Draft-then-refine cuts H3 video latency 27.7x on one GB200
- Source
- x.com
- Date
The real breakthrough in the NVIDIA SANA team’s Sol Engine work on MiniMax H3！ By splitting generation into a 4-step low-res H3 draft and a 3-step LTX refinement pass at target resolution with Sol-Attn, they’ve crushed 10s 768p latency on a single GB200 from 414s down to 14.93s (27.7x speedup). Replacing heavy VAE decodes with TAEH3/TAEHV while holding the latents stable for refinement is a masterclass in co-designing sampling topology with hardware kernel acceleration. When inference latency collapses this dramatically, unit economics fundamentally shift: a single node can suddenly serve 378K videos a month at 97%+ GPU margins. This is how high-fidelity AI video moves from asynchronous batch rendering to near-instant, interactive infrastructure. Huge respect to the team for setting a new engineering bar for our open-weights ecosystem! 🫡🩵
Context
NVIDIA's H3 Super Acceleration pipeline saves work by generating a small draft before refining it. It runs MiniMax H3 for four steps at 896×512, then applies three LTX refinement steps at the target resolution. The engineering report reports 14.93 seconds for a ten-second 768p clip on one GB200, compared with a 414.1-second SGLang baseline.
That 27.7-fold speedup combines separately measured stage timings and excludes model loading and warmup. The report also says the changed generation path can alter detail, motion or audio, so the result represents a speed-quality tradeoff.
Checking sign-in…
Loading comments…