Product Introduction

What is a TPU Bare Metal Instance?

TPU bare metal instances attach TPU accelerator cards to ordinary cloud servers. While providing super computing power, they also possess the characteristics of flexible provisioning and on-demand use of ordinary cloud servers, suitable for scenarios such as AI training, AI inference, scientific computing, and video rendering.

TPU bare metal instances provide the same usage and management methods as ordinary cloud servers, such as combining services like private networks, security groups, key pairs, and load balancing to implement rich business architectures, and flexibly and conveniently purchase and manage TPU bare metal instances.

On the basis of ordinary bare metal servers, RDMA networks are added, providing computing specifications of 8 TPU cards and up to 114 vCPU cores, which can significantly improve network performance and increase the acceleration ratio of large-scale clusters, usable for business scenarios such as high-performance computing, artificial intelligence, and machine learning.