MiniMax H3 Text-to-Video Successfully Runs on a Single DGX Spark
Scobleizer · x · 2026-08-04
Developer @aijoey successfully deployed the MiniMax H3 text-to-video and audio workflow on a single NVIDIA DGX Spark.
- Technical Details: Overcame initial out-of-memory errors by resolving SM121 compatibility issues, utilizing online FP8 quantization, and pinning a specific vLLM-Omni build.
- Performance: Generating a video currently takes about 2.5 minutes.
- Open Source: The author has published the full configuration recipe, patches, and troubleshooting logs on GitHub for reproducibility.
More from Infra
- AI Shipping Labs Launches 'Inference Engineering' Book Club on Aug 10 — Al_Grigor · 2026-08-04
- FMS 2026 Kicks Off: Samsung, SK Hynix, Nvidia to Discuss Memory-Centric AI Vision — AccBalanced · 2026-08-04
- AMD Faces AI Earnings Test as Open Models Threaten Closed Labs' Margins — AccBalanced · 2026-08-04
- RTX 5090 Hits 83°C Running MiniMax H3 Locally — Careless-Constant-33 · 2026-08-04
- AMD MI355X runs Kimi K3 on single node, 3.8x throughput of B200 setup, better cost-performance than B300 — 机器之心 · 2026-08-04
- Spectrum Acceleration for MiniMax H3 in ComfyUI: Up to 34% Lower Inference Time — marres · 2026-08-04