跳到正文
原文
SemiAnalysis_· @SemiAnalysis_ · X·· 1 天前AI 评分61

SemiAnalysis:Google TorchTPU vLLM 在 TPUv7 FP8 上以 4 芯片在多数性能/TCO 前沿胜过 GB300 NVL72 FP4

AI 导读

SemiAnalysis 指出,Google TorchTPU vLLM 在 TPUv7 FP8 上仅用 4 芯片,就能在多数性能/TCO 前沿上胜过 GB300 NVL72 FP4;在低交互性场景下 GB300 FP4 仍占优,但 TPUv7 FP8 依然胜过 GB300 FP8。

正文 · 原文

What is surprising is that Google TorchTPU vLLM on TPUv7  FP8 with just 4 chips was able to beat GB300 NVL72 FP4 on performance per TCO across most of the pareto frontier. At low interactivity, gb300 fp4 still wins but at low interactivity tpuv7 fp8 still beats gb300 fp8. (3/4)

来源:SemiAnalysis_ · x.com