36,000 AMD GPUs create the largest model of the universe

The Universe is too big and old to observe the processes occurring in it in real time. Meanwhile, only observations give true ideas about the world in which we live. The output is in the simulation. Supercomputers can recreate a model of the Universe within certain limits, but in return they require the use of considerable resources, which, fortunately, are available to scientists today.

Image source: Argonne National Laboratory

Several years of preparation and tuning of algorithms made it possible to create the largest digital model of part of the Universe in history. Work completed in November 2024. The Frontier supercomputer at Oak Ridge National Laboratory (ORNL) used its 9,000 nodes, each powered by a 3rd generation AMD EPYC processor and four AMD Instinct 250X GPU accelerators, to create a model of the expanding universe of over 31 billion Mpc3 (megaparsecs cubic).

«

Thus, it is not difficult to understand that the project called ExaSky – the largest simulation of the Universe in history – will help scientists better understand the physics and evolution of the Universe, including research into the nature of dark matter. The model allows you to quickly view the evolutionary transformations of matter from different angles and from different sides. Correlating what is observed in the real Universe with evolution in the model will help clarify theory and practice, and will also draw attention to nuances that may have escaped understanding.

It will be a year or more before we see any publications based on work with the new model of the Universe, but scientists are already offering to familiarize themselves with a fragment of the model. The video prepared for this shows only one thousandth of the entire model – the volume of space is 311,296 Mpc3 or a cube with sides 64 × 64 × 76 Mpc. This was a real challenge to Frontier’s capacity, the scientists add, but it was worth it.