Dual-GPU Optimization Speeds Up MiniMax-H3 Video Generation 8x
multimodalart · x · 2026-08-10
A developer shared engineering practices for massively accelerating MiniMax-H3 (a 33B omni-modal model for video and audio generation) within the Diffusers framework. Using a dual-GPU setup (48GB+20GB), the project achieved 9.7s per image and 44.2s per 5s video clip, yielding 16x and 8x speedups respectively over the base configuration.
The core optimization targets reducing the "fixed overhead" of frequent model loading and unloading, rather than just accelerating denoising:
- Caching the quantized text encoder to disk
- Keeping the text encoder resident on the second 20GB GPU
- Keeping the transformer resident during decoding (balancing VRAM via VAE fp16)
The project also supports first/last frame specification, reference-based generation, and batch generation. It is open-sourced on GitHub.
Related event: MiniMax H3 Local Long Video Generation and Optimization Practices(7 posts)→
More from Infra
- MLCommons launches MLPerf Endpoints, a benchmark that measures inference endpoints the way buyers actually purchase them — TheKanter · 2026-10-16
- Neocloud M&A decoded: CoreWeave, Nebius, Crusoe, Nscale all bet margins vanish beyond GPU scarcity — AccBalanced · 2026-10-16
- AI-Driven Timing-Aware Power Optimization Uses Learned Ranking With Baseline-Preserving Repair — Amit Sharma · 2026-10-16
- Nscale AI datacenter plagued by coolant leaks, transformer fire, 10 safety violations — saranormous · 2026-10-16
- WSL 3.0 ships with native containers and Linux 6.18 kernel, syscall overhead down ~16% — solyarisoftware · 2026-10-16
- Are 22 GB modded RTX 2080 Ti cards a viable budget local LLM option? — Dave86ch · 2026-10-16