Intel’s artificial intelligence will be built on a supercomputer with 1 trillion parameters

At the ISC23 annual conference in Hamburg, Germany, Intel announced: Your Aurora supercomputer is a generative artificial intelligence model seriesof type ChatGPTbut it was developed specifically for the scientific research community by the Argonne National Laboratory.

According to Rick Stevens, deputy director of the Argonne lab, who shared the supercomputer project with Intel and Hewlett-Packard, the idea is to use the machine’s potential “to generate a resource that can be used for science with application. In collaboration with Department of Energy labs and others.”

The result of joint effort, not just one, but multiple generative AI models for research use. For this, they will be trained in general texts, codes and structured scientific data from Biology, Chemistry, Materials Science, Physics, Medicine and other sources.

What will the generative models produced on Aurora look like?

called Aurora genAI, future artificial intelligence models can reach trillion parametersare adjustable variables that determine how the model processes and responds to inputs. According to Intel, this is almost twice the models used to train Google’s PaLM and almost six times that of OpenAl’s GPT-3.

These generative models will be used in a variety of scientific applications, such as the design of molecules and materials and the synthesis of information from millions of sources for experiments in Systems Biology, Polymers and Energy Materials Chemistry, Climate Science and Cosmology. The models will also be used to accelerate the production of drugs for cancer and other diseases.

At the conference, Intel confirmed the physical delivery of more than 10,000 blades for the Aurora supercomputer. The company also reported that the complete Aurora system built using HPE Cray EX supercomputers will have 63,744 GPUs and 21,248 CPUs and 1,024 DAOS storage nodes.