MiniMax H3 video gen runs locally on M5 Ultra: 768p in ~2m22s with optimizations
bakawolf123 · reddit · 2026-09-22
A Reddit user benchmarked MiniMax H3 video generation (q8 FL2VA, vpipe engine) locally on a Mac Studio M5 Ultra (36/80 cores, 256GB unified memory), generating 5s clips at 6 steps / 24fps.
Results:
- 960x544: 1m35s with Sol attention + int8 GEMM + turbo LoRA, vs 1m48s baseline
- 1366x768: 2m22s optimized, vs 3m56s baseline
Fans only spin up near the end of 768p runs (halfway through without optimizations), making speeds quite usable, though 768p output shows some corruption. A useful snapshot of consumer unified-memory hardware running video gen models locally.
More from Infra
- Gas-for-AI-datacenters forum co-located with CoRL shows energy demand spillover — carlosdponx · 2026-11-12
- a16z podcast: Mike Volpi on Cisco's 70+ acquisition playbook and the AI buildout — a16z · 2026-11-12
- AMD to pass Intel in server revenue by mid-2027 as share gains 6 points a year — BenBajarin · 2026-11-12
- Study: Reflective Data Center Roofs Actually Cool Daytime Ground by 1.8-2.2°C — SumitGup · 2026-11-12
- Tsinghua's BaGuaLu-2 tames MoE expert-load spikes at runtime, accepted to SC26 — jiqizhixin · 2026-11-12
- antirez backs low-bit DeepSeek 4.1 Flash over Qwen 27B dense on Apple M5 Macs — antirez · 2026-11-12