RTX 3090 MiniMax H3 speed test compares FP8 and INT8 ConvRot quantization

Community benchmark compares FP8 scaled vs INT8 ConvRot (W8A8) quantization speed for MiniMax H3. Tested on an RTX 3090 24GB with ComfyUI 0.30.0, CUDA 13.0, SageAttention, 17 Euler steps at 0.3 MP for 2-second clips.