Menu Close

Google adds TPU 8tx and 8i Max to its eighth-generation lineup, the company says

Steel-slate field with the Google G mark at upper left and two large light gray-teal accelerator packages: the left holds a rectangular die under a bright cyan sealed glow with a silver CPU lid seated on its edge; the right holds two silicon dies joined by a bright cyan bridge and flanked by HBM slabs.

Google on Monday announced two new systems in its eighth-generation TPU lineup, TPU 8tx and TPU 8i Max, in a blog post by Amin Vahdat, senior vice president and chief technologist for AI and infrastructure. Reuters reported the announcement the same day and said Bloomberg News had first reported the development earlier Monday.

Google described the pair as new configurations of eighth-generation silicon, not new chips, joining TPU 8t for large-scale training and TPU 8i for real-time reasoning, which the company introduced at Google Cloud Next in April. Vahdat wrote that this is the first year Google has introduced four TPU systems.

TPU 8tx is optimized for confidential inference and small-scale training, part of what Google calls confidential computing, and is built for regulated enterprises and governments, Google said. It uses the same silicon as TPU 8t and is paired with 5th Generation AMD EPYC CPU hosts, “keeping sensitive data and model weights protected and isolated even while in active use.” Reuters said the system “encrypts data at the hardware level” and is aimed at government agencies and companies in regulated industries that need to keep sensitive data protected while it is being processed. Google said TPU 8tx delivers up to 2x lower latency than TPU 8t with a near-zero performance penalty, and that customers can fine-tune models, distill them and train agents with reinforcement learning on the same system without moving sensitive data between clusters.

TPU 8i Max combines two TPU 8i chips in a single package, Google said, doubling compute, on-chip SRAM and high-bandwidth memory bandwidth compared with TPU 8i. The company said the system is designed for massive context windows and for the highest compute density per rack, including cases in which many agents work across a large software codebase or hours of video. Reuters said the design makes it well suited for coding assistants, mathematical reasoning and drug-discovery tasks.

Neither Google’s post nor the Reuters report listed pricing or said when the systems would be generally available.

Google said TPUs, first built for internal services, now serve nearly 100 customers through Google Cloud and, more recently, in customers’ own data centers. Reuters, citing Alphabet, said TPU sales made a meaningful contribution to results in the September quarter, when Google Cloud revenue more than doubled to $30.71 billion, as AI Tech Daily reported. Oracle has said it will offer Google TPUs on Oracle Cloud, AI Tech Daily reported.

Sources

0 0 votes
Article Rating
Subscribe
Notify of
0 Comments
Inline Feedbacks
View all comments
0
Would love your thoughts, please comment.x
()
x