Intel Iris Xe Max Support

[ejf_fold2021]

I am requesting support for latest Intel Tiger-Lake CPUs, XE iGPUs, Xe Discrete GPU

[Joe_H]

Welcome to the folding support forum.

The intro topic to this sub-forum provides directions on how to request a GPU be added - viewtopic.php?f=83&t=26208. However, as the Intel iGPUs are only a beta test platform, I suspect there is no hurry to add those. In addition, with the Tiger Lake processors Intel dropped hardware support for double precision (FP64) calculations in the iGPU. Those are necessary for folding.

As for the XE GPUs, those introduce a new instruction set and may require new support be added to the OpenMM codebase that is used for GPU folding cores. The early information I have seen so far also does not indicate which if any of the XE models support FP64.

[JimboPalmer]

Support for latest Intel Tiger-Lake CPUs, XE iGPUs, Xe Discrete GPU

Welcome to Folding@Home!

Are you asking if future products are supported? No, F@H will need a product to exist to know if it is supported, and if necessary write code to support them.

There is a beta group testing existing Intel igpus in the hope that future products will be similar. You can join the beta team if you are interested.

[ejf_fold2021]

Welcome to the folding support forum.

The intro topic to this sub-forum provides directions on how to request a GPU be added - viewtopic.php?f=83&t=26208. However, as the Intel iGPUs are only a beta test platform, I suspect there is no hurry to add those. In addition, with the Tiger Lake processors Intel dropped hardware support for double precision (FP64) calculations in the iGPU. Those are necessary for folding.

As for the XE GPUs, those introduce a new instruction set and may require new support be added to the OpenMM codebase that is used for GPU folding cores. The early information I have seen so far also does not indicate which if any of the XE models support FP64.

I believe FP64 is a supported configuration with the change of architectures and their new Graphics API, but I am not that knowledgeable about the nuances of the codebase.

[Joe_H]

No hardware FP64 available in the Tigerlake iGPUs, that information is direct from Intel. If Intel emulates FP64 through the driver that might be good enough, but that currently is not an option.

The new instruction set is probably going to be the bigger problem.

[ejf_fold2021]

No hardware FP64 available in the Tigerlake iGPUs, that information is direct from Intel. If Intel emulates FP64 through the driver that might be good enough, but that currently is not an option.

The new instruction set is probably going to be the bigger problem.

It's probably through emulation but I have a notebook with both the Xe-LP and Xe-DG1 (Xe Max) so it would be nice to utilize all the cores to fold.

I am willing to be the Guinea Pig for any betas on Windows.

[JimboPalmer]

Beta forum
viewtopic.php?f=16&t=8
viewforum.php?f=66
Intel Graphics forum
viewforum.php?f=111
How to get a GPU whitelisted
viewforum.php?f=83

[ejf_fold2021]

System 1:

CPU -> i5-1135G7
GPU -> Xe Max (DG1)
OS -> Win 10
Other -> Tiger-Lake U


I am requesting beta testing GPU support of Intel iGPUs and Xe-xx GPUs for compute.

[bruce]

Intel GPUs are not really supported, nor is beta team membership required. They have been intermittently beta tested but the results have not been promising so you'll probably not get any assignments unless you have an AMD or an NVidia GPU. (A number of early AMD GPUs were excluded on the same basis.) Hardware support of FP64 instruction are generally found in GPUs that are designed for Gaming and scientific computing. They are not used to support desktop video so I have my doubts that they will ever be supported on the DG1.

Until we can validate that Hardware FP64 is supported by Intel, we cannot add support. I

Othere intel GPUs (e.g.- UHD Graphics which has FP64 support can be tested for no points by setting ``gpu-beta=true`` in the configuration.) I'm not sure if that will help you.

[JimboPalmer]

https://www.techpowerup.com/gpu-specs/i ... hics.c3737

This seems positive, the known show stoppers are no 64 bit Floating Point math also called Double Precision, and OpenCL 1.2 support and it seems to have OpenCL 3.0 support. So if the drivers match what GROMACS expects, all is well.

https://en.wikipedia.org/wiki/GROMACS

[bruce]

Very encouraging:

"Intel Iris Xe MAX Graphics
FP64 (double) performance
633.6 GFLOPS (1:4)"

"The GPU is operating at a frequency of 300 MHz, which can be boosted up to 1650 MHz,"
I wonder how they plan to keep it cool.

@ ejf_fold2021
What hardware codes are shown by lspci or by GPU-Z?

[ejf_fold2021]

Very encouraging:

"Intel Iris Xe MAX Graphics
FP64 (double) performance
633.6 GFLOPS (1:4)"

"The GPU is operating at a frequency of 300 MHz, which can be boosted up to 1650 MHz,"
I wonder how they plan to keep it cool.

@ ejf_fold2021
What hardware codes are shown by lspci or by GPU-Z?

FInd attached from GPU-z
https://1drv.ms/u/s!AmOlOpxInkhRsAvoomK ... 7?e=EBt1ek

Supposedly the power budget for the chips are balanced and spread through the cpu/iGPU/dGPU. The Xe Max has its own 4GB LPDDRx memory.