Unverified50% confidenceFactExact time
H100的第四代Tensor Core原生支持FP8,峰值算力达3958 TFLOPS,是FP16的两倍
1
Sources
50%
Confidence
Long-term
Relevance
7/22/2026
First Seen
Sources
Related Claims
UnverifiedV100的Tensor Core在FP16精度下峰值算力可达125 TFLOPS,是前代Pascal架构的5倍以上80% similarUnverifiedH100的Transformer Engine支持FP8精度,张量核心在FP8模式下峰值算力可达3958 TFLOPS,FP16为1979 TFLOPS79% similarUnverifiedP100作为数据中心旗舰配备了专用的FP16 SIMD单元,理论上可实现21.2 TFLOPS的FP16算力,是Pascal家族中唯一拥有快速fp16硬件单元的型号79% similarVerifiedH100 GPU采用英伟达Hopper架构,单卡峰值算力达3,958 TFLOPS(FP16精度)76% similarUnverified原生 FP8 硬件加速最早在 NVIDIA Hopper 架构(H100,2022年)的第四代 Tensor Core 中实现,峰值算力达 3958 TFLOPS74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/593790API
curl https://kongchang.com/api/v1/knowledge/claims/593790MCP
get_claim(id=593790)