Open asarnow opened 10 months ago
Hi Daniel (@asarnow ),
Currently TomoBEAR uses maximum available (which is all identifiable) CPU workers in your system and is limited by MATLAB parallel profile preferences (if any are set). Regarding the module GCTFCtfphaseflipCTFCorrection
, the default there for "execution_method"
is "parallel"
, which means that module already should run in parallel on N tomograms, where N is the number of the available CPU workers.
If you want to use CTFFIND4, you need to setup two parameters:
"ctf_correction_command"
in the "general"
section of the configurations/defaults.json
file"ctf_estimation_method"
value to "ctffind"
in your input configuration file, like below:
...
"GCTFCtfphaseflipCTFCorrection": {
"ctf_estimation_method": "ctffind"
},
...
If that still doesn't help, probably you may also want to try to check your Parallel Preferences in MATLAB to change "Preferred number of workers in a parallel pool".
Let me know if that was helpful!
P.S. The other parameters (jobs_per_node
and gpus_per_node
) you mentioned are for SLURM submission, which is under refactoring currently... but anyway doesn't matter for this particular issue.
I am having trouble finding where the pool size is specified for various steps. I see
matlab_workers
for DynamoTemplateMatching (=1), and the generaljobs_per_node
(=1) andgpus_per_node
(=2). Running GCTFCtfphaseflipCTFCorrection step with CTFFIND4 two jobs are run simultaneously, which I would like to increase. Does it use the gpus_per_node parameter even with CTFFIND4 or is there another variable that should be used?I am running TomoBEAR in a full MATLAB install, on a workstation with 32x physical cores and 2x A6000 GPUs.
Thanks, -da