3 ms·
CUDA has had managed memory that pages between VRAM and system RAM for a decade. Problem is doing so is unusably slow for AI purposes. Seems like an unnecessary
by kcb 7mo ago
CUDA has had managed memory that pages between VRAM and system RAM for a decade. Problem is doing so is unusably slow for AI purposes. Seems like an unnecessary layer here.
- yjtpesesu2 7mo ago[dead]
- hrmtst93837 7mo ago[flagged]