you need to add support for every cc level explicitly.

the whole section would be something like this:

--generate-code arch=compute_50,code=sm_50 --generate-code arch=compute_52,code=sm_52 --generate-code arch=compute_60,code=sm_60 --generate-code arch=compute_62,code=sm_62 --generate-code arch=compute_70,code=sm_70 --generate-code arch=compute_75,code=sm_75 --generate-code arch=compute_72,code=sm_72 --generate-code arch=compute_80,code=sm_80 --generate-code arch=compute_86,code=sm_86 --generate-code arch=compute_89,code=sm_89

plus whatever other flags you have there

was v31 compiled with explicit support for CC_6.0 in the NVCCFLAGS section of your makefile or build script?


the argument should be something similar to this:

--generate-code arch=compute_60,code=sm_60

this was what I was suggesting when I mentioned recompiling it to add 6.0 support.

That fixed one of my PCs.

Another still errors out on every task
https://srbase.my-firewall.org/sr5/result.php?resultid=145885980

All errors with new version.
https://srbase.my-firewall.org/sr5/result.php?resultid=145853091

Please undo

Fine with 0.28, errors on 0.29.

ok, this was not as planned, reverted back to 0.28, need to check...

All errors with new version.
https://srbase.my-firewall.org/sr5/result.php?resultid=145853091

Please undo

Fine with 0.28, errors on 0.29.

there also appears to be a problem with the cuda100 package.

the mfaktc-linux64-v10.zip contains Windows files (mfaktc.exe and .ini files). it should contain linux binaries and the cuda10 libcudart.so.10 shared object libraries like the cuda11 package does.

the libcudart.so.10 file you can find in FAQs. If the app was dynamically compiled you dont need them in package. The .exe in linux was also compiled so but running under linux. The .ini file is fixed.

when I run ldd mfaktc.exe on the binary it says libcudart.so.10.1 is missing.

in your cuda111 and cuda120 packages you include this file.

I see nothing in the FAQ that references this file. can you link to it? Is your intended workaround for this to create a symlink from libcudart.so.10->libcudart.so.10.1 ?

there also appears to be a problem with the cuda100 package.

the mfaktc-linux64-v10.zip contains Windows files (mfaktc.exe and .ini files). it should contain linux binaries and the cuda10 libcudart.so.10 shared object libraries like the cuda11 package does.

cuda 12+ still supports CC 6.0 devices. why not just recompile the cuda120 app with CC 6.0 support?

otherwise, the only real solution is to transition to Anonymous Platform, and force the cuda100 app to run on everything with an app_info.xml file.

cuda12 app support cc6.1, see 1070ti I dont know why the Tesla doesnt run cuda120 WUs, anonymous patform is not allowed.

i know your app supports 6.1. but CUDA 12 from nvidia still supports down to CC 5.0

I'm saying you should recompile it to add support for 6.0 also. Tesla P100 has CC level 6.0.

6.1 is used on the mainstream pascal GPUs (GTX 10-series, Quadro P-series)

cuda 12+ still supports CC 6.0 devices. why not just recompile the cuda120 app with CC 6.0 support?

otherwise, the only real solution is to transition to Anonymous Platform, and force the cuda100 app to run on everything with an app_info.xml file.

I'm trying to say that the situation is worse now for some reason. With your help, I was able to process tasks for a time (on P100 only computers) using the first app_config you posted, but now that one no longer helps, nor does the second app_config with plan_class help either. I can't process tasks on P100s at all now, whether they are alone in the computer, in pairs, or mixed with another GPU. Nothing works now.

My humble request would be for you to revert whatever changes were made to the server side, so I can see if at least the P100-only computers can process tasks again. That would be a large improvement.

Also my P100-only machines are having problems again despite having got them working with the initial app_config you posted.

It could be that since the P100 is the second GPU, it's presence isn't being properly detected on the server side. That's just a wild guess based on the fact that the list of hosts always claims the computer has x number of whatever the first GPU is, instead of what's really there.

I'm sorry, after getting rid of the app_config and resetting the project, the client is still getting cuda120 tasks which error out immediately on the P100.
Perhaps it's not worth the trouble, not many people have these old compute GPUs.