Problem detecting GPUS

Hi, I just tried installing cryosparc2 and I got this error. I am using Ubuntu 18.04 and cuda 10. Should I try a lower cuda version?


Autodetecting available GPUs…
Traceback (most recent call last):
File “bin/”, line 191, in
gpu_devidxs = check_gpus()
File “bin/”, line 73, in check_gpus
num_devs = print_gpu_list()
File “bin/”, line 7, in print_gpu_list
pycuda._driver.Error: cuInit failed: unknown error

First i’d check the NVIDIA Driver to make sure it’s seeing the devices.

What’s the output of nvidia-smi -q -a ?

All the devices appear.

How does the look (cryosparc2_worker/ ? is it pointing to CUDA path properly?

how does the path look , is CUDA in the pathing properly?

you can do env | grep cuda to check where cuda shows up in the environmental variables.

Assuming it was a problem with the cuda path in my setup, can I just run the install again? Or should I delete the current installed files first?

If you checked and CUDA isn’t in the path, Reinstalling CUDA won’t fix path not being added.

if it’s installed at /usr/local/cuda (smlink there, check via ls) then add export PATH=$PATH:/usr/local/cuda to add it.

Sorry, I meant once I fix the path issue, should I re-run the whole cryosparc2 install and should I delete any files first.

I’d try doing the reinstall script again so it re-writes/refreshes - if the PATH was missing after it was re-added.

When you do the CUDA Installation you’re supposed to set the PATH / LD_LIBRARY_PATH environmental variables (it’s listed as an instruction after CUDA install finishes – on the terminal screen)

may want to set LB_LIBRARY_PATH also (not sure if Cryosparc uses that variable, but it might)

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH=/usr/local/cuda/lib for example.


If you’d like to reinstall cryoSPARC, you can delete cryosparc2_master and cryosparc2_worker folders, and start again (after re-extracting the tar archives).