Unverified50% confidenceFactExact time
NVIDIA从Hopper架构（H100）开始原生支持FP8计算
1
Sources
50%
Confidence
Long-term
Relevance
9/2/2026
First Seen
Sources
Related Entities
Related Claims
Unverified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 TFLOPS80% similarUnverifiedNVIDIA H100是基于Hopper架构的数据中心GPU，其SXM5版本TDP高达700W79% similarUnverifiedNVIDIA H100 GPU原生支持FP8格式，其吞吐量相较FP16提升约2倍78% similarVerifiedNVIDIA的MIG技术可在A100/H100等高端GPU上实现硬件级别的资源隔离76% similarVerifiedNVIDIA Hopper架构（H100/H200）首次加入FP8 Tensor Core，并引入Transformer Engine自动管理精度转换75% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/843034API
curl https://kongchang.com/api/v1/knowledge/claims/843034MCP
get_claim(id=843034)