NVIDIA optimizes MiniMax H3: Single GB200 produces 378K videos monthly
songhan_mit · x · 2026-08-22
NVIDIA's SANA team released a super-acceleration scheme for the MiniMax H3 video generation model. Using Sol-Engine and a low-res draft + high-res refinement strategy, they achieved 22.2x (5s video) to 27.7x (10s video) speedup on a single GB200. Economically, at full utilization, the GPU-only gross margin exceeds 97%, capable of producing 378,000 five-second 768p videos per month.
More from Infra
- Muse Spark Open runs locally on a Mac M3 Ultra via llama.cpp: frontier model on desktop — pcuenq · 2026-10-16
- Muse-Spark open-weights drop: 816B-A42B MoE model gets 359GB MLX 3-bit build — ns123abc · 2026-10-16
- vLLM ships Day-0 LoRA RL for Muse Spark Open, lifting AIME from 0.56 to 0.85 — vllm_project · 2026-10-16
- DSpark draft model for Muse Spark Open proposes 7 tokens per pass with 23-42% acceptance — vllm_project · 2026-10-16
- Cloudflare Stream's WebRTC ultra-low-latency live streaming is now Generally Available — ritakozlov · 2026-10-16
- YC-backed Overshoot launches inference optimization, claims $5M saved for AI teams — ycombinator · 2026-10-16