Google unveiled two specialized systems from the eighth generation of its Tensor Processing Units, the TPU 8tx and TPU 8i Max, bringing the number of configurations available to businesses for training and running AI models to four. The TPU 8tx isolates customer data and proprietary model parameters through hardware-level confidential computing, with Google claiming it can halve latency compared with the TPU 8t while imposing almost no performance penalty. The TPU 8i Max combines two TPU 8i chips, doubling computing power, on-chip memory and high-bandwidth memory throughput to address the “memory wall” limiting autonomous agents. Google also claims its open-source TPU Sync software can cut certain data-transfer delays and response startup times by more than 90%, but it disclosed neither pricing nor availability dates, and its performance figures do not cover Nvidia’s latest products.
Source: Read the original article
