Unverified50% confidenceFactExact time
FP8 is a low-precision computing format natively supported by the Blackwell architecture, including two variants—E4M3 and E5M2—and the Tensor Core's FP8 compute throughput is twice that of FP16.
1
Sources
50%
Confidence
Long-term
Relevance
8/8/2026
First Seen
Sources
软件优化让单张B200媲美专用推理芯片
Related Claims
UnverifiedFP8 is a data type natively supported by the NVIDIA Blackwell architecture, available in two formats: E4M3 and E5M2.80% 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.77% 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.76% similarUnverifiedTensorRT supports multiple precision calibrations including FP32, FP16, INT8, and FP874% similarUnverifiedThe NVIDIA Blackwell architecture (B200/GB200) supports FP4 computation, theoretically doubling Tensor Core throughput once again.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/707604API
curl https://kongchang.com/api/v1/knowledge/claims/707604MCP
get_claim(id=707604)