X-MinimaxH3: Local acceleration for MiniMax H3 on consumer GPUs
This_Temporary_8537 · reddit · 2026-08-30
An indie dev open-sourced X-MinimaxH3, accelerating MiniMax H3 on consumer NVIDIA GPUs (tested on RTX 4090):
- Benchmarks: Native 720p→1440p second sampling takes 112s/223s/334s for 5s/10s/15s clips.
- Technique: Reuses latent states for second sampling instead of frame-by-frame upscaling; uses a quality-aware scheduler to allocate compute budgets dynamically.
- Profiles: Supports 8GB (W4A8), 16GB (INT8), and 24GB (INT8) via ComfyUI, REST API, or Web UI.
- Integration: Connects to ComfyUI via HTTP to avoid double model loading.
More from Infra
- SkyRL ships Day-0 RL support for Beam with +62% LoRA decode throughput in vLLM — vllm_project · 2026-10-16
- DFlash2 delivers up to 3.03x decode-only speedup for Beam across five benchmarks — vllm_project · 2026-10-16
- vLLM's Beam serving kernels cut GB200 forward time from 8.15 to 6.47 ms at TP8 batch 128 — vllm_project · 2026-10-16
- Reflection's 501B MoE Beam hits 463 tok/s on 8x GB200 with NVFP4 at day 0 — zhyncs42 · 2026-10-16
- Nemotron 3.5 Bolt runs 16 concurrent streams on a single DGX Spark — NVIDIAAI · 2026-10-16
- Why your computer can't store 0.2: an IEEE 754 primer — Franc0Fernand0 · 2026-10-16