AI Briefing
KO

llama.cpp Adds Blackwell NVFP4 Support

·2026.04.29 17:31

Key point

llama.cpp b8967 adds native NVFP4 support for Blackwell.

Details

llama.cpp release b8967 added native NVFP4 support for Blackwell.

  • Key change: Blackwell native NVFP4 support was added to ggml-cuda.
quantizationllama.cppnvfp4blackwell
Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.

Related

RedditK2-Horizon GGUF Quantizations ReleasedRedditGewell Inference Engine Optimized for Gemma 4 31B ReleasedTLDR AINVIDIA Releases NVFP4 Quantized Model of DeepSeek-V4-Pro-0813TLDR AIUnsloth Dynamic 3.0 GGUF ReleasedNVIDIA BlogNVIDIA Reveals Five Companies in AI-Powered Clean Energy ProjectsPyTorchKRPrismML Releases 'Ternary Bonsai 2 27B', a Ternary Quantized Model Maintaining 98.2% of FP16 Performance