Radeon 7800 XT not used for folding (Fedora 41)

[Bud Specter]

Hi,

I compiled and installed the FAH client from the latest code based on the RPM build instructions: https://github.com/FoldingAtHome/fah-cl ... ING-RPM.md

Folding with CPU cores works but I want to use my dGPU, Radeon 7800 XT, instead. It seems to be recognized but when I start folding it is not used at all. Is there a way to debug this or is there info I could provide? In "machine details" I see that my dGPU is listed as "not supported" which I find weird because it worked with FAH client v7.

Below is my machine log with my discrete GPU enabled and CPU disabled. It says "machine state fold" but it is not downloading and processing any work units.

Code: Select all

15:23:40:I1: Version: 8.4.10
15:23:40:I1: Author: Joseph Coffland <joseph@cauldrondevelopment.com>
15:23:40:I1: Org: foldingathome.org
15:23:40:I1: Copyright: 2023-2024, foldingathome.org
15:23:40:I1: Homepage: https://foldingathome.org/
15:23:40:I1: License: GPL-3.0-or-later
15:23:40:I1: URL: https://v8-4.foldingathome.org/
15:23:40:I1: Date: Jan 29 2025
15:23:40:I1: Time: 16:17:06
15:23:40:I1: Revision: cc3168a54beafdd0a502d8c599256ef3d1485bfc
15:23:40:I1: Branch: master
15:23:40:I1: Compiler: GNU 14.2.1 20250110 (Red Hat 14.2.1-7)
15:23:40:I1: Options: -Wsuggest-override -faligned-new -std=c++14 -fsigned-char -flto -O3
15:23:40:I1: Platform: linux 6.12.10-200.fc41.x86_64
15:23:40:I1: Bits: 64
15:23:40:I1: Mode: Release
15:23:40:I1: Args: --config=/etc/fah-client/config.xml
15:23:40:I1: --log=/var/log/fah-client/log.txt
15:23:40:I1: --log-rotate-dir=/var/log/fah-client/
15:23:40:I1: Config: /etc/fah-client/config.xml
15:23:40:I1:****************************** CBang ******************************
15:23:40:I1: Version: 1.7.2
15:23:40:I1: Author: Joseph Coffland <joseph@cauldrondevelopment.com>
15:23:40:I1: Org: Cauldron Development
15:23:40:I1: Copyright: Cauldron Development, 2003-2024
15:23:40:I1: Homepage: https://cauldrondevelopment.com/
15:23:40:I1: License: LGPL-2.1-or-later
15:23:40:I1: Date: Jan 29 2025
15:23:40:I1: Time: 16:16:34
15:23:40:I1: Revision: 70a6b569a63b2d3e354b8a732fff81d31505232e
15:23:40:I1: Branch: master
15:23:40:I1: Compiler: GNU 14.2.1 20250110 (Red Hat 14.2.1-7)
15:23:40:I1: Options: -Wsuggest-override -faligned-new -std=c++14 -fsigned-char -flto -O3
15:23:40:I1: Platform: linux 6.12.10-200.fc41.x86_64
15:23:40:I1: Bits: 64
15:23:40:I1: Mode: Release
15:23:40:I1:***************************** System ******************************
15:23:40:I1: CPU: AMD Ryzen 5 7600 6-Core Processor
15:23:40:I1: CPU ID: AuthenticAMD Family 25 Model 97 Stepping 2
15:23:40:I1: CPUs: 12
15:23:40:I1: Memory: 30.43GiB
15:23:40:I1:Free Memory: 28.95GiB
15:23:40:I1: OS Version: 6.12
15:23:40:I1:Has Battery: false
15:23:40:I1: On Battery: false
15:23:40:I1: Hostname: rhea
15:23:40:I1: UTC Offset: 1
15:23:40:I1: PID: 1289
15:23:40:I1: CWD: /var/lib/fah-client
15:23:40:I1: Exec: /usr/bin/fah-client
15:23:40:I1:*******************************************************************
15:23:40:I2:<config/>
15:23:40:I1:Opening Database
15:23:40:I1:F@H ID = Cd9cfCvu8nzviD6IfFPNzJvEU5bMOXIgKYjyuMjnhKU
15:23:40:I3:Loading default resource group
15:23:40:I1:Listening for HTTP on 127.0.0.1:7396
15:23:40:I3:Loaded 0 wus.
15:23:40:I3:gpus = {
15:23:40:I3: "gpu:03:00:00": {"vendor": 4098, "device": 29822, "type": "amd", "supported": false, "description": "Navi 32 [Radeon RX 7700XT/7800XT]"},
15:23:40:I3: "gpu:122:00:00": {"vendor": 4098, "device": 5710, "type": "amd", "supported": false, "description": "Raphael [Ryzen 7000 Series iGPU]"}
15:23:40:I3:}
15:24:18:I1:Machine state fold

[muziqaz]

Rocm-opencl-sdk is needed and then usermod render and video for your username. Restart the PC and you should be good

[muziqaz]

Hm, probably I'll create a new thread with the guide for all major distros when I'm home

[Bud Specter]

Thanks for the quick reply!

I added my user to the groups "render" and "video". Additionally, I installed the packages "rocminfo" and "rocm-opencl" via DNF. I could not find a Fedora package for rocm-opencl-sdk. (I also found some Fedora docs: https://fedoraproject.org/wiki/SIGs/HC)

I rebooted my machine. Both rocminfo and clinfo commands list my 7800 XT. I think that's good! However, FAH client still does not use my card for folding.

Edit: I found an AMD document stating that the integrated GPU, if one exists, should be disabled: https://rocm.docs.amd.com/projects/inst ... applicable
I have to do that as well I guess.

[muziqaz]

Remove that rocm-opencl.
Let me get home, and I'll post you the repo name for you to grab actual working stuff

[muziqaz]

It seems I am travelling around London the whole night today...
Here is a guide. This is copy pasta from another forum comment I wrote, since I'm on the bus:

remove rocm-opencl.
add https://repo.radeon.com/amdgpu-install/6.2.3/ to your repo list (if you know how to do it).
or
go to https://repo.radeon.com/amdgpu-install/6.2.3/ choose your distro (probably el or rhel), click on OS version (I think 41 is based on RHEL9.4, you might figure that out, or know yourself).
Download amdgpu-install package from there. Install it, it will add above mentioned repo to your sources.
refresh them. and then install rocm-opencl-sdk
then
sudo usermod -a -G render yourlinuxusername
sudo usermod -a -G video yourlinuxusername
reboot the pc.
check clinfo output. it should show opencl platform as AMD platform and opencl device should be your GPU.

If after reboot fahclient still cannot see your gpu as supported, get rocm icd loader. That should fix it.
Once things are working, go to your software sources and remove amdgpu-install related repo, since that is known to throw a fit or two is kept as repo. It is not being updated anyways, so it was used to get ROCm repo to your system.
Also if you see ROCm 6.3.x available, do not use it, this one is for HPC market, AMD seems to think it is ok to mix HPC ROCm with consumer one...

[Bud Specter]

Hey!

The user groups are already configured. I removed the rocm-opencl package, added the amdgpu-install repo and installed rocm-opencl-sdk.
Rebooted. Clinfo output was OK. Still FAH webclient listed the dGPU as unsupported and work units are not processed.

I installed OpenCL-ICD-Loader but it reported a package conflicht with ocl-icd. Now, clinfo output does not report any platform!

Thanks for your efforts, however, I'll uninstall ROCm packages completely and try to get back to a clean state and will ask in the Fedora forums for additional experience with this problem. I'll report back if I find anything.

[muziqaz]

Get the sdk back on, and copy and paste clinfo output here

[Marcos FRM]

... usermod render and video for your username ...

Is it necessary? The client process is executed under the fah-client user account, which is already part of the render and video groups.

[muziqaz]

... usermod render and video for your username ...

Is it necessary? The client process is executed under the fah-client user account, which is already part of the render and video groups.

That never worked
sudo usermod -a -G video linuxusername
sudo usermod -a -G render linuxusername
commands were always a requirement for AMD to be able to fold. Always