Non-uniform refinement for simulated dataset

Hi CryoSPARC team,

I simulated a dataset for testing (boring standard params) and while 2D classification, ab initio and homogeneous/heterogeneous refinement work without issue, NU Refine and Local (with NU) Refine both crash with the same error:

Traceback (most recent call last):
  File "cryosparc_master/cryosparc_compute/", line 95, in
  File "cryosparc_master/cryosparc_compute/jobs/local_refine/", line 799, in
  File "cryosparc_master/cryosparc_compute/jobs/nonuniform_refine/", line 51, in
  File "cryosparc_master/cryosparc_compute/jobs/nonuniform_refine/", line 40, in
  File "cryosparc_master/cryosparc_compute/jobs/nonuniform_refine/", line 328, in
  File "cryosparc_master/cryosparc_compute/jobs/nonuniform_refine/", line 402, in
RuntimeError: cuMemHostRegister(mem=0x7e454815b640, num_bytes=2040000, CU_MEMHOSTREGISTER_PORTABLE): CUDA ERROR: (CUDA_ERROR_HOST_MEMORY_ALREADY_REGISTERED) part or all of the requested memory range is already mapped

Have been running dozens of NU Refinements from real data on this system over the last few weeks and they all run correctly.

I’ve simulated it again at a higher mag and larger box, as I wonder whether the small box/moderate angpix is causing it to hiccup, but 2D not complete yet.

Also, would like to ask again when simulating data will be untied from needing an SSD cache?


Seems box size/pixel size related as larger box dataset appears to be fine so far…