Benchmarking MiniMax H3: Doubling Video Length Nearly Triples Render Time
jtreminio · reddit · 2026-08-05
A developer benchmarked the MiniMax H3 video generation model, revealing a non-linear growth pattern in rendering time: doubling the video length nearly triples the render time. For instance, a 5-second clip takes 2 minutes, while a 30-second clip takes 43 minutes.
The author explains the underlying mechanism: H3 chops video into a grid, and due to the $O(n^2)$ complexity of the attention mechanism, every second added increases the computational cost of all previous frames. This means the marginal rendering cost per second of video continuously rises.
Related event: MiniMax H3 Video Generation Shows Non-linear Time Scaling(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