AI 导读
SemiAnalysis 表示，在同等口径的 InferenceX 对比中，Google TPUv7 Ironwood 在新增的 TorchTPU 外部推理栈上每美元性能比 NVIDIA Blackwell Ultra 高 50%。TorchTPU 把原生 PyTorch 带到 TPU，Google 同时开源了一批 Pallas 推理 kernel，SemiAnalysis 认为这为 TPU 快速外部化打下基础，并称 TPU 外部化正朝正确方向全速推进。
正文
ALERT🚨🚨: On apples-to-apples InferenceX Comparisons, TPUv7 Ironwood achieves 50% better perf per dollar than Blackwell Ultra on the new TorchTPU external inference stack!
TorchTPU brings native PyTorch to TPUs, & along with Google open-sourcing a bunch of their Pallas inference kernels, Google has laid out a solid foundation for TPU to rapidly externalize. We at SemiAnalysis strongly believe that TPU externalization is heading in the right direction and moving full steam ahead.
来源：@SemiAnalysis_ · x.com