Google announced the sixth generation of its Tensor Processing Unit (TPU) for data centers, codenamed Trillium, at its I/O 2024 Developer Conference today. While a specific launch date wasn’t revealed, Google assured attendees that Trillium will be available later this year.
The Trillium TPU also has doubled memory bandwidth
Google CEO Sundar Pichai emphasized Google’s longstanding commitment to AI innovation, stating, “Google was born for this moment. We have been a pioneer in GPUs for more than a decade.”
Pichai then unveiled the impressive performance gains of Trillium. Compared to the fifth generation TPU, Trillium boasts a staggering 4.7 times improvement in computing power per chip. Google achieved this by enhancing the chip’s matrix multiplication unit (MXU) and boosting the overall clock speed. Additionally, Trillium benefits from doubled memory bandwidth.
Trillium incorporates Google’s third-generation SparseCore technology, described as “a purpose-built accelerator for common large-scale tasks in advanced ranking and recommendation workloads.” This allows Trillium TPUs to train models faster and deliver lower latency when serving those models.
Energy efficiency was another key focus for Google. Pichai highlighted Trillium as the company’s “most energy-efficient” TPU yet. This is particularly significant considering the ever-growing demand for AI chips, which can have a substantial environmental impact. Google claims Trillium offers a 67% improvement in energy efficiency compared to the previous generation.
RELATED:
- Google tries to run ChromeOS on Android devices to compete with Samsung DeX
- Tecno Spark 20 Pro key features revealed: Visits Google Play Console
- Xiaomi 13 Ultra Premium Camera Phone is now only $799
- Lenovo Legion Y700 2024: Latest Gaming Tablet with Enhanced Display now available at Giztop
- HyperOS Update Is Available for These Xiaomi, Redmi, and Poco Devices in India
(Via)
