Hi! I am struggling with NU-homogeneous refinement of my dataset. The job appears runs out of memory. I tried changing the refinement box size, but that dramatically worsened the final resolution of my map. I saw in a past post that lowering the computational minibatch size may help the job complete. How exactly do I lower the computational minibatch size from 2000 (default) to 1000? I do not see it as an input parameter in the GUI. Thanks so much for your help.
Hi @Dnbrawley,
Sorry for the delayed response. Can you post the traceback of the error when the job ran out of memory? It will help to tell which phase of processing ran out of memory.
Most likely in Non-uniform refinement, the job ran out of memory in the “local processing” phases, which do not depend on the computational minibatch size, but rather depend only on the box size of the volume.
When you found that reducing the box size dramatically reduced the refinement resolution, were you already close to the Nyquist rate of the original box size?
What box sizes are you using, and which GPU model?