Google on Monday unveiled two new AI chip systems, the TPU 8tx and the TPU 8i Max, expanding its eighth-generation Tensor Processing Unit portfolio to four systems announced in a single year — a first for the company.
The TPU 8tx is built for secure and confidential inference, and is also suited to small-scale training. It pairs the same silicon as Google's TPU 8t training chip with fifth-generation AMD EPYC host processors, keeping data and model weights protected while they are being processed. Google says it is aimed at businesses and government agencies that handle sensitive information.
The TPU 8i Max takes the opposite approach: it packs two TPU 8i inference chips into a single package, doubling the compute, on-chip SRAM and high-bandwidth memory bandwidth of the original 8i. Google positions it for massive context windows and the highest compute density per rack, where AI models must weigh large amounts of information in a single request.
The two new systems join the TPU 8t, introduced in April for large-scale training, and the TPU 8i, introduced at the same time for real-time reasoning. Google says the 8t delivers up to 2.7 times better training performance than the previous generation, while the 8i offers up to 80 percent better inference price performance.
Google also released TPU Sync, an open-source library designed to help manage working data across chip memory, server memory and shared storage. The company did not disclose pricing or general-availability dates for either new system.
The launch lands as Google presses its challenge to Nvidia's dominance in AI accelerators. Nvidia's founder and chief executive, Jensen Huang, is under scrutiny after the company's board opened an independent investigation into his admission of an affair with a woman alleged to be a Chinese spy, and shareholders sued the company last week over its disclosures.
