Unable to run GPU workunits

[mfpotter]

For the past several months, I have been unable to run any workunits for my GPU (I AM able to run CPU workunits). Prior to this, I did not h ave any issues. I am currently running the latest Windows client (7.6.13). My OS version is Windows 10 Pro, version 1909. My computer is a Microsoft SurfaceBook with the 'Performance Base' (Intel i7-6600U CPU/NVIDIA GeForce GTX 965M GPU/16GB RAM). Every time a GPU workunit is downloaded, entries such as shown below are generated in the Log window of the client:

10:49:53:WU02:FS01:0x22:ERROR:exception: Error initializing context: clGetDeviceInfo (-5)
10:49:53:WU02:FS01:0x22:Saving result file ..\logfile_01.txt
10:49:53:WU02:FS01:0x22:Saving result file science.log
10:49:53:WU02:FS01:0x22:Folding@home Core Shutdown: BAD_WORK_UNIT
10:49:54:WARNING:WU02:FS01:FahCore returned: BAD_WORK_UNIT (114 = 0x72)

The client will then attempt to download a new workunit, and everything repeats as above, until, after some number of repeated attempts, the client has 'had enough', and the slot is listed as "FAILED".

As I wrote, I used to be able to run GPU workunits, and nothing has changed wrt my hardware. I appreciate any assistance.

Best regards,
Mike

[ajm]

It can be a mere driver issue: it can happen that windows updates the drivers, using more "universal" versions of them that FAH cannot really use. In this case, reinstalling the drivers downloaded from Nvidia would solve the issue.

But it also could be a problem linked to the fact that your Surface Book has two GPUs. The i7-6600U has an Intel HD 520 embedded (iGPU) and of course you have the GTX 965M. Depending on how you configure your device, FAH may be led to try and use the Intel iGPU and that just cannot work. I don't have a straightforward solution as I have no experience with these devices, but I would try and disable the iGPU, then delete the GPU slot in Advanced Control, quit and restart Advanced Control, and recreate the GPU slot.

[mfpotter]

ajm,

Thank you for your reply.

The version of the NVIDIA display adapter currently installed is the same as it was in December 2019, when I was still able to run GPU workunits, so I do not believe that a driver update has "broken" the ability for my client to run GPU workunits.

Also, I have never disabled the integrated GPU on the i7 chip, and as I wrote, I was able, as of a few months ago, to run GPU workunits, so it doesn't seem to me that there is a conflict between which GPU (integrated or dedicated) that FAH is trying to access.

Best regards,
Mike

[ajm]

You indeed didn't have to disable the iGPU when you installed FAH, of course, and it would only be temporary here, for the duration of the process I was exposing above. Hence you may also try to uninstall FAH (incl. data) and reinstall it. If the problem resides there, as the error "clGetDeviceInfo (-5)" suggests, it would also be solved that way. I was just trying to propose a way to solve it without uninstalling, which would perturb the CPU folding.

But maybe someone else will propose a better solution. It might help that you post your log, then, as described here: viewtopic.php?p=327412&f=24#p327412

[anandhanju]

You may have luck "locking" your FAH GPU slot to the GTX 965M using the GPU Index in the Advanced Control.

[ajm]

To reduce the "luck" factor, Mike could peruse the old config.xml stored in %AppData%\FAHClient\configs\ (to get there, just copy this in the address line of an Explorer window).

Explanations: the file config.xml contains the personal configuration of the donor, including that of the different slots. It is stored in %AppData%\FAHClient\
Each time that configuration is changed (or that one quits/restarts FAH), FAH puts the old one in %AppData%\FAHClient\configs\ and renames it config-<date-code> (eg config-20200629-144759.xml), and creates a new file config.xml in %AppData%\FAHClient\
A total of 16 such old config files are kept.
So there probably is one or several config-<date-code>.xml file in %AppData%\FAHClient\configs\ that would indicate how the GPU slot ought to be configured in order to work correctly. By comparing the present configuration of the slot with previous ones that we know were working (see "Date modified" of the files), we could deduce how to adapt the GPU index.

Not exactly simpler but somehow elegant.

[Joe_H]

Two possibilities exist. One is as mentioned, the iGPU being active instead of your GTX645m. The other may be a side effect of internal testing to add support for using the the iGPUs present on Intel processors. Post the beginning 100-200 lines of you log file and which it is can be determined.

For the second, a numer of Intel iGPUs have been added to the GPUs.txt file and some have been enabled for the internal testers. Unfortunately, due to an issue with how the client handles updates to the supported list, your system may now have the folding slot connected to the Intel GPU instead of the nVidia. The fix is to take that slot, Pause it, and add the 'pause-on-start' parameter to the slot and set that to 'true'. Then add a slot for the GTX 645m

One other possibility exists, a number of the early nVidia GPUs based on the Fermi chips have been disabled. They do not supprt OpenCL 1.2, and there are other issues involved in supporting them.

[bruce]

The third possiblity is that your BIOS is deciding to select the Intel iGP when it believes power saving is more important than GPU performance. I don't have one of those dual GPU laptops, but I understand that there is a BIOS setting that selects the Perfornamce GPU. Is that the same as disabling the iGP?