Unverified50% confidenceFactExact time
FP4采用1位符号位、2位指数位和1位尾数位的布局
1
Sources
50%
Confidence
Long-term
Relevance
7/7/2026
First Seen
Sources
Related Claims
UnverifiedNVFP4 采用 E2M1 编码（2位指数、1位尾数、1位符号位），以每16个元素为一组配置独立的 FP8 缩放因子77% similarUnverifiedFP8 has two main sub-formats — E4M3 (4-bit exponent + 3-bit mantissa) and E5M2 (5-bit exponent + 2-bit mantissa) — optimized for forward propagation and backward propagation respectively.69% similarUnverifiedNVFP4 采用 E2M1 编码（2 位指数、1 位尾数、1 位符号位），数值网格在零附近更密集、在大数值区间自动稀疏67% similarUnverifiedThe Blackwell architecture introduces fifth-generation Tensor Cores with native support for FP4×FP4 matrix multiply-accumulate operations, using FP32 or FP16 precision for the accumulator.63% similarUnverifiedNVIDIA designed a dedicated Tensor Core instruction mma.kind::mxf4 for NVFP4 in the Blackwell architecture (SM 100/103, such as the B100 and B200 series), and adopted an FP8 scaling factor mechanism with one scale per group of 16 elements.58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/185206API
curl https://kongchang.com/api/v1/knowledge/claims/185206MCP
get_claim(id=185206)