Google adds TPU 8tx for confidential inference and TPU 8i Max for memory-heavy work, making four new TPU systems this year

Google said on 9 November: “We've expanded our eighth generation TPU portfolio with TPU 8tx and TPU 8i Max”, which it describes as marking four new TPU systems in a single year. They join TPU 8t for training and TPU 8i for inference.

TPU 8tx “is optimized for secure and confidential inference” and is built for regulated enterprises and governments running frontier or proprietary models, Google says, with up to 2x lower latency than TPU 8t. TPU 8i Max is aimed at what Google calls the memory wall: it doubles the compute, SRAM and HBM bandwidth of TPU 8i by fitting twice the silicon into the same rack footprint.

Google says its TPUs now have nearly 100 customers, including frontier labs, trading firms, retailers and universities. The post gives no price and no availability date for either system. Google's TPUs are its own alternative to Nvidia's GPUs for training and running AI models.

• Claimed TPU 8tx and TPU 8i Max mark four new TPU systems in a single year. Google
• Claimed TPU 8tx is built for secure and confidential inference, with up to 2x lower latency than TPU 8t; TPU 8i Max doubles TPU 8i's compute, SRAM and HBM bandwidth. Google
• Claimed Google's TPUs have nearly 100 customers, including frontier labs, trading firms, retailers and universities. Google