Gpu batch size in refinement

Hello, I was wondering if increasing the gpu batch size from default to ~GPU_MEMORY_SIZE (Our gpu has 24 GB memory, thus say changing default batch size 500 to 2000 for 512pix particles) in refinement jobs e.g., local refinement will significantly increase the speed?

If so, will this impact the results or not? Thanks a lot.

If you’re read limited, it’ll help a bit. Not by a factor of four, though. :wink:

If other settings remain unchanged, it should not effect results obtained, although of course you will see variance between runs for cryo-EM data but the general trends should remain.

1 Like

Thanks a lot! Very helpful to know.