Pycuda.driver in Ubuntu 18.04 - CryoSparc 2.11

Hi -

I’m having trouble after upgrading from Cryosparc2.10 to 2.11.
I have a standalone system - i.e. the master is also the worker. The master installs fine, the worker installation gets stuck with “No module named pycuda.driver”.

OS is Ubuntu 18.04 LTS.
I’ve tried various Nvidia drivers, and also Cuda 9.2 and 10.
Tried installing pycuda with pip, and also with the conda under the worker install directory.
Nothing has helped so far.

I see that there was a similar problem a year ago, by jkang and Vinson. probably with an older version of Crypsparc. I didn’t see any answers there.

Questions:

  1. Has anyone managed to get Cryosparc v.2.11 working standalone under Ubuntu 18? Which versions of Nvidia driver and Cuda are you using? Are there any other crucial packages I might have not installed?

  2. I think that Cryosparc v.2.10 worked. It it possible to get the 2.10 packages - master and worker?

Thanks for any assistance.

Gershon,

We have I think a related problem. Upgrade from Suse 13.2 to Leap 15.1 and upgrade of Cryosparc 2.9 to 2.11. Initially we had issues with the pip install of pycuda, after several iterations we had issues with /usr/lib64/crti.o “unrecognized format”, it just would not build, looking through we ended up removing the softlink /usr/local/cuda which was pointing to cuda-10.1, and after that we were able to run the cryosparc_worker connect command successfully, where we had previously been unable to load/find(?) the driver.py file.

At this point the jobs in the queue are waiting on GPU, and I’m unclear as to why, or where I might find daemon log files what would help me unravel the issue.

If you have any advice on pycuda/driver.py build or where to find log files.
I hope that our experience is of some use to you, but we do not have a running v2.11.

thank you,
Brian

Hi @Gershon,

Has anyone managed to get Cryosparc v.2.11 working standalone under Ubuntu 18? Which versions of Nvidia driver and Cuda are you using? Are there any other crucial packages I might have not installed?

We have installed cryoSPARC on a machine with similar specs (CUDA 10.1) and found no issues. All dependencies are installed with cryoSPARC except CUDA, GLIBC >= 2.12 and gcc >= 4.4 (Note: On a fresh Linux installation, you may need to run sudo apt-get install build-essential) and
bash as the default shell for cryosparc_user.

I think that Cryosparc v.2.10 worked. It it possible to get the 2.10 packages - master and worker?

You can specify the version of cryoSPARC you’d like to upgrade/downgrade to by using cryosparcm update --version=<cryosSPARC_version>