“This is Aurora GPT, an artificial intelligence focused on storing scientific knowledge with a capacity of over a trillion parameters” | Daily menu

he Argonne National Laboratory (ANL) American is working with Intel and HP to create a generative artificial intelligence (AI) model, which they refer to as Aurora GPT, which focuses on storing scientific knowledge with more than a trillion parameters.

The capabilities of supercomputers are currently being used for Solve big problemsHowever, ANL aims to provide “new opportunities for the development of science and engineering.”

To do this, the US Argonne laboratory is training Aurora GPT AI in cooperation with Intel and HP. It is an artificial intelligence model that is trained on the Aurora supercomputer – one of the most powerful computers currently – within the framework of the concept of storing all possible scientific knowledge.

Turn when artificial intelligence Once it is ready, it will have a “chatbot” interface where researchers will be able to ask questions to facilitate access to knowledge in different fields, be it biology, physics, mathematics and engineering.

Specifically, as ANL explains on its website, it will allow researchers to run a wider range of workloads, including machine learning and data-intensive tasks, in addition to “traditional simulation and modeling campaigns.”

With this new project, he also aims to expand capabilities using exascale computing, that is, with the ability to perform at least one exaflop operation per second. Exaflop operations have the performance that allows them to compute at least one trillion floating-point operations per second. In this way, search times can be accelerated.

This exascale powerhouse is powered by Intel, with the GPU architecture consisting of six Intel Max Series GPUs at 128GB each. Likewise, it integrates two Intel Xeon CPU Max Series processors with 64GB each.

At the moment, Aurora GPT is in the early stages of development, so, as shown, its capacity is currently limited to 256 nodes, however, in the future it will be expanded to 10,000 nodes of capacity available on the Aurora supercomputer.

Despite all this, the laboratory stressed that Argonne developers continue to work on implementing a wide range of activities “to prepare Aurora for science from day one.”