Hi all,
We currently have a master-worker setup of cryosparc v4.4.1 in addition to the v3.3.2 installed on our cluster. I have successfully imported NU refine and patch CTF jobs into our v4.4.1 (on the same file storage server) and run rbmc and 3D flex. I would now like to import the rbmc particle stacks back onto v3.3.2 to do further NU refinements, local refinements and rounds of CTF refinement as we currently have far greater GPU capability and flexibility on our v3.3.2.
I ran 2-3 rbmc jobs per particle stack (needing to remove duplicates >50Å once or twice first) and joined the stacks using particle sets tools. I then exported each joined rbmc particle stack and imported into v3.3.2. I then attempted to run an NU refine job on v3.3.2 and, after trying to estimate the scale of initial reference, the job failed with the error ‘pycuda._driver.LogicError: cuInit failed: system has unsupported display driver / cuda driver combination’.
Based on the error message itself and when others on the forum have seen this, it suggests that the metadata associated with the particle stacks is incompatible with the current CUDA toolkit and/or NVIDIA drivers installed on each worker node (i.e. the older versions on our v3.3.2 setup).
Is there any job I can run on v4.4.1 after rbmc to circumvent this error before importing back into v3.3.2 or is it simply not possible to continue processing particle stacks on v3.3.2 from a job run in v4.4.1 without upgrading NVIDIA drivers and CUDA toolkit for the v3.3.2 version? I would just upgrade the NVIDIA drivers and install fresh CUDA toolkits on each node of our cluster then point cryosparcw to the new installations, but our cluster is set up in a very unusual way by someone who is no longer around, so I don’t want to take the risk by causing any issues with our current working setup of v3.3.2.
Many thanks!