Local Motion Correction Issue

lorago92 · October 28, 2022, 5:03pm

Hello,

I run Local Motion Correction with extraction box size of 512pix as job parameter and used the output particles for 3D Helical reconstruction. The first iteration of the recontruction made me realise that the particles had different pixel size from the particles I used originally. The original ones were 1/2 f-cropped in patch motion correction (data acquired in super res mode gain y-flipped) and extracted with a box size of 512pix. In opposite, the particle output from Local Motion Correction were not 1/2 f-cropped anymore.
Logically, I thought to re-run Local Motion Correction doubling the box size to 1024pix and, afterwards, downsample 1/2. However, the 1052pix box size Local Motion Correction job gets killed after processing 2-3 micrographs with the following error.

Is there any way you could help to fix this issue?

Cheers
Lorenzo

wtempel · October 31, 2022, 6:18pm

Hi Lorenzo,

What model(s) of GPU are you using?
Is it ensured that no other programs are running on the GPU while that GPU is working on the CryoSPARC job?

lorago92 · October 31, 2022, 7:07pm

Hello,

I have two GPU models available: GeForce RTX 2080 Ti and Quadro RTX 6000.
It’s not. I can definitely try.

Thank you

lorago92 · October 31, 2022, 7:38pm

Hello,

I have the same issue even as a single user of GPUs.

Cheers
Lorenzo

wtempel · October 31, 2022, 7:56pm

Does the error occur even if the job is sent directly to the RTX 6000 card (guide)?

lorago92 · October 31, 2022, 9:05pm

Hello,

Yes, it does. Though I processed more micrographs (about 100 vs 20 with the RTX 2080). However, the issue is the same.

wtempel · November 2, 2022, 4:18pm

If the problem you describe here pertains to the same dataset as the one under Asymmetric Helical refinement (sym or convergence problem), where the resolution is apparently not Nyquist-limited, you may consider down-sampling the movies. Movie down-sampling is not currently implemented in CryoSPARC and would therefore have to be performed with an external program.
Again assuming that cufftAllocFailed occurred during processing of the dataset described in your other topic, you may be able to treat the filament asymmetric unit as a single particle, and reduce the extraction box size accordingly, aiming to extract (and pad) the asymmetric unit, but not necessarily entire filaments.