Installing 3DFlex Dependencies on HPC

Hi together,

we recently updated cryosparc to the latest version and want to use 3DFlex from now on. However, I was wondering if we just need to install CUDA Toolkit 11.7, PyTorch with CUDA Toolkit 11.7 and PyCUDA with CUDA Toolkit 11.7? We are using cryosparc in a HPC environment and have all these dependencies already compiled for each individual architecture. If nothing else is required I would rather adapt the submission scripts to load the respective dependencies as modules. Please let me know if you still suggest to install these dependencies agein.


Configuring the 3DFlex prerequisites inside the CryoSPARC environment was a non-trivial exercise. It may be possible to simulate this environment by reusing components such as CUDA toolkit and PyTorch already available elsewhere on the system, but, unfortunately, we do not have the bandwidth to significantly assist in such an exercise.
Should you decide to try it, please update this topic on your progress.

Thank you for your reply. I am going to test it on one of our GPU Nodes and compare it with the recommended procedure.