Google Launches New Generation AI Chip

Google introduces its sixth-generation TPU, Trillium, offering 4.7x higher compute capacity and 67% higher energy efficiency. Trillium TPUs are part of Google Cloud's AI Hypercomputer, revolutionizing how AI models are trained and served.

Share
Google Launches New Generation AI Chip

Generative AI is transforming how we interact with technology while opening up major efficiency gains for businesses. But these advances require ever-greater computing power, memory, and communication to train and fine-tune the most capable models and to serve them interactively to a global user base. For over a decade, Google has been developing custom hardware specifically for AI, Tensor Processing Units (TPUs), to push the boundaries of what's possible at scale and efficiency.

Trillium: The Next Generation of TPU

Google has recently announced Trillium, the sixth-generation TPU, the most powerful and energy-efficient TPU yet. Trillium TPUs achieve an impressive 4.7x increase in compute capacity per chip compared to its predecessor, TPU v5e. Memory capacity and bandwidth have doubled, as has Interchip Interconnect (ICI) bandwidth. In addition, Trillium is equipped with third-generation SparseCore, a specialized accelerator for processing the ultra-large embeddings common in advanced ranking and recommendation workloads.

More Efficient and More Sustainable

Trillium TPUs make it possible to train the next wave of foundation models faster and serve these models with reduced latency and lower cost. They're also the most sustainable TPUs yet, with over 67% higher energy efficiency compared to TPU v5e. Trillium can scale up to 256 TPUs in a single high-bandwidth, low-latency pod. With multislice technology and Titanium Intelligence Processing Units (IPUs), Trillium TPUs can scale to hundreds of pods and connect tens of thousands of chips in a building-scale supercomputer interconnected via a multi-petabit-per-second data center network.

Continued Innovation in AI

Google began work on the world's first custom-built AI accelerator, TPU v1, back in 2013, followed by the first Cloud TPU in 2017. Without TPUs, many of Google's most popular services, such as real-time voice search, photo object recognition, and interactive language translation, wouldn't be possible. Trillium TPUs will power the next wave of AI models and agents, and Google looks forward to helping its customers take advantage of these advanced capabilities.

Technical Development and Partnerships

Trillium TPUs are part of Google Cloud's AI Hypercomputer, a groundbreaking supercomputer architecture designed specifically for advanced AI workloads. It integrates performance-optimized infrastructure (including Trillium TPUs), open-source software frameworks, and flexible consumption models. Google's partnership with Hugging Face and other players underscores the shared commitment to accelerating AI development.

Customer Stories and Future Outlook

Customers like Essential AI, Nuro, Deep Genomics, and Deloitte are already using Cloud TPUs to innovate and improve their operations. Google DeepMind will use Trillium TPUs to train and serve future generations of Gemini models faster and more efficiently than ever before.

Learn More About Trillium

Google Cloud TPUs are at the forefront of AI acceleration, custom-built and optimized to enable large-scale AI models. Sixth-generation Trillium TPUs, the result of over a decade of research and innovation, will become available later this year. To learn more about Trillium TPUs and AI Hypercomputer, visit Google's website