All news

Google launches two new TPUs: TPU 8tx, designed with AMD for confidential computing, and TPU 8i Max with double the memory

Prepared by Sintelligent with the help of SI. Published in SI: . Updated: .

SI accelerators are beginning to specialise: instead of one universal chip, variants are appearing for protecting data while in use and for huge contexts, which makes the alternative to GPUs more practical.

Google has expanded the eighth generation of its tensor processors with two variants. TPU 8tx is built on the same silicon as TPU 8t but with hosts on fifth-generation AMD EPYC processors: it is designed for training and running models on confidential data and protects data and model weights at the hardware level while they are in use, and also allows fine-tuning small models without moving data off the device. TPU 8i Max combines two TPU 8i inference chips in a single package, doubling compute, SRAM and memory bandwidth and fitting twice as much silicon into the same rack. This is already the fourth TPU system this year, Google says, and the number of TPU customers is approaching one hundred, including frontier labs and trading firms. The specifications come from Google's claims.

History

Google unveils TPU 8tx and TPU 8i Max

2 events
  1. Google launches two new TPUs: TPU 8tx, designed with AMD for confidential computing, and TPU 8i Max with double the memoryOpen now
  2. Google unveils TPU 8tx for confidential computing and the dual-chip TPU 8i Max