Google's self-developed Trillium AI accelerators are now generally available for use.
Officially launched on December 11, the accelerators are designed for artificial intelligence (AI) workloads.
Trillium is Google's sixth-generation Tensor Processing Unit (TPU). Unveiled in May 2024, Trillium was described as being Google's "most advanced TPU," capable of achieving a 4.7x increase in peak compute performance per chip compared to the TPU v5e.
According to Google, Trillium achieves this by expanding the size of the chip’s matrix multiple units (MXUs), increasing the clock speed, and comes equipped with a third-generation SparseCore – a specialized dataflow accelerator that processes large embeddings often found in ranking and recommender systems.
In addition, the TPU offers a 67 percent increase in energy efficiency over the previous generation.
The TPUs were used to train its Gemini 2.0 AI model, and are a key component of Google Cloud's AI Hypercomputer.
In addition to the general availability of Trillium, Google is also enhancing the AI Hypercomputer's open software layer including frameworks such as JAX, PyTorch, and TensorFlow.
Early customers of Trillium include AI21 Labs.
"As long-time users of TPUs since v4, we're incredibly impressed with the capabilities of Google Cloud's Trillium. The advancements in scale, speed, and cost-efficiency are significant. We believe Trillium will be essential in accelerating the development of our next generation of sophisticated language models, enabling us to deliver even more powerful and accessible AI solutions to our customers," said Barak Lenz, CTO, AI21 Labs.
