I'll try rolling the driver back to 440 and see what happens.
Hardware and temps I doubt are an issue given the card works fine on other prime number projects but it won't hurt to try just the same.

Went to 455 drivers instead. Still no go.

Back to 435 drivers, still no go.

Still running PrimeGrid tasks with no drama (not at the same time of course)

ldd mfaktc.exe confirms I have all required libraries installed correctly.

Same issue on my machine. mfaktc was happily running (Manjaro Linux), then since I updated the nvidia drivers, it keeps throwing "ERROR: cudaGetLastError() returned 702: the launch timed out and was terminated".

I noticed the nvidia drivers were updated from 450.66 to 450.80 so I rolled back, but no luck...
The only way I can make it work is by disabling the "Interactive" option in X.org config as described in https://nvidia.custhelp.com/app/answers/detail/a_id/3029/~/using-cuda-and-x but it makes the computer pretty much unusable...

Also worth noting that older version of mfaktc I was using before outside of SRBase (mfaktc-0.21.linux64.cuda65) is still working fine, without changing X.org config.

Also worth noting, it looks like from Linux kernel 5.9, the nvidia-uvm module won' t load due to licensing issues (it wants to use other bits of GPL code I think). and this causes CUDA applications to fail... So stay with 5.8 until Nvidia fixes their drivers.

Not sure if this is related, but I converted a Windows 7 host to Linux Mint and it successfully completed more than 40 GPU tasks on an NVIDIA GTX 960 but then started to fail with computation errors.

The error message is:

./mfaktc.exe: error while loading shared libraries: libcudart.so.10.1: cannot open shared object file: No such file or directory

Host ID is 211122

Anybody seen this before and know if it's a simple problem to fix?

Rebooting didn't solve it.
____________

The FAQ isn't easy to understand :(

The GPU info is:

CUDA: NVIDIA GPU 0: GeForce GTX 960 (driver version 455.38, CUDA version 11.1, compute capability 5.2, 1993MB, 1952MB available, 2618 GFLOPS peak)


The FAQ says I will be sent Cuda111 and Cuda100 tasks.

How do I stop the Cuda100 tasks from causing computation errors?

The FAQ says to copy a missing file to /usr/lib64 -- but that file is deleted when you install Cuda11 Nvidia Drivers.

I tried this last night with a libcudart.so.10.1.243 copied from another host and renamed to 10.1, but it didn't work in /usr/lib64.

So I created the missing directories and copied it to /usr/local/cuda/lib64 (the location on the other host) and that didn't work either.

Now trying again with your file copied to both locations.

Still getting computation errors.

Do you know that when you install the latest Nvidia driver it removes the entire cuda directory from /usr/local ?

It's not just this file that is missing. Everything has gone.

The other cuda files are in /usr/lib (not lib64) and in subdirectories by host architecture.

OK, this now works!

The file needs to go in /usr/lib

Thank you for sending the file link.

A little confusing, but also rewarding to use with old Macs and PCs :)

My next job is to get boinc to recognise a Thunderbolt eGPU. Ubuntu has no problem utilising the eGPU, but boinc says "no GPU found" no matter if it is Nvidia or ATI :(


I have a question about your 3080.

Have you tried running multiple tasks at the same time?

I would have thought that two, or perhaps even 4 at a time would be more efficient because the time spent ending/beginning each tasks is significant when processing tasks so quickly?
____________

Didn't manage to get 3080 Ti running. Had to use driver 465 and re-added libcudart.so.10.1 to /usr/lib/ but got:
process exited with code 195 (0xc3, -61)</message>

rebirther: do we need additional changes to libcudart or to application to use cuda 11.3? file in order to work with 465 driver?
smi: Driver Version: 465.27 CUDA Version: 11.3

host: https://srbase.my-firewall.org/sr5/show_host_detail.php?hostid=211970
boinc does not detect card fully but running great on other projects.

https://srbase.my-firewall.org/sr5/show_host_detail.php?hostid=211970

installed just driver then ocl-icd-libopencl1 and nvidia-opencl-dev