AMD is back in the big leagues of graphics cards! With these words, you can start reviewing the new generation of Radeon RX 6000. In recent years, we have become accustomed to the dominance of NVIDIA in the graphics market. If AMD had powerful competitors in the budget segment, then GeForce reigned supreme in the top segment. There was an attempt to win back the position with the release of the Radeon VII, but this model failed to show impressive results and was released in an extremely small edition. An important step was the release of the Radeon RX 5700 series based on the new RDNA architecture, which was later improved and finalized. As a result, we got full-fledged flagships in the face of the Radeon RX 6900 XT and Radeon RX 6800 XT on the new GPU with twice the number of compute units and hardware support for ray tracing.
Architecture AMD RDNA 2
At the moment, three new generation video cards are presented — Radeon RX 6900 XT, Radeon RX 6800 XT and Radeon RX 6800. All of them are based on the Navi 21 GPU in its various variations. The chip is divided into 4 Shader Engine clusters of 10 Compute Engine (CU). In fact, this is a double version of Navi 10 — 80 CU for the new GPU versus 40 for the old one.
Each CU still operates 64 cores with 4 texture units. The full Navi 21 boasts 5120 cores and 320 texture units with 128 ROPs. But this configuration is only used in the Radeon RX 6900 XT, the Radeon RX 6800 XT has 8 CUs disabled, the Radeon RX 6800 has 20 CUs and part of the ROP disabled.
The processors are manufactured using the 7nm process technology at TSMC factories, like the previous generation. At the same time, AMD engineers redesigned the block structure and optimized the Big Navy pipeline for more compact dimensions and better power efficiency. And this made it possible to raise frequencies without switching to new technological standards. If we talk about specific numbers, then the increase in energy efficiency was more than 50%, and with the same power, RDNA2 can operate at 30% higher frequencies.
The frequencies of the new models have really become higher, and now more than 2 GHz in nominal value is the norm for them. The older Radeon RX 6800/6900 XT have Boost up to 2250 MHz with an average Game Clock value of 2015 MHz, the minimum base value is 1825 MHz. It should be understood that the frequencies depend on power limitations and temperatures, therefore, they may differ in different applications. But, as we will see below, they are closer to a higher Boost Clock value.
The full characteristics of the new video cards in comparison with previous models are presented in the table:
|Video adapter||Radeon RX 6900 XT||Radeon RX 6800 XT||Radeon RX 6800||Radeon VII||Radeon RX 5700 XT|
|Core||Navi 21||Navi 21||Navi 21||Vega 20||Navi 10|
|Number of transistors, million pieces||26800||26800||26800||13200||10300|
|Process technology, nm||7||7||7||7||7|
|Core area, sq. mm||520||520||520||331||251|
|Number of stream processors||5120||4608||3840||3840||2560|
|Number of texture blocks||320||288||240||240||160|
|Number of render units||128||128||96||64||64|
|Ray Accelerator Cores||80||72||60||—||—|
|Base core frequency, MHz||1825||1825||1700||1400||1605|
|Game Clock frequency, MHz||2015||2015||1815||—||1755|
|Boost Clock, MHz||2250||2250||2105||1750||1905|
|Memory bus, bit||256||256||256||4096||256|
|Memory frequency, MHz||16000||16000||16000||2000||14000|
|Memory size, GB||16||16||16||16||8|
|Supported version of DirectX||12 Ultimate (12_2)||12 Ultimate (12_2)||12 Ultimate (12_2)||12 (12_1)||12 (12_1)|
|Interface||PCI-E 4.0||PCI-E 4.0||PCI-E 4.0||PCI-E 3.0||PCI-E 4.0|
|Recommended price at the time of announcement, $||999||649||579||699||399|
Now let’s talk more about the main architectural changes. The Compute Engine has special Ray Accelerator blocks for accelerating ray tracing — one block per CU and 80 for the entire GPU. The Ray Accelerator performs ray intersection and geometry calculations, up to four intersections per clock. Blocks are simpler than RT-cores in NVIDIA processors, but this helps to save the transistor budget.
This is the first generation of AMD Radeons that are suitable for ray tracing games, and they should work in all previously released projects. According to AMD’s own tests, the potential of older models allows you to play with tracing effects at a resolution of 2560×1440.
It is also important to note that the new game consoles from Microsoft and Sony use graphics on a similar architecture, therefore, we can count on some optimization of tracing for the hardware capabilities of the RDNA2 GPU.
Another important change was a redesigned memory system and the addition of an additional L3 Infinity Cache. This fast 128MB cache reduces the number of external memory accesses. Data exchange with Infinity Cache is 1024 bits per clock at a variable frequency up to 1.94 GHz.
The throughput of Infinity Cache is several times faster than with 14 Gb/s GDDR6 memory on a 256-bit bus, and even higher than it would be on a 512-bit bus. This approach made it possible to simultaneously reduce power consumption, which is very important for such a massive complex chip.
In addition to ray tracing through the new GPU, it also supports other DirectX 12 Ultimate API features. These are Variable Rate Shading, Mesh Shaders and Samper Feedback.
Variable Rate Shading technology allows you to change the quality of shaders for different frame segments. The most obvious example is racing games, where blurring is actively used on the periphery of the frame and precision is not needed. NVIDIA has supported the technology since the Turing generation. VRS is already supported in a number of games, starting with Wolfenstein II: The New Colossus. AMD cites DIRT 5 as a newer example.
Mesh Shading optimizes the work with geometry through a special graphics pipeline. Samper Feedback improves texture streaming.
Not only hardware improvements are important, but also software innovations. The company is developing the AMD FidelityFX toolkit as part of the GPUOpen open platform. They allow you to add a number of effects using software optimizations for the Radeon GPU. In particular, players already know the FidelityFX CAS (Contrast Adaptive Sharpening) technology to improve picture clarity.
Now they will be replenished with new effects. FidelityFX Ambient Occlusion uses the optimized Combined Adaptive Compute Ambient Occlusion (CACAO) algorithm to faithfully shade objects in dynamic scenes. Provides several quality presets to select the optimal mode.
FidelityFX Screen Space Reflections uses AMD Stochastic Screen Space Reflections (SSSR) technology to create high quality screen space reflections.
FidelityFX Variable Shading uses a special variant of VRS to improve performance.
When using tracing, artifacts appear in the form of noise. NVIDIA solutions implement deep learning-based noise reduction, and this operation is performed quite efficiently thanks to special tensor blocks. Navi 21 does not have special blocks for AI algorithms, but it introduces its own version of FidelityFX Denoiser noise reduction.
Another performance optimization tool is FidelityFX Downsampler. FidelityFX HDR Mapper technology is also introduced to make HDR work effectively with FreeSync monitors.
Announced technology FidelityFX Super Resolution, which should become an alternative to NVIDIA DLSS. This is a smart scaling method based on the DirectML API, which is part of DirectX 12. Details have not yet been disclosed. It can be assumed that without special blocks, the efficiency of this method will be somewhat worse than NVIDIA DLSS. But it will be an open source algorithm for DirectX 12, which could make it more mainstream and popular.
At the moment, FidelityFX technologies are announced for 35 games. They are supported by recent DiRT 5, Godfall and World of Warcracft: Shadowlands (FidelityFX CAS Fidelity, FX Ambient Occlusion, etc.), they will also be in Far Cry 6. All voiced games must support ray-traced shadows, which is available on all AMD solutions Radeon RX 6000.
AMD Radeon Boost
Among the new features in Radeon Software is Radeon Boost technology for faster gaming performance. Its essence is the dynamic reduction of resolution during sudden movements.
When exploring locations slowly, you will get the highest quality picture, but if you decide to move the mouse to quickly turn around, the resolution will decrease to provide improved performance. A compromise option that will add smoothness in dynamic moments with a short-term deterioration in visualization.
The technology is supported in many popular games — Fortnite, Overwatch, GTA V, Battlefield V, Destiny 2, The Witcher 3 and others.
AMD Radeon Anti-Lag
The following technology is designed to improve comfort in fast-paced eSports games. Radeon Anti-Lag reduces the delay between the moment you press a key and the reaction on the screen.
It’s not hard to guess that this is a response to NVIDIA Reflex.
AMD Smart Access Memory
Smart Access Memory technology is introduced, which allows the processor to access the entire amount of video memory through the PCI Express bus. Previously, a 256 MB segment was allocated for addressing. AMD Smart Access Memory allows all 16GB of the Radeon RX 6800/6900 to be addressed.
According to AMD tests, this technology provides acceleration from 0 to 15%. It can be enabled on AMD X570 and B550 motherboards in conjunction with Ryzen 5000 processors. That is, the combination of AMD components will ensure their maximum performance in games.
Smart Access Memory is a variation of Resizable BARSupport available on Intel solutions. And there are already test results for the Radeon RX 6800 XT on the LGA1200 platform, which we wrote about in the news. It is also known that NVIDIA will soon present its analogue of this technology without reference to AMD or Intel.
Video processing and image output
The video processing block has been updated. The new Radeons provide hardware accelerated video playback up to 8K at 24 or 30 frames. Added support for the AV1 codec. Supports hardware encoding up to 4K at 90 frames (H.264) or 4K at 60 frames (H.265).
We must admit that Radeon is still inferior to GeForce RTX, since Ampere support encoding even at 8K.
For image output, an HDMI 2.1 port is provided with support for formats up to 8K @ 60 Hz and 4K @ 144 Hz, there are two DisplayPort and one USB Type-C with support for virtual reality helmets. FreeSync and Variable Refresh Rate technologies are supported.