Estelle Hutschka

Jules Verne, Frontier, Jupiter… They all have one thing in common: they’re supercomputers, capable of performing 1 billion billion floating-point operations per second. And they will soon be joined by Dojo, Elon Musk’s supercomputer, via his Tesla company. Dojo is meant to be the world’s most powerful supercomputer and will be used to train the artificial intelligence models behind the autopilot systems in Tesla vehicles.  But with commissioning scheduled for late 2024, will the billionaire’s ambition be achievable?

In this L’usine nouvelle article, Inria researcher Jean-Yves Berthou and CEA researcher Jérôme Bobin, both program directors of the NumPEx PEPR, discuss the subject. For the two researchers, the project is not impossible, but will undoubtedly involve compromises. Will Tesla be able to assemble all the hardware needed to build the supercomputer by October 2024? What will the supercomputer’s real capabilities be?

Today’s booming digital technologies require the processing and storage of vast amounts of data. For this, supercomputers are essential. The development of artificial intelligence, machine learning and quantum computing all depend on these structures. Whether it’s the rapid development of medicines and vaccines, improving the aerodynamics of aircraft and other vehicles to reduce energy consumption, or simulating a nuclear explosion for deterrence purposes, the applications are diverse. But above all, they raise political and societal stakes for the major powers: Europe, the USA and China are engaged in a frantic race for supercomputers.