Update - Still figuring this out, disregard for now
I have a 3DVar Display job running on GPU 0. Meanwhile, I have a Het Refine in a different project that’s failing repeatedly due to CudaMemaAlloc. I assumed I would need to reboot the computer, since GPU 1 is the primary GPU and can’t be reset. However, I noticed that the Het Refine was actually queueing on GPU 0, and if I manually selected GPU 0, the job would start instead of waiting for the 3DVar Display to finish.
If I force the 2nd job on to GPU 1, it seems OK.