Page MenuHomePhabricator

Beamforming GPU memory allocation unstable
Closed, ResolvedPublic


Given an image of 18 slices, and using 16 as the size of a batch loaded onto the GPU, the beamformer runs out of resources. A temporary fix is to set the batch size such, that the image is either completely in one batch, evenly subdivided, or the last batch is relatively large.

As an example, the image A7E3.PA.rf.pc.0000822.nrrd in the pulse corrected set of experiment A7E3 does not beamform because of this reason, when using 16 as the batch size.