NU-refine CU-MemAlloc error with small number of particles

Troubleshooting
1

Hi,

I am trying to manually generate a ResLog plot (because the current Reslog plot does not incorporate refined CTF params or NU-refine), by selecting ever smaller sets of particles and running NU-refine jobs on each.

NU-refine jobs with larger numbers of particles run fine and complete normally, but jobs with very small numbers of particles (in this case 1750, C3 sym) die with a GPU memory error. This seems like a bug.

Cheers
Oli

image
image1888×462 102 KB

2

Hi @olibclarke, what’s the box size for this dataset?
I see that it has reached 6.3A at the iteration that fails.
For the jobs that succeed, what resolution do they get to?

3

Hi Ali,

The box size is 320px. The successful runs (with more particles) go to higher resolution - e.g. 3.9 at the same iteration where this one dies.

Cheers
Oli