He Argonne National Laboratory (ANL) Aurora is working with American Intel and HP to develop a generative artificial intelligence (AI) model they refer to as GPT, which focuses on storing scientific knowledge with more than a trillion parameters.
The capabilities of supercomputers are currently being harnessed Solving big problemsHowever, ANL aims to provide “new opportunities to advance science and engineering”.
To do this, American lab Argonne has teamed up with Intel and HP to train Aurora GPT AI. It is an AI model that is trained on the Aurora supercomputer – one of the most powerful computers currently available – under the concept of storing all possible scientific knowledge.
In turn, when AI Once ready, it will have a ‘chatbot’ interface where researchers can ask questions to access knowledge in different areas, be it biology, physics or mathematics and engineering.
Specifically, it will allow researchers to run a variety of workloads, including machine learning and data-intensive tasks, as well as “traditional simulation and modeling campaigns,” as ANL describes on its website.
With this new project, it aims to expand capabilities with exascale computing, meaning it is capable of performing at least one exaflop operation per second. Exaflop operations are capable of computing at least one trillion floating point operations per second. In this way, research time can be accelerated.
This exascale power is powered by Intel, with a GPU architecture of six Intel Max Series GPUs of 128GB each. Likewise, it combines two Intel Xeon CPU Max Series processors with 64 GB.
Currently the Aurora GBT is in the early stages of development, so, as explained, its capacity is currently limited to 256 nodes, however, it will be expanded to 10,000 nodes with the capacity available on the Aurora supercomputer in the future.
With all of this, the lab emphasized that Argonne developers are continuing to work on a wide range of activities to “prepare Aurora for science from day one.”
