Skip to content
AI IntelligenceSep 26, 2026AI Intelligence
Article

Google TPU runs Kimi 57% faster than Nvidia GPU using DeepSeek framework

A startup team founded by vLLM creators has demonstrated that Google's TPU runs the Kimi model 57% faster than Nvidia's GPU. The optimization was achieved by utilizing the DeepSeek inference framework.

Frontier EditorialSource: 量子位
01

Source Brief

Google TPU runs Kimi 57% faster than Nvidia GPU using DeepSeek framework: A startup team founded by vLLM creators has demonstrated that Google's TPU runs the Kimi model 57% faster than Nvidia's GPU. The optimization was achieved by utilizing the DeepSeek inference framework.

02