Verified75% confidenceFactExact time
H100 SXM5的HBM3带宽约为3.35TB/s，FP16峰值算力约989TFLOPS
3
Sources
75%
Confidence
Long-term
Relevance
7/7/2026
First Seen
Sources
NVFP4量化实战：NVIDIA模型优化器压缩大模型指南
rss6/26/2026
Related Claims
Partially VerifiedA100 80GB采用HBM2e显存，带宽约2039GB/s；H100 SXM采用HBM3，带宽约3.35TB/s，FP64 Tensor Core算力约67 TFLOPS，是A100（约19.5 TFLOPS）的3倍以上86% similarVerified英伟达H100采用的HBM3内存带宽高达3.35TB/s84% similarUnverified英伟达H100基于Hopper架构，采用HBM3内存带宽高达3.35TB/s，单卡FP16算力可达近2000 TFLOPS81% similarUnverifiedNVIDIA H100的HBM3显存带宽高达3.35TB/s，而Ryzen AI Halo受限于LPDDR5x带宽通常在100-300GB/s量级75% similarUnverifiedNVIDIA H100的HBM显存带宽约为3.35 TB/s，在自回归生成任务中利用率往往不足10%74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/187833API
curl https://kongchang.com/api/v1/knowledge/claims/187833MCP
get_claim(id=187833)