Chris Hall, Senior Director Software Development and head of AMD's ROCm project, which is a software stack for AI development and roughly equivalent to Nvidia's CUDA platform, explained that FSR Redstone—which could be called FSR 5 when it hopefully releases later this year—was developed using code from a ROCm project known as AMD ML2CODE (Machine Learning to Code).
Apparently, FSR Redstone's neural rendering technology is "converted" into compute Shader code. "This means that FSR Redstone's neural rendering core can also run on GPUs made by other companies," Hall said (via machine translation).
In other words, it doesn't require AI or matrix math cores, just as Nvidia's DLSS technology requires Tensor cores. It can run on conventional GPU shaders. "ML2CODE-based frameworks, such as FSR Redstone, can be seamlessly integrated directly into DirectX or Vulkan graphics pipelines with minimal latency. We believe that the ML2CODE solution is the best way, at least for now, to integrate and deploy 3D graphics and AI technologies." Hall says.
That's a pretty explicit indication that AMD favours running AI-enhanced upscaling on the shader pipeline as opposed to fitting out GPUs with dedicated AI cores, the latter being Nvidia's approach over multiple generations of RTX GPUs and their Tensor cores.
However, Hall does foresee future AMD graphics architectures that will have dedicated AI cores that are compatible with Microsoft's DirectX Cooperative Vector technology. "Cooperative Vector is a DirectX programming model. It's a framework that provides a way for programmable shader units to perform calculations using other accelerators. In other words, it's a framework that allows shader programs to utilize AI accelerators within the GPU. This is an excellent approach for implementing various AI styles.
"However, currently, significant delays can occur unless the GPU has an architecture specialized for Cooperative Vector technology. It's true that we're actively working on supporting Cooperative Vector," Hall says.
The biggest gaming news, reviews and hardware deals
Keep up to date with the most important stories and the best deals, as picked by the PC Gamer team.
All of this scans perfectly in the context of AMD's existing GPU technology. Even AMD's very latest RDNA 4 GPUs, such as the Radeon RX 9070 XT, don't have dedicated hardware AI cores equivalent to Nvidia's Tensor cores. Instead, RDNA 4 has optimisations to the shader engines to improve matrix math performance. AMD describes this capability in the shader engine as an "AI accelerator," but it's not entirely clear whether that constitutes discrete hardware you can physically point to in the chip as opposed to add capability in the general purpose shader ALUs.
"ML2CODE-based Compute Shader approach is highly versatile and delivers optimal performance on many current GPUs, including NVIDIA GPUs," Hall says, adding that the two approaches of running ML upscaling code either in the shader pipeline or on dedicated AI cores, "may merge or be integrated in the future. I'm not sure when that will happen, though."
Could older GPUs like the RX 7800 XT support the Redstone update for AMD's FSR upscaling tech? (Image credit: Future)
Of course, the obvious caveat here is that "can" is not the same as "will." We can confidently say from Hall's comments that it does not seem like there are any technical reasons why AMD's FSR Redstone couldn't run on Nvidia and indeed Intel GPUs. But we can't say there won't be commercial or political reasons why AMD doesn't make that happen.
For gamers, it's definitely preferable if upscaling and related technologies like frame generation are platform agnostic. That way you could buy an Nvidia or AMD GPU then use the best upscaler available for a given game. What's more, the way that AMD has built FSR Redstone to run on shader cores opens up the possibility that it might be compatible with older AMD GPUs, like the RDNA 3-based RX 7000 series, including the RX 7800 XT.
However, Nvidia's approach currently locks the DLSS platform down to its own GPUs. So, there's no immediate prospect of a fully open approach to upscaling. Indeed, locking its own upscaling down while AMD takes a more open approach seems to hand the advantage to Nvidia. After all, if you want to buy a GPU that supports both of the main GPU vendors' upscaling tech, for the foreseeable future it looks like your only chance will be Nvidia graphics cards.
Jeremy has been writing about technology and PCs since the 90nm Netburst era (Google it!) and enjoys nothing more than a serious dissertation on the finer points of monitor input lag and overshoot followed by a forensic examination of advanced lithography. Or maybe he just likes machines that go “ping!” He also has a thing for tennis and cars.
You must confirm your public display name before commenting
Please logout and then login again, you will then be prompted to enter your display name.
