Cloud OnAir: CETV: Introducing Cloud TPU Pod, a 11.5-petaflop supercomputer for deep learning

Опубликовано 11 декабря 2018, 23:50

Cloud TPU v2 Pod uses a custom high-speed network that connects second-generation TPUs into a single unit computing resource that delivers 11.5 petaflops of performance per pod. In this session, we will walk you through the architecture of Cloud TPU Pod, the distributed management of model parameters, its benefits in training deep neural nets, and customer examples.