AnalysisVisual AIAugust 29, 2026

MiniMax H3 runs faster on consumer GPUs via VDN-H3 and sparse attention

Independent developers report speedups for MiniMax H3 video generation on consumer hardware. VDN-H3 on M5 Pro 24GB reaches ~2.6× faster H3 generation, while auto-scheduled sparse attention on a single RTX 4090 achieves 112s/223s/334s for native 720p to 1440p second sampling.

2 sources

More stories today

Open the live feed