search

cmosig / sentle

Sentinel-1 & Sentinel-2 data cubes at large scale (bigger-than-memory) on any machine with integrated cloud detection, snow masking, harmonization, merging, and temporal composites.
MIT License
22 stars 2 forks source link

memory leak for GPU memory during cloud detection #18

Closed cmosig closed 1 month ago

cmosig commented 3 months ago

It seems like with each inference batch the GPU memory usage is growing by about 2MiB and never decreasing. Theoretically it should remain constant.

Requires minimal reproduction example.

cmosig commented 1 month ago

magically fixed in after dropping dask

cmosig commented 1 month ago

still there...

cmosig commented 1 month ago

cannot reproduce, VRAM usage increases a bit in the beginning but then settles at a fixed amount. E.g., 30659MiB for 50 workers.