Unverified50% confidenceFactExact time
Cerebras WSE-3片上内存高达44GB SRAM，带宽达到21 PB/s，比GPU的HBM内存快约1000倍
1
Sources
50%
Confidence
Long-term
Relevance
7/8/2026
First Seen
Sources
GPT旗舰模型集成Codex：750 Token/秒推理速度详解
bilibili声译看世界7/7/2026
Related Claims
UnverifiedCerebras WSE将约40GB的SRAM直接集成在晶圆上，片上带宽超过20 PB/s82% similarPartially 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79% similarVerifiedA100 GPU的HBM带宽约2TB/s，而片上SRAM带宽可达19TB/s，相差近10倍77% similarUnverifiedGPU HBM带宽通常在2-3 TB/s量级，H100为3.35 TB/s，而PCIe 5.0连接的Host DRAM有效带宽仅约64 GB/s，两者相差近50倍77% similarUnverifiedP100采用4颗HBM2堆叠，总带宽达732 GB/s，远超同期消费级GTX 1080的320 GB/s76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/291490API
curl https://kongchang.com/api/v1/knowledge/claims/291490MCP
get_claim(id=291490)