Google introduced the TPU 8tx and TPU 8i Max, two configurations joining its eighth generation of Tensor Processing Units, which now serves nearly 100 customers. The TPU 8tx uses hardware-based confidential computing to isolate customer data and model parameters during processing, with Google claiming it can halve latency compared with the TPU 8t. The TPU 8i Max combines two TPU 8i chips, doubling computing power, on-chip SRAM and high-bandwidth memory throughput while fitting twice as much silicon into the same rack footprint. Google is also open-sourcing TPU Sync, software for moving AI working memory between chips, servers and storage that it says can reduce certain data-transfer delays and response startup times by more than 90%. The announcements came as competitors pressured Nvidia and its shares fell 9.3% on Thursday and another 5% on Monday amid a personal scandal involving CEO Jensen Huang.
Source: Read the original article
