MiniMax H3 Needs 50 GB — Here's How It Runs on a 12 GB GPU
fa6637 · reddit · 2026-11-12
A Reddit post shares a YouTube walkthrough showing how MiniMax H3, whose weights take roughly 50 GB, can be squeezed onto a consumer GPU with only 12 GB of memory via quantization. The video covers the compression setup and real runtime results, useful for anyone running large open-weight models locally.
More from Infra
- Curated 9-paper reading list for LLM quantization beginners, from QLoRA to BitNet — East-Muffin-6472 · 2026-11-12
- llama.cpp PR #31915 fixes Gemma4 tool calling, user reports it "completely fixed" agentic coding — jacek2023 · 2026-11-12
- Nvidia board reportedly working with outside firm to find Jensen Huang successor — kimmonismus · 2026-11-12
- RDNA3-tuned llama.cpp fork doubles Qwen3.8-27B speed on RX 7900 XTX to ~100 tok/s — Jebbyk1 · 2026-11-12
- Ming Chi Kuo: Huang is irreplaceable to Nvidia; Rubin mass production slips to 2027 on HBM4 — zephyr_z9 · 2026-11-12
- Breakdown: Only 3% of Anthropic's 2.68M Chips Are Nvidia, Lowest in AI — teortaxesTex · 2026-11-12