amdgpu pro card no longer folding

[muziqaz]

OK let's start from the beginning and basics, since I see you are willing to install fresh OS.
Make sure you only have that GPU present on the system (if on Intel CPU, disable its iGPU, please). Install fresh ubuntu/kubuntu (go for 24.04). Do not go to AMD website for anything. Follow this guide to the letter: viewtopic.php?t=43222 except instead of https://repo.radeon.com/rocm/apt/6.4.3, add https://repo.radeon.com/rocm/apt/latest for repository. Check out this comment incase you hit some gpg key issue: viewtopic.php?p=371796#p371796
After successful repo addition, sudo apt update, and see if there is no errors
then:
sudo apt install rocm-opencl-runtime && sudo apt install rocm-hip-runtime (select 7.2, which will ensure you are pulling packages from radeon repository and not from ubuntu own)
after that
sudo apt install ocl-icd-opencl-dev (this will come from ubuntu repository)
Then download and install latest fahclient from foldingathome.org/beta. Download deb package, then cd to the Download folder in terminal and:
sudo apt install ./fah-client-8.5.6-somethingsomething.deb
Then before folding, go to fah system info, and look if you see True next to Supported, and then if you see OpenCL as supported, and HIP as supported. And GPU description should be gfx1010-xnacks or something like that.

Then you can try folding, keep an eye on which project and which core you are trying to fold. If you receive fahcore23 WU, pause folding after it crashes, and apply the remedy from this thread:
viewtopic.php?t=42789 <-- only for core23. Core27 is fine without it.

Now keep in mind, that every time you restart the PC, you will have to run this command:
sudo systemctl restart fah-client
To enable fahclient to see the GPU.

If things are not folding after this at all. Either your GPU is broken, or workstation cards require some special sauce to work :/

[nada391]

thanks ,
might aswell try this approach one more time, curious to find out.

it still does however not explain wy it wud not function not even on windows 11 25h2 or win 10 22h2 (tried both)
there are some other error involved, but failing on both (tried only on one computer9)> on windows 10 tried even without installing any driver from AMD, just the windows driver that was already present and recognized the card correctly.

most probably the system is instable altogether.

[muziqaz]

Ah, if it is also failing on Windows, then yes, system has issues

[nada391]

OK let's start
sudo apt install rocm-opencl-runtime && sudo apt install rocm-hip-runtime (select 7.2, which will ensure you are pulling packages from radeon repository and not from ubuntu own)

thios would lead to error on latest ubuntu 24.04 lts noble

The following information may help to resolve the situation:

The following packages have unmet dependencies:
rocm-hip-runtime : Depends: rocminfo (= 1.0.0.70200-43~24.04) but 5.7.1-3build1 is to be installed
E: Unable to correct problems, you have held broken packages.

https://ibb.co/HLKKmy2M

[muziqaz]

OK let's start
sudo apt install rocm-opencl-runtime && sudo apt install rocm-hip-runtime (select 7.2, which will ensure you are pulling packages from radeon repository and not from ubuntu own)

thios would lead to error on latest ubuntu 24.04 lts noble

The following information may help to resolve the situation:

The following packages have unmet dependencies:
rocm-hip-runtime : Depends: rocminfo (= 1.0.0.70200-43~24.04) but 5.7.1-3build1 is to be installed
E: Unable to correct problems, you have held broken packages.

https://ibb.co/HLKKmy2M

I cannot remember how exactly, but there is a way to disable/lock out everything related to rocm from ubuntu's own repository. I remember I had to do this in Fedora, as fedora had some outdated packages in its own repo, while radeon repo had latest packages. It was clashing with each other until I instructed apt to not touch rocm stuff on fedora repo

[nada391]

thank you,

it seems a bit difficult to acheive;

will wait for the release of ubuntu 26.04 and then try again to reinstall; maybe the driver situation is bettered then, as it has been suggested that for newer cards the driver will auto-install by then (tried yesterday a testing.iso of 26.04 and it did so for a nvidia-card t400).

am happy to report the card can still fucntion,
since it folds on windows 2022 server - am wondering however if the numbers are correct, since am getting sometimes quite higher number than on linux, even thrice higher number (maybe i am cheating?=)

like here shown
924,965 304,456 (linux) 937,519 (windows) Navi 24

taken from here
https://folding.lar.systems/gpu_ppd/overall_ranks

[muziqaz]

In official Ubuntu repository source file following line can exclude rocm stuff:
exclude=rocm*

[nada391]

back to square1 after one-two week of folding succesfully,
yesterday or day before installed once v7 instead of v8 (for testing purpose) -
and received windows update in the process.

now the card would neither fold on v7 nor v8 - with all sorts of error.
(for windows server 2022,2025)

Code: Select all

17:30:33:I1:WU5:Attempting to create OpenCL context:
17:30:33:I1:WU5: Configuring platform OpenCL
17:31:16:I1:WU5:ERROR:Discrepancy: Forces are blowing up! 0 0 

Code: Select all

7:25:49:I1:WU2:UPLOAD 94% 109.12MiB of 115.98MiB
17:25:52:I1:WU2:UPLOAD 95% 110.48MiB of 115.98MiB
17:25:54:I1:WU4:An exception occurred at step 248: Particle coordinate is NaN. For more information, see https://github.com/openmm/openmm/wiki/Frequently-Asked-Questions#nan
17:25:54:I1:WU4:Max number of attempts to resume from last checkpoint (2) reached. Aborting.
17:25:54:I1:WU4:ERROR:114: Max number of attempts to resume from last checkpoint reached.
17:25:54:I1:WU4:Saving result file ..\logfile_01.txt
17:25:54:I1:WU4:Saving result file science.log
17:25:54:I1:WU4:Saving result file state.xml.bz2
17:25:54:I1:WU4:Saving result file xtcAtoms.csv.bz2
17:25:54:I1:WU4:Folding@home Core Shutdown: BAD_WORK_UNIT
17:25:54:E :WU4:Core returned BAD_WORK_UNIT (114) 

Code: Select all

7:25:03:I1:WU4:Completed 0 out of 5000000 steps (0%)
17:25:03:I1:WU4:An exception occurred at step 0: Potential energy error of inf, threshold of 20000
17:25:03:I1:WU4:Reference Potential Energy: -564085 | Given Potential Energy: inf
17:25:03:I1:WU4:ERROR:98: Attempting to restart from last good checkpoint by restarting core.
17:25:03:I1:WU4:Folding@home Core Shutdown: CORE_RESTART
17:25:04:I1:WU4:Core returned CORE_RESTART (98) 

Code: Select all

17:33:04:I1:WU7:ERROR:Discrepancy: Forces are blowing up! 52667 0
17:33:04:I1:WU7:Saving result file ..\logfile_01.txt
17:33:04:I1:WU7:Saving result file science.log
17:33:04:I1:WU7:Saving result file state.xml.bz2
17:33:04:I1:WU7:Folding@home Core Shutdown: BAD_WORK_UNIT
17:33:05:E :WU7:Core returned BAD_WORK_UNIT (114) 

the computer has the newest Radeon Pro software released for this card already installed.

[muziqaz]

Windows update overwrote your AMD drivers.
And for the love of protein, stop randomly installing other clients on the troublesome system for "testing purposes".
If it ain't broken, DO NOT fix it