Intel XeSS does not need per-game training

WCCF Tech had an interview with Intel Principal Engineer Karthik Vaidyanathan about XeSS, the AI-based supersampling technology that will be used by the ARC Alchemist GPUs. This technique will attempt to rival NVIDIA's DLSS and the interview definitely covers a lot of ground.

No per-game training needed

Vaidyanathan explains Intel's objective from day one was to ensure XeSS did not need training on a per game basis. He also confirms that the XeSS demo (video embedded below) was not used as part of the training process of XeSS. Vaidyanathan claims XeSS can be rolled out to a lot of games quickly because it's very generalized. He also added that XeSS will have multiple quality models, allowing gamers to select between more performance or higher image quality.

You mentioned Unreal Engine 5 and it produces some of the highest quality shadow geometric lighting fidelity and when you're investing so much in your render, you really don't want to lose any of that quality when you scale from that - those pixels to your target resolution - and that's been our objective from day one. And also you don't want to have a solution that's fragile that requires training for every game that someone ships that's also been our objective from day one. -- Intel's Karthik Vaidyanathan

XeSS can work on older and non-Intel GPUs

There are two versions of Intel's XeSS; one version uses DP4a and the other version uses Xe Matrix eXtensions (XMX). The latter is the fastest and is Intel-specific. But DP4a is supported by NVIDIA and AMD GPUs, so it could potentially run on supporting cards. Intel's integrated graphics support it since the Skylake generation, so backwards compatibility may arrive too.

Karthik: Yes, we require inference capabilities but matrix acceleration is not the only form of inference capability that is available on GPUs. If you go all the way back to- I think Skylake- we had dot product acceleration, which is DP 4.4 – there’s various names for it. Nvidia has had this I think, since Turing and AMD has this now on RDNA2. So even without Matrix acceleration you can go quite far. It might not be as fast as matrix acceleration, but certainly meets the objective. And as I said, the objective is to maintain the fidelity of your render and achieve smooth frame rates. So, when it comes to older models, on older internal GPUs, we've had dot product acceleration (DP4a) for a while now. Microsoft has enabled this through Shader Model 6.4 and above and on all these platforms XeSS will work. -- Intel's Karthik Vaidyanathan