MiniMax H3 C++/GGML Implementation Benchmarks: 10s Video in 119-130s on RTX 5090
Acceptable-Cycle4645 · reddit · 2026-08-08
The author shares progress on a C++/GGML implementation of MiniMax H3 and seeks performance baselines. Current benchmarks on RTX 5090 show 480x832, 10 steps, 10s video generation takes 119-130s with 24GB peak VRAM. Different memory-saving modes trade off speed: level 2 uses 11.4GB but takes 183s, level 3 uses 12.6GB and takes 163s. The author notes audio quality at 10 steps is better than the Python reference, and TTS word error rate is lower.
More from coding & agent
- Cloudflare on Ensuring Dashboard Agent Quality with Evals — threepointone · 2026-08-08
- "AI Expert" != Gets S**t Done: Dev Rants on Theoretical AI Practices — BenSimonDev · 2026-08-08
- WebMCP Gains Traction and Knowing When to Stop an Agent Loop — rseroter · 2026-08-08
- Developer Claims the 'async' Keyword is Now Effectively Legacy — samgoodwin89 · 2026-08-08
- AI Computer Use Poses Serious Risks: Agents Reported Deleting Files and Breaking OS — ericelliott_ · 2026-08-08
- AI Automatically Tracks Historical Feedback and Notifies Customers: Indie Dev Workflow — gabriel1 · 2026-08-08