Unrealistic WU for AMD iGPU

Moderators: Site Moderators, FAHC Science Team

Post Reply
UdoA
Posts: 12
Joined: Sun Feb 07, 2021 10:26 am

Unrealistic WU for AMD iGPU

Post by UdoA »

Hello,

I have been folding for some time now and also bought a little PC with the AMD Ryzen 7 Pro 4750G CPU which now works quite well.
The folding on this CPU for the CPU part works quite well, but the folding on the integrated GPU, which up to now was o.k. now gets an issue.
For some about a week or so, I get WU which are way beyond the capability of the GPU. The credit is/was at most 120000 being mostly in the area of 109000 points.
Now I get WU which are worth about 220000 credit but take about a day more than the deadline permits.

Anyone know what is going on there and how to get around this?
It is kind of annoying to repeatedly kill the WU which is also not what is intended.

Thanks in advance for your suggestions.

Edit: The PC works exclusively for folding@home on a 24/7 basis under Linux in minimal configuration to allow maximum utilization for the folding.

Greetings from Cologne, Germany, Udo
JimboPalmer
Posts: 2573
Joined: Mon Feb 16, 2009 4:12 am
Location: Greenwood MS USA

Re: Unrealistic WU for AMD iGPU

Post by JimboPalmer »

Currently, GPUs are ranked by capability, not performance. Because the 4750G is recent, it has modern features and is lumped in with other modern AMD GPUs.

F@H has tried to move to a performance metric. They are having issues with it but want to move to WUs the GPU can complete.
Tsar of all the Rushers
I tried to remain childlike, all I achieved was childish.
A friend to those who want no friends
UdoA
Posts: 12
Joined: Sun Feb 07, 2021 10:26 am

Re: Unrealistic WU for AMD iGPU

Post by UdoA »

Thanks for the answer.
Looks like no way to wrok around this except for killing overwhelming WUs.

BTW: There have been no beta-packages of the software for some time now. Don't they develop it any further at the moment?
JimboPalmer
Posts: 2573
Joined: Mon Feb 16, 2009 4:12 am
Location: Greenwood MS USA

Re: Unrealistic WU for AMD iGPU

Post by JimboPalmer »

There is one developer, who writes the client software, the server software, and the website. As I understand it, he is currently rewriting server software to match features already implemented in the client. Assigning WUs by performance would be part of that. It won't need more info from your PC, just new server software and a new GPUs.txt file.

The Cores, the science software, is written by the researchers. They seem to be focused on Intel GPUs.
Last edited by JimboPalmer on Sat Jul 24, 2021 3:03 pm, edited 1 time in total.
Tsar of all the Rushers
I tried to remain childlike, all I achieved was childish.
A friend to those who want no friends
Jonazz
Posts: 357
Joined: Sun Jan 17, 2010 2:08 pm

Re: Unrealistic WU for AMD iGPU

Post by Jonazz »

JimboPalmer wrote: The Cores, the science software, is written by the researchers. They seem to focused on Intel GPUs.
I don't think that's correct :D
JimboPalmer
Posts: 2573
Joined: Mon Feb 16, 2009 4:12 am
Location: Greenwood MS USA

Re: Unrealistic WU for AMD iGPU

Post by JimboPalmer »

Jonazz wrote:
JimboPalmer wrote: The Cores, the science software, is written by the researchers. They seem to be focused on Intel GPUs.
I don't think that's correct :D
Core_22 supported AMD Navi, that is last major change.

I haven't heard no rumors that F@H is working to support tensor cores or Ray Tracing cores on Nvidia, just the long established CUDA cores.

Intel GPU betas have been ongoing.
Last edited by JimboPalmer on Sat Jul 24, 2021 8:58 pm, edited 1 time in total.
Tsar of all the Rushers
I tried to remain childlike, all I achieved was childish.
A friend to those who want no friends
Joe_H
Site Admin
Posts: 7856
Joined: Tue Apr 21, 2009 4:41 pm
Hardware configuration: Mac Pro 2.8 quad 12 GB smp4
MacBook Pro 2.9 i7 8 GB smp2
Location: W. MA

Re: Unrealistic WU for AMD iGPU

Post by Joe_H »

JimboPalmer wrote:There is one developer, who writes the client software, the server software, and the website. As I understand it, he is currently rewriting server software to match features already implemented in the client. Assigning WUs by performance would be part of that. It won't need more info from your PC, just new server software and a new GPUs.txt file.

The Cores, the science software, is written by the researchers. They seem to be focused on Intel GPUs.
F@h did put out a job posting for another developer position few months ago. I don't know what is going on with that.

In broad strokes though this is how current development is being handled. The one full time developer has been tasked with working on the server code to improve it and add the features needed to allow more granular assignment of WUs to better match with GPU and CPU capabilities both feature and speed wise.

The GPU core software is being worked on by members of the research groups. Most recent changes added support for AMD Navi and Intel iGPUs. The Intel support remains beta as they work on some bugs in the software support and work out how to incorporate them into WU processing. This also involves updates being done to the OpenMM code that the F@h GPU core is based on.

The CPU core was updated with outside developer help to create the A8 folding core. This updates CPU folding to use a newer version of GROMACS code, and added support for ARM chips. The code to fully support Big-Little by adjusting how much computing was assigned to each thread based on the power of the core it was to run on did not work as expected. So the current A8 release does not have that enabled.

The previous A7 folding core creation was given to the full time developer to d a few years ago.
Image

iMac 2.8 i7 12 GB smp8, Mac Pro 2.8 quad 12 GB smp6
MacBook Pro 2.9 i7 8 GB smp3
gunnarre
Posts: 567
Joined: Sun May 24, 2020 7:23 pm
Location: Norway

Re: Unrealistic WU for AMD iGPU

Post by gunnarre »

The main reason why Intel GPU support is a good idea is that Intel is coming out with more capable discrete GPUs in the near-ish future, if things work out as planned.
Image
Online: GTX 1660 Super, GTX 1080, GTX 1050 Ti 4G OC, RX580 + occasional CPU folding in the cold.
Offline: Radeon HD 7770, GTX 960, GTX 950
Post Reply