Running MiniMax Video Model Locally: M3 Ultra Takes 2.6 Hours for 10s Video
cocktailpeanut · x · 2026-08-05
A developer has successfully run MiniMax's video generation model locally on a Mac powered by an M3 Ultra chip.
- Test Parameters: Resolution set to 1280x768, 20 inference steps, generating a 10-second video.
- Performance: The entire generation process took approximately 2 hours and 39 minutes.
- Outlook: The author noted that while future M5 Max chips might offer speed improvements via NAX, the reduction in core count (from 80 to 40) means overall performance gains may be limited.
More from Infra
- OpenAI Engineer Details WebRTC Improvements Powering GPT-Live — juberti · 2026-08-05
- Musk: Starmind Satellite Design Will Improve Ground Data Centers — elonmusk · 2026-08-05
- SpaceX and NVIDIA Plan Up to One Million AI Compute Satellites — ns123abc · 2026-08-05
- 20B Model Maple-Preview Runs at 200+ tokens/s on Mac Mini, Solves IMO Math — tylerbruno05 · 2026-08-05
- Open Source AI Hosting Hits Compute Wall: DeepSeek Flash Offline Amid GPU Crunch — bindureddy · 2026-08-05
- Full Kimi K3 model runs on 16x GB10 cluster at 20+ TPS — ciprianveg · 2026-08-05