Unverified88% confidenceFactTime unknown
A100 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
1
Sources
88%
Confidence
Medium-term (~90 days)
Relevance
-
First Seen
Sources
Related Entities
Related Claims
VerifiedH100 SXM5的HBM3带宽约为3.35TB/s，FP16峰值算力约989TFLOPS86% 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% similarVerifiedA100 GPU的HBM带宽约2TB/s，而片上SRAM带宽可达19TB/s，相差近10倍81% similarVerifiedRTX 4090的理论FP16算力为82.6 TFLOPS，显存带宽为1008 GB/s80% similarUnverified被管制的NVIDIA芯片配备80GB HBM2e内存，内存带宽达2TB/s，NVLink互联带宽达600GB/s79% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/468515API
curl https://kongchang.com/api/v1/knowledge/claims/468515MCP
get_claim(id=468515)