Running MiniMax H3 on 12GB VRAM: quantization, Turbo LoRAs and attention backends compared
Possible_Mood676 · reddit · 2026-09-11
A Reddit user is setting up MiniMax H3 video generation in ComfyUI on an RTX 4070 (12GB VRAM / 32GB RAM), targeting 5-second 768p I2V/First-to-Last clips. They report that unpruned FP8 models cause heavy paging to system RAM, and are polling the community on: quantization choices (pruned INT8 convrot vs GGUF Q4KM/Q3KM, plus text encoder quants) that preserve faces and audio; whether the 4-step LiteX2V v1.1 Turbo LoRA is still the best speed/quality option versus 8-step variants or other LoRA families; and which attention backend (H3 SLA, SageAttention, Comfy Kitchen) performs best on Ada Lovelace cards. The post is a help request without answers yet, but maps the current decision space for low-VRAM local deployment.
More from Infra
- SemiAnalysis: OpenAI reportedly negotiating ~1GW TPU commitment while seeking sites — dylan522p · 2026-10-28
- I tested 11 video upscaling methods for AI video: local, Topaz and cloud compared — lopas99 · 2026-10-28
- Thermo-backed CTO says AI is in its dial-up phase, touts disaggregated compute for an ultrafast 'broadband moment' — beffjezos · 2026-10-28
- DeepSeek-V4.1-Flash Q2 runs at 41 tok/s on one RTX 3090, 90% token match vs API — musicurgy · 2026-10-28
- USearch v3 RC ships after 1,000 days, its libraries now embedded in 550K-star projects — viglovikov · 2026-10-28
- SK Hynix says memory demand will outpace supply through 2030 — Chemical_Hawk_6307 · 2026-10-28