Intel Considers Adopting AMD FidelityFX Super Resolution for Its Upcoming Xe-HPG Graphics Cards

[Tsing]

Image: Intel



Intel’s upcoming desktop gaming GPUs might include support for AMD’s new spatial upscaling technique for improved performance. This is according to a recent Twitter exchange between HardOCP’s Kyle Bennett and Intel graphics chief Raja Koduri, the latter of whom confirmed that blue team was “definitely looking” at leveraging AMD’s newly unveiled FidelityFX Super Resolution technology for its Xe-HPG graphics cards. During its Computex 2021 presentation on Monday, AMD stressed that FidelityFX Super Resolution would be available on a wide range of products that not only included its own Ryzen and Radeon range, but NVIDIA’s GeForce graphics cards as well.



Definitely looking at it – the DL capabilities of Xe HPG architecture do lend to approaches that achieve better quality and performance. We will definitely try to align with open approaches to make ISVs job...

Continue reading...

 

[Peter_Brosdahl]

I'm all for an open standard that everyone can get behind but here's the rub of it. It's obvious that each tier of GPU is getting hammered to provide better performance while the workloads increase and at some point that single die can only do so much. Yes chiplets and the like are in the works but it still points to the fact that extra hardware is needed and not more compromises with an open standard. SLI is dead and mGPU is something almost unheard of despite claims of what could be. If AMD pulls out a magic rabbit, and Intel does too, to take on the heavy workload of ray tracing along with whatever is next, without tanking FPS then that's great, but how many really expect that to happen in the next couple of years?

I will give AMD credit with their latest round of GPUs in 2020/2021 but they've still got a ways to go. It's been awesome seeing them edge into top tier enthusiast segments. I'll also give them credit for coming up with interesting sampling techniques that are more efficient. I may not always like CAS but it can help versus other techniques but it still has a performance hit. The rumors are that they hope to be closer to the top and maybe that's part of the plan for ongoing evolution of FidelityFX. If we got the power we can spare some for FX.

Intel still has yet to release anything to the public to do the same but we know some day they plan to. So far we hear rumors of Xe-HPG and it could be great but until its out in the wild proving itself we come back to the same performance limitations of what does exist. Meanwhile NV saw the writing on the wall and came up with a solution that has evolved into a viable alternative. Regardless of how they want to label it, it came down to adding hardware to offload some of the workload. I'm not a fan of proprietary things but sheer rasterization power isn't enough at this point unless that's some pretty spectacular magic oil used for these new sampling techniques.