Primary sourceTeortaxes
View sourceEvent ·Multi-source
SemiAnalysis 实测华为 Ascend 950PR:性能约为 H100 的 44%
SemiAnalysis 拿到华为 Ascend 950PR(950PR_9579)进行实测。该芯片功耗 600W,BF16 算力中位数为 325 TFLOPS,约为 H100 的 44%、B300 的 21%,FLOPS/W 比 Hopper 低 2 倍,实际达到 A100 档位。有效内存带宽约 1.4 TB/s,低于 H100 SXM5 的 3TB/s。不过在计算密集型负载下,950PR 的 MFU 中位数达 86%,超过所有 Nvidia 芯片,内存密集型场景 MFU 为 41.5%,低于 H100 的 47%。
Current read
SemiAnalysis 真拿到了华为 Ascend 950PR 实测:算力只有 H100 的 44%,但算子效率比 Nvidia 还高,数据很细。
11 sources71° AI heatLast updated 2026/10/22 11:30:32
Evidence
11 sourcesRelated source 1The Information
View sourceRelated source 2marktechpost
View sourceRelated source 3arXiv cs.LG
View sourceRelated source 4techcrunch
View sourceRelated source 5kimmonismus
View sourceRelated source 6rohanpaul_ai
View sourceRelated source 7Tech in Asia
View sourceRelated source 8PyTorch
View sourceRelated source 9IT之家
View sourceRelated source 10NVIDIA
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.