Google's Trillium Chips to Deliver 4.7x Performance Boost

Google announced the next generation of its Tensor Processing Units (TPU) at the Google I/O developer conference. The sixth generation chips, named Trillium, will be released later this year.

Google CEO Sundar Pichai stated, "Google has been at the forefront of GPU innovation for over a decade, and we are ready for this moment."

These new TPUs will provide a substantial 4.7x increase in compute performance per chip compared to the previous generation, according to Pichai.

Google achieved this performance boost by expanding the chip's matrix multiply units (MXUs), increasing clock speed, and doubling memory bandwidth for the Trillium chips.

Trillium also includes the third generation of SparseCore, a specialized accelerator for handling large embeddings in advanced ranking and recommendation tasks. This feature will enable Trillium TPUs to train models faster and with lower latency.

Pichai highlighted that the Trillium chips are the most energy-efficient TPUs from Google yet, addressing the growing demand for AI chips. He noted that the industry demand for machine learning compute has increased by a million-fold over the last six years, emphasizing the importance of energy efficiency. Google claims that the new TPUs are 67% more energy-efficient than the fifth-generation chips.

While Google didn't provide specific details about the new chips or pricing information for the Google Cloud, the company has announced plans to offer access to Nvidia's next-gen Blackwell processors by early 2025.

Pichai expressed Google's commitment to investing in infrastructure to power AI advancements and breaking new ground in the field.

Subscribe to the AI newsletter launching on June 5 to receive the latest updates directly to your inbox.