Related to the trouble in:
3DFlex Dependencies; Building pycuda?
Worker installation of v4.1.0 on CPU-only node without GPUs fails, at least in my case (Rocky Linux 8).
There seems noway around a forced attempt to install pycuda on the worker, which fails and with the trick described in above link.
Is CPU-only workers not supported anymore?
Perhaps the possibility for an optional “–nocuda” flag to “cryosparcw update” could be nice?
EDIT: It seems like the CPU-only worker install actually works (CPU-only jobs), although many CUDA and gcc related errors are printed.