Dear Colleagues,
I’m getting errors running motion correction. The micrographs I’m using are from tilt series. ::(unfortunately the tomography didn’t work out like we wanted it to, so we will treat each tilt as a micrograph and process the data through single particle)::
I tried running patch motion correction on the imported movies. I’m hitting a wall with the following error:
Please help?
Best,
CBEN
Hi @CBEN,
What GPUs do you have? This error means that there isn’t enough GPU memory to complete the motion correction job… It’s very common on GPUs that have less than 11 GB of memory.
–Harris
Thanks Harris. That seems to be the problem. We have 4 GPU with 11 GB each. Do you have any suggestions to remedy this? I’ve tried running the job in Low Memory Mode, but that’s been going for quite some time without any progress.
-CBEN
Hi @CBEN,
Are you working with super-resolution movies? That’s the only case where we have found 11GB to be insufficient. Low memory mode is usually the successful work-around. What do you mean by “without any progress”? Is the job state listed as “Running”, or is it something else (e.g. “Launced”/“Started”)?
Harris
Yes. They are super-resolution. I’m attaching a screenshot of the status of this job currently running in Low Memory Mode. It’s been in this state for more than an hour now, and hasn’t moved any further than loading the gain reference.
Hmmm… it definitely should not take that long. Are you able to log into the machine and use nvidia-smi to see if the process is actually using the GPU? If you don’t see any activity, I’d recommend killing the job, clearing it, and restarting it.
Hello Harris. You were right, the GPU wasn’t doing anything, so I restarted it after killing and clearing the job like you recommended. Unfortunately, that didn’t remedy the situation. However, I found a work-around. I excluded more of the later frames, and luckily I don’t need them either. After doing that I’m able to run the job fine, even without Low Memory Mode. Thanks!
-CBEN
Glad you found a workaround! You can also try using output F-crop (set it to 3/4 or 1/2), that may reduce the memory usage as well.
–Harris
