Google has recently announced the general availability of its Trillium AI chip, just months after its preview release. This powerful chip offers over four times the training performance compared to previous versions and has been utilized by Google to train its advanced AI model, Gemini 2.0.

The sixth-generation TPU, Trillium, has improved features including double the HBM capacity and Interchip Interconnect bandwidth. Google reports that the chip delivers a 2.5x enhancement in training performance per dollar, making it an attractive option for enterprises in need of efficient AI infrastructure.

Trillium not only boosts training performance but also enhances energy efficiency by 67% and peak compute performance per chip by 4.7 times. It is optimized for various models, offering improved performance for tasks like image generation and language models.

Additionally, Trillium TPU serves as the core of Google Cloud’s AI Hypercomputer, featuring over 100,000 Trillium chips connected via a Jupiter network fabric. This system provides high bandwidth and integrates with popular machine learning frameworks. With Trillium now widely available, Google Cloud customers can access cutting-edge hardware for high-performance AI applications.

In related news, Google is positioning itself as the leader in delivering business AI value, putting competitors on high alert and collaborating with industry partners to launch secure virtual machines.