Running MiniMax H3 on a Single 3090: Troubleshooting OOMs and Optimizing Args
knoll_gallagher · reddit · 2026-08-05
A developer shared their experience running the MiniMax H3 video generation model locally on a single RTX 3090 (64GB RAM), providing a solution for common Out of Memory (OOM) errors.
- Counter-intuitive Finding: The author discovered that OOMs were actually caused by GPU demands being too low, leading to dynamic memory growing pains, rather than excessive load.
- Optimizations: By lowering the resolution to 0.2mpx and using int8 + kj sage, they successfully generated 30-second videos in about 16 minutes.
- Launch Args: Recommended adding --vram-headroom 3, --disable-pinned-memory, and --disable-cuda-malloc to mitigate memory issues. The full workflow file is also shared.
Related event: Running MiniMax H3 on a Single RTX 3090(2 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