Infrastructure
JUPITER – Train Your AI Model on One of Europe’s Most Powerful Supercomputers
JAIF leverages the computational power of JUPITER, Europe’s first exascale supercomputer, hosted by the Jülich Supercomputing Centre (JSC) in Germany. Designed as one of the world’s most powerful system for AI training, JUPITER achieves more than one ExaFLOP per second in 64-bit precision, that is the equivalent of one quintillion (“1” followed by 18 zeros) calculations per second. A standard personal computer would take several years to carry out a task that JUPITER can perform in 1 second. This exceptional performance makes JUPITER the centerpiece of large-scale AI innovation in Europe.
JARVIS – Leverage Our High-Performance Inference Solutions
Inference is commonly understood as the ability of a trained AI model to recognise patterns and draw conclusions or make predictions. Decision-making is performed through algorithms based on large language models. These models rely on vast language datasets and are trained to statistically predict patterns in word sequences in a way that resembles natural language. Within JAIF, the consortium will provide high-performance inference solutions tailored to customer needs. To this end, it will integrate a novel cloud-based inference service, JARVIS (JUPITER Advanced Research Vehicle for Inference Services), into the Jülich supercomputing infrastructure. JARVIS can be used for generative models, including large language models for text as well as models for image and video generation. The system leverages inference time scaling techniques, enabling enhanced predictive capabilities and improved error correction through longer, simulated reasoning processes. It is designed as a highly scalable platform that will be accessible to thousands of users.
By providing access to this unique infrastructure, JAIF ensures that businesses, public institutions, and researchers can take advantage of our state-of-the-art computing power to develop next-generation AI applications.
