Google's parent company Alphabet has announced the release of its sixth-generation tensor processing unit, Trillium.

Trillium offers a 4.7-fold increase in peak compute performance and has enhanced power efficiency, using 67% less power compared to its predecessor.

Key improvements include larger matrix multiple units, increased clock speed, a new SparseCore accelerator, and 32GB of high-bandwidth memory.

The TPU is designed to support larger AI models, more users simultaneously, and maintains efficiency in performance.

Early adopters of Trillium include Nuro, Deep Genomics, Deloitte, and Google's DeepMind for their AI model training and operations.

Trillium will be integrated with Google Cloud’s AI Hypercomputer project and is set to launch later this year, though pricing and public availability details remain undisclosed.