Because the batch size should be kept low at 4 or 8, the high end consumer GPUs with 16 or 24gb of VRAM are under-utilized. It could be great to find a usage for the large amount of Vram that is left-over, like main RVC branch using it for caching dataset.
From what I understand, pushing the batch size at 16 or 24 (even for large datasets of 1 hour) isn't recommended, so it could be nice to find usage of this Vram.
aitronssesin
commented
1 month ago
Because the batch size should be kept low at 4 or 8, the high end consumer GPUs with 16 or 24gb of VRAM are under-utilized. It could be great to find a usage for the large amount of Vram that is left-over, like main RVC branch using it for caching dataset. From what I understand, pushing the batch size at 16 or 24 (even for large datasets of 1 hour) isn't recommended, so it could be nice to find usage of this Vram.
Desktop Details: