Resurrecting explicit GPU binding for v2

solved

#1

It was really handy to have the option to explicitly bind jobs to GPUs. Version 1 had it, v2 does not. We need it to prevent collisions with other applications using the GPU. Sorry for resurrecting this but the topic that had this has been archived.


Setting for "CUDA device numbers"?
#2

Hi @abreuer,

Thanks for bringing this up again - you may already know this but it is currently possible to select GPUs that cryoSPARC will schedule jobs on (by default, all GPUs are enabled). At least temporarily that may help to allocate particular GPUs to particular programs.


#3

Hi @abreuer

Following up to what @apunjani said, you can find the documentation on limiting the GPU’s that cryoSPARC will schedule jobs on at the end of this section here: https://cryosparc.com/docs/reference/install/#installation-standalone-worker

bin/cryosparcw connect --worker <worker_hostname> --master <master_hostname> --port <port_num> --update --gpus [0,1,2,3]


#4

Hi @abreuer,

This feature has been released in cryoSPARC v2.12+!