Hi Everyone,
I am running the Cryosparc2 2.15 with CUDA 8.0. It seems working well.
But when I check Nvidia-smi, My 2080 graphics indicate Cuda 10.0 working with.
Is there any advantage when I upgrade CUDA in Cryosparc2?
I just curious about what is the most impact for working with the higher CUDA version.
Thanks,
Jinseo Park
Hi @Jinseo,
How did you determine that cryoSPARC is running with CUDA 8.0? Why I ask is usually when you have a default CUDA installed (CUDA 10.0 in your case), if cryoSPARC is not also compiled against that version, jobs will fail (meaning it is more likely that cryoSPARC is already running with CUDA 10).
Can you double check in cryosparc2_worker/config.sh if cryoSPARC is in fact using CUDA 8.0?
If it is, it’s definitely worth recompiling cryoSPARC with CUDA 10.0. In general, it’s better to use the latest software with your hardware- historical bugs may have been fixed, and overall stability may increase. You also of course get access to the latest features. If you’re looking for the official release notes about CUDA 10, read this article:
To recompile cryoSPARC with a different CUDA version, see this post:
https://guide.cryosparc.com/setup-configuration-and-management/management-and-monitoring/cryosparcw#cryosparcw-newcuda-less-than-path-greater-than
Please note that although I do recommend using the latest CUDA Toolkit, we don’t support CUDA 11.0+ as of v2.15.0 and v2.15.2-live_privatebeta just yet- the next release will have support for it!
Hi Stephan,
My cryosparc 2 was installed with CUDA-8.
When I checked config.sh, path to CUDA still was indicating CUDA-8.0 folder.
But /usr/local/ has multiple CUDA versions. The latest is CUDA-10.0 also linked with /usr/local/cuda.
So I re-complied using command “/bin/cryosparcw newcuda /usr/local/cuda”.
It complied successfully, and config.sh CUDA path was changed to /usr/local/cuda
Now it is working well.
Thanks,
Jinseo