Job attempts to run on GPU not in CUDA_VISIBLE_DEVICES

[wtempel]

@Yehuda

Also, all 8 GPUs are listed in the job builder.

Please can you post a screenshot, and updated output form the command

cryosparcm cli "get_scheduler_targets()"

@Yehuda

After reconnecting worker, two GPUs are in use. A third job was queued until one of the running jobs was done.

Was not this the intended behavior?

@Yehuda

A job cloned from one of the completed jobs is running much slower.

Please compare timings of the completed subroutines between both jobs and post the discrepancies, in addition to details like job type, cache settings, if applicable, etc.