Google today announced Ironwood, its 7th generation Tensor Processing Unit (TPU), which is a high-performance, scalable custom AI accelerator, and the first designed specifically for inference . So, what should you know about it?
For your information, Ironwood represents a major shift in AI development and the infrastructure that supports its advancement. It marks a shift from a responsive AI model, which provides real-time information for humans to interpret, to one that proactively generates deep knowledge and interpretations. This is what we call the “era of inference,” where AI agents will proactively ingest and produce data to deliver knowledge and answers, not just data.
Scheduled to launch later this year for Google Cloud customers, Ironwood will be available in two sizes based on AI workload needs - a 256-chip and 9,216-chip configuration. Each individual chip has a memory capacity of 192GB RAM (6x Trillium) with High Bandwith Memory (HBM) bandwidth reaching 7.2 TBps (4.5x Trillium) and offers a peak computing capacity of 4,614 TFLOPs.
Additionally, when scaled up to 9,216 chips per pod for a total of 42.5 Exaflops, Ironwood is said to have the capacity to support more than 24 times the computing power of the world's largest supercomputer, El Capitan, which offers just 1.7 Exaflops per pod.
Moreover, Ironwood is designed to efficiently manage the complex computing and communication requirements of "thinking models," which include Law Language Models (LLM), Mixture of Experts (MoE), and advanced thinking tasks. With Ironwood, developers can also use Google's Pathways software suite to harness the combined computing power of tens of thousands of Ironwood TPUs in a reliable and easy way.
What are your thoughts on this news? Comment below, and stay tuned for more news like this at TechNave.
COMMENTS