AMD may be planning to take a different direction with FSR in the future, and if a clue highlighted on Twitter is anything to go by, Team Red’s intention is to develop the framerate boosting technology along AI-powered lines – as well as Nvidia DLSS.
This theory is based on a tip tweeted by well-known hardware leaker Greymon55, who detected a new commit in the LLVM repository that is about the introduction of WMMA (Wave Matrix Multi-Accumulate) instructions in GFX11.
AI accelerate! It should work with FSR3.0 and other features. https://t.co/Wh31UxdoaIJune 29, 2022
Okay, this sounds like a lot of gobbledygook, so how about a translation? Essentially, GFX11 refers to AMD’s next-gen RDNA 3 GPUs (and Radeon Pro cards), with WMMA instructions being a way to really boost machine learning (AI) operations. So this could point to FSR – perhaps in version 3.0 – making use of such AI cuts to get better quality upscaling results.
One theory that has emerged behind this is that perhaps RDNA 3 graphics cards offer some sort of built-in hardware functionality along the lines of Nvidia’s tensor cores – dedicated AI processors in RTX graphics cards, which are used as muscle to trigger DLSS. .
Analysis: Will AMD take the next step into the avenue of AI?
Obviously, we need to be very careful when drawing big conclusions from a single commit. And that certainly doesn’t mean that high-end RDNA 3 graphics cards will necessarily come with specific hardware designed to accelerate AI workloads (like Nvidia’s tensor cores).
There are some important things to consider here. Let’s not forget that AMD just released FSR 2.0 and with this technology many new moves have been made, including the transition to temporal upscaling, offering huge improvements over spatial upscaling (as used by FSR 1.0). We’ve seen this make a big difference in supported games, and comparisons indicate some pretty impressive image quality results for FSR 2.0 compared to DLSS 2.0, with Team Red catching up in style.
Furthermore, AMD argued that FSR 2.0 is an effective rival to DLSS without the need for any AI-powered acceleration (or even dedicated hardware on the GPU to improve these operations – which takes up space that could be used to boost other aspects of the GPU). GPU actuation). In fact, recently AMD has greatly downplayed the influence and importance of AI in upscaling technologies…
So for AMD to be looking to outfit its RX 7000 graphics cards with some sort of equivalent for tensor cores seems like a pretty old stretch, especially since we haven’t heard anything from the rumor mill about it so far – and these newer generations GPUs aren’t too far off the mark. launch (possibly arriving in October).
Certainly, RDNA 3 boards can make use of these WMMA instructions in some way to further optimize FSR 2.0 perhaps, but any implementation of FSR 3.0 using machine learning and dedicated hardware combined with this is certainly a good way to go. What this seems to point out is that AI is indeed the way AMD is looking ahead (despite recent comments and doubts cast about exactly how much difference it really makes to DLSS).
In that case, the plan might be to keep FSR 2.0 in use going forward after the next generation implementation (FSR 3.0) arrives, just like with the release of FSR 2.0, games are still getting FSR 1.0 support (since the latter might not be as good, but it offers wider coverage for lower spec GPUs).
Who knows, ultimately, but what’s clear enough is that AMD built and pushed FSR to be an open standard and usable not only on their own Radeon graphics cards (even on slightly older models) but also on Nvidia and Intel GPUs. So whatever happens, presumably AMD won’t want to turn their backs on this core philosophy for frame rate boosting in the future.
Through Wccftech (opens in new tab)