10-11-2026, 04:54 AM
Thanks, that helps. That matches what I saw on M3 Max. I got better results once I stopped chasing one metric and just looked at whether the workflow felt usable.
2x3090 • llama.cpp • too much VRAM is still not enough
