- Trillium has reached general availability just months after soft launch
- Powerful AI chip delivers more than four times the training performance
- Google uses it to train Gemini 2.0, the company’s advanced AI model
Google has been developing Tensor Processing Units (TPUs), its custom AI accelerators, for more than a decade, and a few months after becoming available in preview, it has announced that its sixth-generation TPU has reached general availability and is now available. available for rent. .
Trillium doubles the capacity of HBM and the bandwidth of Interchip Interconnect, and was used to train Gemini 2.0, the tech giant’s flagship AI model.
Google reports that it offers up to a 2.5x improvement per dollar in training performance compared to previous generations of TPU, making it an attractive option for companies looking for an efficient AI infrastructure.
Google Cloud AI Hypercomputer
Trillium offers a number of other improvements over its predecessor, including four times the training performance. Energy efficiency has increased by 67%, while maximum computing performance per chip has increased by 4.7 times.
Trillium also naturally improves inference performance. Google tests indicate three times the performance for imaging models like Stable Diffusion XL and almost double the performance for large language models compared to previous generations of TPUs.
The chip is also optimized for integration-intensive models, and its third-generation SparseCore provides better performance for dynamic and data-dependent operations.
Trillium TPU also forms the basis of Google Cloud’s AI hypercomputer. This system has more than 100,000 Trillium chips connected through a Jupiter network fabric that offers 13 petabits/s of bandwidth. It integrates optimized hardware, open software, and popular machine learning frameworks including JAX, PyTorch, and TensorFlow.
With Trillium now generally available, Google Cloud customers have the opportunity to access the same hardware used to train Gemini 2.0, making high-performance AI infrastructure more accessible for a wide range of applications.
Look
