VC-Attention: training-free low-bit attention hits 1.9x on B200, beating FlashAttention-4
xiuyu_l · x · 2026-09-17
Nunchux AI, with researchers from MIT, CMU, UC Berkeley, Stanford and NVIDIA, released VC-Attention: fast, accurate low-bit attention without retraining.
- On MiniMax-H3 video generation (243 frames, 1344×768): 1.59x speedup on B200 and 1.51x on B300 over BF16 FlashAttention-4; the proprietary Nunchux Attention extension reaches 1.91x / 1.83x.
- Better fidelity than SageAttention2: 20.2 dB mean PSNR vs 19.9 dB across 100 prompts.
- Two innovations: V-Smooth reduces value quantization error (enabled on the first quarter of denoising steps); ExpCast-FP8 accelerates softmax.
- Compatible with existing sparse attention methods; targeted at multimodal/world-model inference where attention dominates cost.
Blog and technical report available.
Related event: VC-Attention: Training-Free Low-Bit Attention Speeds Up Video Generation(5 posts)→
More from Infra
- To dethrone Postgres, new databases must prioritize ecosystem integrations from day one — mgill25 · 2026-11-02
- ESP32-powered 4-inch touchscreen builds a TUI dashboard for monitoring a DGX inference cluster — CATLLM · 2026-11-02
- OpenRouter crosses 10 quadrillion annualized token run rate, doubling every 10.5 weeks — deedydas · 2026-11-02
- Palmer Luckey asks: can you use datacenter RAM for things prohibited by the operator's terms of service? — avaitopiper · 2026-11-02
- Disable MTP for 20-30% faster local LLM inference on limited RAM — InternalMode8159 · 2026-11-02
- Kioxia up 216% in a year but down 68% from its June 2026 peak — zephyr_z9 · 2026-11-02