Which model to pick for RTX 5090?
minimax_h3_ref2va_int8_convrot.safetensors (34 GB)
minimax_h3_ref2va_pruned_int8_convrot.safetensors (21 GB)
For an RTX 5090?
Your GPU is 32gb of VRAM so you should go with the minimax_h3_ref2va_pruned_int8_convrot.safetensors
The non-pruned int8_convrot (which I downloaded by mistake lol) did run on my 4090, although there was definitely slowdown from swapping to system RAM. With a warm prompt encoding, the default T2V workflow took about 100 seconds (obviously more if the text encoder has to run.) Your 5090 oughta hold up better since it has to offload less.
Both would work technically, the pruned version is a bit more experimental. See blog post: https://blog.comfy.org/p/minimax-h3-day-0-support-in-comfyui
We found that the model's modulation weights (~40% of the total parameters) could be pruned and replaced with a functionally equivalent lookup table, dramatically shrinking the memory footprint with no loss in output quality.