Unverified50% confidenceFactExact time
NVFP4在NVIDIA Blackwell架构的Tensor Core中获得了原生硬件支持，无需额外的反量化计算开销
1
Sources
50%
Confidence
Long-term
Relevance
8/24/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedNVIDIA 在 Blackwell 架构（SM 100/103，如 B100、B200 系列）中为 NVFP4 设计了专用 Tensor Core 指令 mma.kind::mxf4，并采用每 16 个元素为一组的 FP8 缩放因子机制79% similarUnverifiedNVIDIA Blackwell架构（GB200/B200系列）在硬件层面引入了原生FP4张量核心77% similarPartially VerifiedNVFP4是NVIDIA推出的4位浮点（FP4）量化格式，专为Blackwell架构GPU的硬件加速能力设计76% similarUnverifiedNVIDIA Blackwell架构（B200/GB200）支持FP4计算，理论上将Tensor Core吞吐量再次翻倍76% similarUnverifiedMXFP4已获得NVIDIA Blackwell架构GPU的硬件原生支持76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/795627API
curl https://kongchang.com/api/v1/knowledge/claims/795627MCP
get_claim(id=795627)