Google announced that Ironwood, its seventh generation of Tensor Processing Units (TPUs), will be made generally available in the coming weeks.
The new TPU v7 delivers a 10x peak performance improvement over TPU v5, and offers 4x better performance per chip for both training and inference workloads compared to TPU v6.
With this release, Google Cloud customers will soon be able to leverage TPU v7 to accelerate their AI workloads at scale.
The chip is said to deliver a ten-fold peak performance improvement over TPU v5, along with 4x better performance per chip for both training and inference workloads compared to TPU v6.
Tensor Processing Units (TPUs) are custom-built chips designed specifically to power AI workloads. In addition to offering them to Google Cloud customers, the company also uses these chips to train and deploy its own AI model families, including Gemini, Imagen, and Veo.
Additionally, several large-scale Google Cloud customers have been leveraging TPUs to power their AI workloads.
Anthropic, the company behind the Claude family of AI models, has long utilised TPUs through Google Cloud and recently expanded its partnership with Google to deploy over 1 million new TPUs.
In India, multinational conglomerate Reliance recently announced its new venture, Reliance Intelligence, which will operate on Google Cloud infrastructure powered by TPUs.
“With Ironwood, we can scale up to 9,216 chips in a superpod connected through a breakthrough Inter-Chip Interconnect (ICI) network running at 9.6 Tb/s,” said Google in the announcement.
This configuration also provides access to 1.77 petabytes (PB) of shared High Bandwidth Memory (HBM), enabling significantly faster data processing for large-scale AI workloads.
TPUs are designed to deliver higher performance efficiency compared to GPUs. According to a Google study, TPU v4 is 1.2× to 1.7× faster than an NVIDIA A100 GPU, while consuming 1.3× to 1.9× less power.
Recently, Google also unveiled a new research initiative called Project Suncatcher, which aims to explore the feasibility of scaling AI compute in space using solar-powered satellite constellations equipped with TPUs.
According to D.A. Davidson analysts, cited by MarketWatch, the combined value of Google’s TPU business and its DeepMind AI research unit could be estimated at around $900 billion.
Industry experts suggest that if Google eventually begins offering TPUs as standalone hardware systems outside of Google Cloud, it could pose significant competition to the GPU market, challenging major players such as NVIDIA and AMD.
