Google Unveils Its Most Powerful AI Processor, Ironwood, with Up to 4.6 Quadrillion Operations Per Second

At its Cloud Next conference this week, Google unveiled a new dedicated AI chip, Ironwood. It’s the company’s seventh-generation AI processor and the first TPU optimized for inference — running trained AI models. The processor will be used in Google Cloud and will be available in two configurations: servers with 256 of these processors and clusters with 9,216 of these chips.

Image source: Google

«Ironwood is our most powerful, most productive, and most energy-efficient TPU yet. It is designed to accelerate AI model inference at scale across cloud infrastructure,” said Amin Vahdat, vice president of Google Cloud, commenting on the processor announcement.

Ironwood’s announcement comes amid growing competition in the proprietary AI accelerator space. While Nvidia dominates the market, Amazon and Microsoft are also pushing their technology. The former has developed the Trainium, Inferentia, and Graviton AI processors that power its AWS cloud infrastructure, while Microsoft uses its own Cobalt 100 AI chips in Azure cloud instances.

Google says Ironwood has a peak computing performance of 4,614 teraflops, or 4,614 trillion operations per second. So a cluster of 9,216 of these chips would offer a performance of 42.5 exaflops.

Each processor has 192GB of dedicated RAM with 7.4 Tbps of memory bandwidth. The chip also includes an advanced, specialized SparseCore core for handling the types of data common in “advanced ranking” and “recommender systems” workloads (like an algorithm that suggests clothes you might like). The TPU architecture is optimized to minimize data movement and latency, which Google claims results in significant power savings.

The company plans to use Ironwood in its modular AI Hypercomputer computing cluster as part of Google Cloud.