Using gpu-split 18,18,18,18,18,18,18,18 on 8x RTX 3090s, 21/20/19/19/19/5 GB (6x cards, 2x unused) actual usage with overhead taken account of.
gpu-split 18,18,18,18,18,18,18,18
No action required, just notating in these for people to see VRAM usage in these quants. Thanks for these, @Dracones !
· Sign up or log in to comment