Google introduced two new eighth-generation tensor processing unit systems on Monday, November 9, expanding its artificial intelligence hardware lineup for customers that need either stronger protection for sensitive data or more memory capacity. The TPU 8tx and TPU 8i Max join two systems announced in April, according to Google, as the company seeks a wider role in the market for computing power used to run AI.
Google said the new offerings are system configurations built around existing chips, rather than entirely new chip designs. Its TPUs began as hardware for internal services and now serve customers through Google Cloud and, according to the company, in customers’ own data centers. Google said it has nearly 100 TPU customers. Reuters reported that the expansion adds to competition with Nvidia in AI accelerators.
The TPU 8tx is designed for governments and regulated businesses that process confidential information. Google said it uses the same silicon as the TPU 8t, paired with fifth-generation AMD EPYC host processors, and protects data and model parameters while they are in use. Google claims its network design can cut latency by up to half compared with the 8t, with little performance penalty.
The TPU 8i Max targets applications that must handle large amounts of information at once, such as AI agents working across software code or lengthy video. It places two TPU 8i chips in one package. Google says the system doubles computing capacity, on-chip SRAM and high-bandwidth memory throughput compared with the 8i, allowing more processing within the same data-center rack space.
The two additions bring Google’s eighth-generation portfolio to four systems: the 8t for large-scale model training, the 8i for lower-latency reasoning, the 8tx for confidential processing and the 8i Max for large context windows. Google says it is also expanding PyTorch and other developer support to make its hardware easier for customers to use. The performance figures are company claims and were not independently verified.
Google did not set out a general availability date or prices in its announcement. Customers will need to assess whether the security, memory and software features suit their workloads as the company develops its cloud and data-center offerings. Reuters said Google’s September-quarter cloud revenue more than doubled to $30.71 billion, underscoring the commercial stakes of its AI infrastructure push.
Comments