AMD RDNA 3 Graphics Architecture

The new Radeon RX 7600 XT is based on the latest RDNA 3 graphics architecture by AMD, which aims to take advantage of newer architecture-level advancements, newer components such as the AI accelerators, 2nd gen Ray accelerator, and MDIA, to provide generational speedups. For the higher end models in the Radeon RX 7000 series, starting the RX 7700 XT and upward, AMD is using 5 nm EUV foundry process, at least where it matters—the shader engines. The RX 7600 series, on the other hand, is based on the monolithic Navi 33 silicon that's based on the slightly older 6 nm DUV foundry node, which AMD thinks has sufficient electrical and thermal characteristics for a mid-range GPU that can be produced in a cost-efficient manner, with room for price-cuts. The silicon comes with a die-area of 204 mm², and a transistor count of 13.3 billion.

The 6 nm Navi 33 silicon, as we mentioned, physically features a maximum of 32 CU (compute units). This works out to 2,048 stream processors. Each CU contains two AI accelerators, and a 2nd gen Ray accelerator, which works out to 64 AI accelerators, and 32 Ray accelerators in all for the RX 7600 XT. Other critical specs include 128 TMUs, and 64 ROPs. The chip's 128-bit GDDR6 memory interface drives 16 GB of memory on the RX 7600 XT, which ticks at 18 Gbps, giving it 288 GB/s of memory bandwidth on tap. This is cushioned by a 32 MB on-die Infinity Cache. AMD hasn't released an architectural diagram for the Navi 33, yet.


Much of the architectural innovation is this generation is with the RDNA 3 Dual-Compute Unit (or Compute Unit pair). The "Navi 33" GPU physically features 32 compute units spread across two Shader Engines. AMD claims that at the same engine clocks, the RDNA 3 CU offers a 17.4% IPC increase over the RDNA2 CU.


The new RDNA 3 CU introduces multi-precision capability for the 64 stream processors per CU: operating either as 1x SIMD64 or 2x SIMD32 units. The Vector Unit that houses these SIMD units can either function as a SIMD execution mechanism, or as a Matrix execution unit, thanks to the new AI Matrix Accelerator, which provides a 2.7x matrix multiplication performance uplift versus conventional SIMD execution. Also added are support for the Bfloat16 instruction-set, and SIMD8 execution. The GPU hence enjoys AI hardware-acceleration that can be leveraged in future feature-additions relevant to gamers, such as FSR 3.0. Game developers will also look for ways to exploit accelerated AI, now that all three brands feature it (NVIDIA Tensor cores and Intel XMX cores).


AMD's first-generation Ray Accelerator, introduced with the RDNA2 architecture, was the result of a hasty effort to catch up to NVIDIA with a DirectX 12 Ultimate GPU, where they developed a fixed-function hardware to calculate ray intersections, and offloaded a large chunk of RT processing to the generationally-doubled SIMD resources. With RDNA 3, they've refined the Ray Accelerator to achieve an 80% ray tracing performance uplift over the previous generation, when you add up the Ray Accelerator count, their higher engine clocks, and other hardware-level optimizations, such as early sub-tree culling, specialized box sorting modes, and reduced traversal iterations.


There is a 50% ray intersection capacity improvement for RDNA 3 thanks to these optimizations, and cycles-per-ray reduction. Besides these, AMD has also made several improvements to the geometry- and pixel-pipes, with the introduction of the new multi-draw indirect accelerator (MDIA), which reduces CPU API and driver-level overheads by gathering and parsing of multi-draw command data. At the hardware-level 12 primitives per clock is now supported compared to 8 per clock on RDNA2, thanks to culling. The core-configuration overall enables 50% more rasterized performance per clock.


AMD has significantly improved the Display Engine of "Navi 33" over the previous-generation in terms of connectivity. The new Radiance Display Engine comes with native support for DisplayPort 2.1, which enables 8K output at up to 165 Hz refresh-rate, or 4K at up to 480 Hz, with a single cable. AMD has refined its FSR 2 algorithm to support 8K (i.e. render at a lower resolution with FSR-enhanced upscaling), to make it possible to enjoy the latest AAA titles at playable frame-rates on 8K displays. The RX 7600 XT gets two full-size DP 2.1 connectors, besides an HDMI 2.1b, and a USB-C with DP 2.1 passthrough. Unlike with the RX 7600, board partners cannot opt to have DisplayPort 1.4a for RX 7600 XT cards. The "Navi 33" silicon receives full hardware-accelerated AV1 encode and decode capabilities. With this generation, AMD is also introducing SmartAccess Video, a feature that lets the AMD driver leverage the hardware encoders of the RDNA2 iGPU of Ryzen 7000 desktop processors, for additional encoding performance.

FidelityFX SuperResolution 3 Fluid Motion Frames (FSR 3 and FMF)

As part of the Radeon RX 7800 XT and RX 7700 XT announcements, AMD finally announced the much awaited FidelityFX Super Resolution 3 and Fluid Motion Frames. FSR 3 is being announced as a technological rival to NVIDIA DLSS 3 Frame Generation. The premise with both technologies is the same—to effectively double frame-rates by generating alternate frames without running them through the entire graphics rendering pipeline, it's just that the two technologies differ in their approach to this goal.

FSR 3 builds on FSR 2 with its updated super resolution upscaler promising generational quality improvements at a every given rendering resolution. Fluid Motion Frames (FMF) isn't the entirety of the FSR 3 feature-set, but is its most important feature-addition. FMF is a frame interpolation technology much like the one consumer televisions come with. Alternate frames are generated as an approximate of two frames. Where FMF differs from DLSS 3 Frame Generation is that while NVIDIA uses a hardware component called optical flow accelerator and the GPU's AI acceleration to generate an intermediate frame without involving the graphics rendering pipeline, FMF uses a certain amount of the graphics rendering pipeline. At a hardware level, FMF uses the main SIMD machinery of the GPU, leveraging asynchronous compute. As with DLSS 3 FG, FSR 3 FMF comes with added latency. NVIDIA counteracts this with Reflex, while AMD uses Radeon AntiLag+. Both technologies try to keep the frame queue short to reduce whole system latencies.

One major advantage FSR 3 FMF enjoys over DLSS 3 FG is that it works on any modern DirectX 12 GPU that supports async compute, since it doesn't require a specific hardware component the way DLSS 3 FG requires the Optical Flow Accelerator on NVIDIA "Ada" GPUs. The only limiting factor here is the performance. To be more specific, AMD says that all Radeon GPUs from RX 5700 series onward; and all GeForce GPUs from RTX 20-series onward, should support FSR 3 FMF. Also, FSR 3 FMF is as easy to integrate with games as FSR 2 is. The first games implementing FSR 3 FMF should arrive in Fall 2023. AMD is also working to extend FMF to Radeon Super Resolution, the driver-level technology that enables performance upscaling to even games that don't support FSR.

HYPR-RX

HYPR-RX is an interesting new feature in the AMD Software (control center) application. It is a one-click performance boosting technology that works with any DirectX 11 or DirectX 12 game. The software is a cocktail of Radeon Boost, Radeon AntiLag+, and Radeon Super Resolution, and applies the three features on any running game as needed, automatically. Radeon Boost improves performance by dynamically reducing the render resolution of a game when there's too much motion on the screen (and hence not enough detail needed). Radeon Super Resolution improves frame-rates as it applies FSR on the output of a game rendered at a lower resolution (including the frames lowered in resolution by Radeon Boost. AntiLag+ counteracts the latency added by these two, by shortening the frame queue. AFMF is part of the HYPR-RX feature-set, which can be activated with a single click.