Google Unveils 100-Petaflop TPU 3.0 AI Chip Requiring Liquid Cooling
Google announces its third-generation Tensor Processing Unit at the 2018 I/O conference, delivering a massive 100 petaflops of performance. The new chip represents a major leap in artificial intelligence processing power and introduces liquid cooling to Google's hardware stack.
Google reveals its third-generation Tensor Processing Unit (TPU 3.0) at the 2018 I/O developer conference, showcasing a massive leap in artificial intelligence hardware. The new chip delivers an impressive 100 petaflops of computing power, which is more than eight times the performance of the previous generation. This significant upgrade allows researchers and developers to train complex machine learning models much faster than before.
To manage the extreme heat generated by this unprecedented level of performance, Google introduces liquid cooling to its data centers for the first time. The company states that traditional air cooling is no longer sufficient for these high-density AI workloads. This shift highlights the growing physical challenges associated with pushing the boundaries of computational speed and efficiency in modern server farms.
The announcement emphasizes Google's strategy of integrating artificial intelligence into virtually every product and service it offers. By designing its own custom silicon specifically for machine learning tasks, Google bypasses traditional chipmakers to optimize its cloud infrastructure. The rollout of TPU 3.0 through Google Cloud ensures that external developers also gain access to this cutting-edge processing capability.