You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Jan 7, 2025. It is now read-only.
Run some performance tests with cuDNN to see which batch size provides the best performance while still fitting within 4GB of memory on both Kepler and Maxwell. Apparently a multiple of 16 (or 32) is the most efficient.
The text was updated successfully, but these errors were encountered:
lukeyeager
changed the title
Update default batch sizes for cuDNN v3
Update default batch sizes for cuDNN
Jun 22, 2015
Run some performance tests with cuDNN to see which batch size provides the best performance while still fitting within 4GB of memory on both Kepler and Maxwell. Apparently a multiple of 16 (or 32) is the most efficient.
The text was updated successfully, but these errors were encountered: