During the next Cloud Conference this week, Google presented the latest generation of the Accelerator TPU AI chip.
The new chip, called Ironwood, is Google’s seventh generation TPU and is the first optimized for conclusions-that is, runs AI models. It is scheduled to start some time later this year for Google Cloud customers, Ironwood will come in two configurations: a 256 chip cluster and a 9,216 chip cluster.
“Ironwood is the most powerful, capable and energy -efficient TPU yet,” writes Google Cloud VP Amin Vahdat in a blog post on TechCrunch. “And it is designed for the thought of power. AI models on a scale.”
Ironwood arrives as competition in the AI Accelerator is heated. Nvidia may have the lead, but technological giants, including Amazon and Microsoft, are pushing their own solutions at home. Amazon features Trainium, Inferentia and Graviton processors, AWS available and Microsoft hosts Azure for the Maia 100 AI chip.
Ironwood can deliver 4,614 computing power tflops to the top, according to Google’s internal comparative assessment. Each chip has 192GB of dedicated RAM with a bandwidth approaching 7.4 Tbps.
Ironwood has an enhanced specialized, sparsecore core, to process data types that are common in “advanced” and “composition” workloads (eg an algorithm that suggests costume you may like). The TPU architecture was designed to minimize data movement and delay in the chip, resulting in energy saving, Google says.
Google plans to integrate Ironwood with AI HyperComputer, a hinged computer complex in Google Cloud, in the near future, Vahdat added.
“Ironwood represents a unique revolution in the era of conclusions,” Vahdat said, “with increased computational power, memory capacity, … networking developments and reliability.”
Updated 10:45 am Pacific: A previous version of this story that is incorrectly mentioned In Microsoft’s Cobalt 100 as a AI chip. In fact, the Cobalt 100 is a general -purpose chip. Microsoft’s Maia 100 is a AI chip. We corrected the report.