Posted on Thursday, May 18 2017 @ 13:37 CEST by Thomas De Maesschalck
At yesterday's Google I/O event, the search giant showed off its second-generation tensor processing unit (TPU). This new model can be used for both training and inference and a board with four of the new Cloud TPUs promises up to 180 teraflops for machine learning tasks. Google claims its custom chip is significantly faster than GPU-based solutions and plans to offer a Cloud platform for commercial developers.
Google said that the new ASIC handily beats GPUs on training. The company’s latest large language-translation models take a full day to train on 32 of the current top-end GPUs. The same job runs in six hours on one-eighth of a pod, presumably eight TPUs.
Google started deploying the first-generation TPU in 2015. It is used for a wide variety of the company’s cloud services, including search, translation, and Google Photos.
Full details
EE Times.
