The School of Laptop Science and Informatics on the College of Ljubljana (UL FRI) has launched FRIDA, a brand new high-performance computing infrastructure designed for synthetic intelligence, machine studying, massive language fashions, scientific computing and large-scale information processing.
FRIDA is constructed as a modular container information middle with hybrid cooling, combining air cooling with liquid cooling immediately on chips to deal with the excessive thermal calls for of recent AI servers. It’s an educational supercomputer platform for researchers, college students and business companions engaged on superior computing tasks.
The system is concentrated on AI workloads and is meant to assist the event and testing of latest applied sciences whereas the infrastructure will assist tasks together with the additional improvement of the Slovenian massive language mannequin GaMS and permit firms and public establishments to check AI purposes and prototype new programs as described in a press release from the College of Ljubljana.
The launch occasion was attended by representatives from academia, authorities and firms concerned in know-how improvement.
“FRIDA immediately helps the mission of FRI, one of many key carriers of technological improvement in Slovenia, because it connects prime analysis data with state-of-the-art infrastructure and permits breakthroughs within the area of UI and supercomputing,” stated Affiliate Prof. Dr. Mojca Ciglarič, Dean, UL FRI. “It affords the Slovenian financial system a platform for testing future applied sciences that we imagine symbolize a serious aggressive benefit. FRIDA helps innovation and opens up new prospects for interdisciplinary tasks by researchers with enterprise and society.”
The supercomputer’s core contains 104 GPU accelerators from seven generations, together with 64 latest-generation items related via a high-performance community. The infrastructure makes use of NVIDIA Blackwell B200 and B300 AI graphics processing items and is at present essentially the most highly effective AI-focused computing system in Slovenia.
The system enhances European high-performance computing initiatives and supplies further computing capability for analysis and business tasks. It’s also designed to scale back AI mannequin coaching instances from weeks to hours or days, relying on the workload and ship to 708 petaflops of computing efficiency for lower-precision operations, with a possible peak efficiency of as much as 1.42 exaflops.