10-18-2026, 07:50 AM
Did a small comparison this afternoon. My result was mixed: first run looked bad, second run looked normal. I expected the model choice to matter more than it did.
Mostly CUDA, sometimes wrong, usually measured
