters in a more difficult mode, then no one will cope with the task, except for the flagship GeForce. Note the tiny difference between the Polaris 10 and the Fury X. Given that the game uses more than 7GB in this mode, this difference is not that surprising. Here, rather, the performance of the GeForce GTX 970 raises a question - we expected the worst results from the accelerator.

The Witcher 3: Wild Hunt

Playing The Witcher 3 at 2K will be tough, but the 30 fps bar is easily overcome by the new Radeon. And this is also an impressive result for a representative of the middle class. The advantage over the younger GeForce is at the level of 4-9%; in overclocking, the opponent wins back a little.

Tom clancy "s the division

The Division is also beyond the strength of the Radeon RX 480 in 2K mode, but we can compare opponents in extreme conditions. And again our hero is better, although in overclocking the GeForce breathes in the back again. The difference between the Radeon RX 480 and the Radeon R9 Fury X is up to 38% in average frame rate.

Total War: Warhammer

New test in a new game. A special benchmark with DirectX 12 support was used.

The results clearly speak in favor of the Radeon RX 480. The rival is still weaker after increasing its frequencies. Scalability of performance during overclocking is weak for both participants, which may be due to the peculiarities of the benchmark.

XCOM 2

Last playtesting in XCOM 2. The game can bring older video cards to their knees with heavy anti-aliasing. We'll restrict ourselves to the Ultra profile with simple FXAA.

Initially, the Radeon RX 480 is closer to the level of a forced rival. But the best frequency potential of the second allows him to equalize the chances after overclocking.

3DMark 11

The Radeon RX 480 is 5% behind its competitor in this test, overtaking it only after increasing frequencies.

3DMark Fire Strike

But here the situation is different, and the Radeon RX 480 is immediately ahead with a margin of more than 6%. When it comes to overclocking, the opponent comes forward again.

Energy consumption

The measurements were made according to the previously described method, but without taking into account the data of the older video cards in Total War: Attila.

Almost identical indicators are for Radeon RX 480, GeForce GTX 970 and GeForce GTX 1070. It seems to be not a very significant achievement for Radeon, but against the background of gluttonous Radeon R9 290/390 this is a serious result. The sharp increase in power consumption during overclocking is not encouraging. It looks like every additional percentage to the core frequency will be difficult.

conclusions

Based on the test results, we can note similar results for video cards Radeon RX 480 and GeForce GTX 970. De facto, in nominal value, the advantage is more often on the side of AMD's new products, but the rival wins back when overclocked. In DirectX 12, the situation is more unambiguous and it is clearly in favor of the Radeon RX 480. On the Radeon side, there is a large amount of memory, which some games can already use. Due to such a volume, one can even observe a funny situation in Rise of the Tomb Raider, where it is possible to catch up with the Radeon R9 Fury X. But in general, it is not worth equating the Radeon RX 480 and the Radeon R9 Fury X, these are solutions of different levels. It's nice to note that the potential of the video card allows you to play not only in Full HD, it pulls out many games even in 2K mode. In its price category, the Radeon RX 480 looks great - faster than its main competitor, more promising in DirectX 12 and at the same time cheaper.

The new 14nm process technology provides a low level of power consumption, but the video card cannot be called cold. In order to make the Radeon RX 480 the most affordable offer on the market, the manufacturer saved a little on cooling. The native cooler copes with the nominal mode, but it has no room for overclocking. Also, during overclocking, power consumption rises sharply. It looks like the initial frequencies are close to the maximum, and then there is not much to squeeze out. But experimenting with good cooling makes sense, you will benefit from it. You just have to wait for non-reference versions of Radeon RX 480 or spend money on CBO.

Among the advantages of the Radeon RX 480, it is worth mentioning improved VR support, the ability to work with HDR and hardware video encoding / decoding of ultra-high definition video. And if in terms of performance this is not the most powerful offering from AMD, then it is definitely the most progressive at the moment.