Intel XeSS should work on GPU AMD RDNA 2 and Nvidia Ampere, Turing and Pascal

Intel xess should work on gpu amd rdna 2 and nvidia ampere, turing and pascal

Intel XeSS is an enhanced technology of artificial intensification enhanced by intellectual intelligence that Intel hopes to be a pen on the cover of its Alchemist graphics cards when they finally face AMD and Nvidia. It’s a relatively well-understood proposition for gamers, unlike DLSS or FidelityFX Super Resolution, but we also know that Intel intends to offer this frame rate-boosting technology to its competitors ’GPUs.

Exactly which graphics cards competitors will benefit from the technology will be reduced to supporting DP4A instructions.

So let's get to the bottom of it. In Intel Alchemist GPUs, Intel XeSS will run on XMX acceleration, architecture-specific array engines that play a role similar to Nvidia's core cores. This deep learning acceleration will help XeSS improve your game performance.

However, for non-Intel Xe-HPG graphics cards, this acceleration can be done using DP4A instructions at a slightly higher expense (meaning it won’t be as powerful as XMX-accelerated XeSS). These DP4A instructions are used to multiply 8-bit integers (one byte, INT8) accumulated in a 32-bit integer and then run on a GPU's ALU. They are also used to speed up certain operations that do not require high precision, that is, deep learning.

Modern GPUs support DP4A (4-element signed full point product and buildup) in line with Microsoft's Shader model 6.4, while other older GPUs do not. This means that some GPUs can speed up certain instructions required by XeSS, while others cannot.

Expected XeSS compatibility
Architecture Compatible
Intel Xe-HPG Yes – XMX (confirmed)
Intel Xe-LP Yes – DP4A (confirmed)
Ancient Intel Architectures (Jan 11, Jan 9) No
AMD RDNA 2 Yes – DP4A (confirmed)
AMD Vega No
AMD Vega 7 nm Possible: DP4A
AMD Polaris No
Nvidia Ampere Yes, DP4A
Nvidia Turing Yes, DP4A
Nvidia Pascal Yes, DP4A
Older Nvidia architectures No

AMD has confirmed that RDNA 2 graphics cards support DP4A, in a comment on ComputerBase. This is achieved through Rapid Packed Math, a feature originally introduced with AMD’s Vega architecture, although AMD says only the RX-6000 series will support DP4A instructions.

You don't just have to take AMD's word for it; the RDNA 1 ISA Reference Guide (PDF Warning) does not include DP4A support, nor first-generation Vega. While RDNA Guide 2 notes "Operations with added ALU products accelerate inference and deep learning" as a change of function with the latest architecture.

It is worth mentioning that Vega of 7 nm, that is within Radeon VII, support list for DP4A operations, which could mean that it also supports XeSS, and owners of this card may feel justified in purchasing it instead of an RX 5700 XT.

Intel does not expect a significant loss of performance between XMX and DP4A acceleration. (Image credit: Intel)

AMD, however, points out that older GPUs could theoretically run XeSS with ALP FP32 / FP16, but that would be slower. We’ll probably never find out if that’s the case, as Intel has already largely removed a more general FP32 / FP16 implementation of XeSS in a recent interview with WCCFTech, citing possible performance questions.

I have contacted AMD for further clarification on DP4A support and will update this article if I see anything new.

Going forward, at Nvidia, and it looks like the company’s Ampere and Turing architectures will support DP4A. There is no doubt about it. In fact, the Pascal architecture should also support DP4A, actually as first architecture that supports instruction for the green team.

This is good news for 10 Series owners, as Nvidia's DLSS technology only supports 20 Series RTX cards or later.

In the aforementioned interview, Intel’s Karthik Vaidyanathan also states that he will not be looking to program specifically for Nvidia’s core cores, as it would require custom programming.

I've also contacted Nvidia to see if it will share more information about supported GPUs.

(embed) (/ embed)

As for the older Intel kit, it looks like the Xe-LP-based iGPUs found on 11th-generation mobile chips should support the instruction and therefore work with XeSS. I haven’t seen any evidence to show that any previous iGPUs or desktop iGPUs will be supported, so for me, the cut will be the latest Intel Xe-based chips.

However, it appears that a large number of older discrete graphics cards will be open to use by Intel's XeSS technology. This should be a big hit for Intel technology, but I'm confident it can deliver higher frame rates without too much loss of clarity, even with DP4A over XMX. Success will also be based on game support, and Nvidia has a several-year lead in that regard.

If all goes well for Intel, however, XeSS is sure to increase the pressure on AMD and Nvidia.

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *