Google presented Ironwood, its brand new homemade chip dedicated to artificial intelligence. And this time, it is not a simple stroke of polish on the previous model. Ironwood was designed specifically for inference, that is to say to run the AI after their training. This is the moment when the model, like Chatgpt or Gemini, answers user’s questions. It is a very demanding type of calculation, and Google has clearly decided to go loudly.
A chip to run AI models, not to train them
Ironwood is Google’s seventh generation of TPU (Tensor Processing Unit). These chips are not sold individually: they are used internally and made available to customers via Google Cloud. From a technical point of view, Ironwood combines several new features: each chip can reach 4,614 teraflops, and up to 9,216 chips can be grouped to form a giant “pod”. On this scale, the system reaches 42.5 Exaflops, which far exceeds the most powerful supercomputers of the moment.
The architecture of the chip has been redesigned to avoid bottlenecks. No more memory (192 GB per chip), more flow (7.2 to/s of bandwidth), better communication between fleas … Everything is there so that the huge IA models can run smoothly. And despite this rise in power, Ironwood is twice as economical as Trillium, the previous generation.
With Ironwood, Google is mainly aimed at companies that want to run complex AI in the cloud. The group highlights its ten -year experience in the field, and the fact that its own models, like Gemini or Alphafold, are already running on these chips.
Google also offers a software environment called Pathways, developed by Deepmind, which allows you to easily control thousands of chips in parallel. It thus becomes possible to execute very gourmet models without taking the lead with the infrastructure. Ironwood even incorporates a specialized component, called sparsecore, to accelerate certain types of calculations, for example in recommendation systems. What interest the finance, research or e-commerce sectors.
No direct sale on the horizon, therefore. But in the AI power race, Google clearly places its tokens!
🟣 To not miss any news on the Geek newspaper, subscribe to Google News and on our WhatsApp. And if you love us, .
