PROJECT BACKGROUND

The primary project constraint was spatial limitation, requiring installation of the data center on a rooftop terrace. This demanded detailed structural assessment, weight optimization, and careful distribution of all infrastructure components. At the same time, the solution had to achieve reliability, redundancy, and energy efficiency levels comparable to conventional brick-and-mortar data centers-within a significantly more compact, prefabricated format. The hybrid center, for which the client coined the acronym FRIDA, was delivered as a turnkey solution-from conceptual design and engineering, through manufacturing and factory acceptance testing, to transport, on-site installation, and commissioning-in less than six months. The modular architecture ensures rapid deployment, minimal spatial footprint, high physical security, weather resilience, and cost-effective scalability for future technology and capacity upgrades. With the FRIDA data center, the faculty gained direct access to additional supercomputing capacity-an environment that ensures stable operation of high-performance AI infrastructure and enables researchers to execute complex workloads, train large-scale deep neural networks, and develop large language models such as GaMS, the first Slovenian open LLM.

 PROJECT RESULTS

To meet these requirements, NTR Engineering designed and executed a modular hybrid computing center that integrates all key components of a modern HPC and AI environment within a purpose-built, oversized container data center module. The system was engineered as a standalone, fully integrated unit comprising the IT white space, modular UPS system, electrical power distribution, technical cooling infrastructure, fire protection, physical security systems, and an integrated DCIM monitoring platform.

The interior accommodates eight air-cooled IT cabinets equipped with active rear door heat exchangers (RDHx), along with provisioned capacity for additional liquid-cooled IT cabinets with pre-installed CDU connections. This hybrid architecture ensures optimal utilization of advanced GPU accelerators, enabling operation at significantly higher power densities typical of modern AI workloads. The cooling system is based on high-efficiency adiabatic chillers combined with RDHx units configured in N+1 redundancy. The design enables energy-efficient operation across a wide range of ambient conditions and is prepared for future waste heat utilization. The solution supports sustainable, energy-efficient AI operations while maintaining high reliability and operational resilience.The entire power infrastructure is supported by a modular UPS system with lithium-ion batteries, ensuring high efficiency, rapid response times, and long-term operational stability.

»With the deployment of FRIDA, the Faculty of Computer and Information Science has gained direct access to supercomputing capabilities that enable our researchers to train significantly larger deep neural network models and accelerate results in computationally intensive algorithms. The modern, energy-efficient modular architecture enabled rapid deployment, allowing the first research workloads to run within just a few months.«

Assoc. Prof. dr. Mojca Ciglarič
Dean, UL FRI

 Modular Hybrid Data Center for Artificial Intelligence - FRIDA
(PDF | EN)

 Modularni hibridni podatkovni center za umetno inteligenco - FRIDA
(PDF | SL)

The primary project constraint was spatial limitation, requiring installation of the data center on a rooftop terrace. This demanded detailed structural assessment, weight optimization, and careful distribution of all infrastructure components. At the same time, the solution had to achieve reliability, redundancy, and energy efficiency levels comparable to conventional brick-and-mortar data centers-within a significantly more compact, prefabricated format. The hybrid center, for which the client coined the acronym FRIDA, was delivered as a turnkey solution-from conceptual design and engineering, through manufacturing and factory acceptance testing, to transport, on-site installation, and commissioning-in less than six months.

The modular architecture ensures rapid deployment, minimal spatial footprint, high physical security, weather resilience, and cost-effective scalability for future technology and capacity upgrades.

With the FRIDA data center, the faculty gained direct access to additional supercomputing capacity-an environment that ensures stable operation of high-performance AI infrastructure and enables researchers to execute complex workloads, train large-scale deep neural networks, and develop large language models such as GaMS, the first Slovenian open LLM.