As artificial intelligence adoption continues to rise in enterprise and academia, legacy data centers worldwide are seeking innovative solutions to manage high-density AI workloads within existing power budgets. The University of Pisa, a prestigious public institution and research center, is no exception. By deploying the latest technologies, the University is driving sustainability in data center infrastructure, positioning itself at the forefront of this crucial endeavor.
The University of Pisa's Liquid Cooling Journey: Pioneering Sustainable Data Centers
In 2016, the University embarked on a mission for sustainable computing by developing the Green Data Center in the Regional Park of Migliarino, approximately 8km from the center of Pisa. The Green Data Center deploys leading-edge technology, housing many servers from Dell Technologies, and high-performance processors from top chipmakers Intel, NVIDIA, AMD and more.
Recently, the Green Data Center expanded to scale the existing data center infrastructure and deploy HPC applications. This expansion divided the data center into hybrid-cooled and traditional air-cooled sections. “We are at the design limit for power and AC, so we decided to move to a new design that is hybrid. We still use precision air cooling, but we moved from 15kW to 30kW and integrated air cooling with a liquid converter to make liquid cooling available in the data center,” explains Maurizio Davini, the Chief of Technology at the University of Pisa.
Additionally, the center must navigate location-specific challenges, as Davini notes; “Our site is complicated to implement liquid-to-liquid technology because of the quality of water, and the fact that we are inside a natural park is complicated to install outside tanks or machine for water treatment.”
The expansion utilized Vertiv solutions to transform the previously air-cooled data center into a combined air and liquid-cooled facility. Despite this upgrade, the center continues to seek optimization methods that fit their unique design and location parameters.
Uncomplicating Liquid Cooling Deployments with Single-Phase, Self-Contained Liquid Cooling Systems
In collaboration with Dell Technologies and Intel, the University of Pisa established a cluster to test artificial intelligence workload performance without GPUs. After several months of testing using stock air-cooled heat sinks, they sought a solution to enhance power savings and cluster efficiency.
The solution came from JetCool’s SmartPlate System; a self-contained liquid cooling solution that requires no facility-level liquid-cooled infrastructure. This direct-to-chip, single-phase liquid cooling solution is a UL-certified, best-in-class solution that enables immediate time-to-value for customers with an average of 15% power savings. As a Dell OEM solution, SmartPlate System can be implemented in new or retrofitted on select Dell PowerEdge servers, supporting 850W in a 1U and 1,200W in a 2U.
Retrofitting Existing Servers with SmartPlate Systems
SmartPlate Systems are designed as a plug-and-play solution, making retrofitting seamless. Working with E4 Computer Engineering, JetCool’s local partner in Pisa, the University retrofitted an HPC cluster powered by Dell PowerEdge R760s. Since the solution operates at the server level, the installation process did not disrupt ongoing operations. Davini commended the swift installation, stating, “The solution is ‘liquid cooling for the rest of us [traditional data centers]’ in the sense that it’s easy to install, it’s easy to maintain, and you don’t need infrastructure.”
Davini and his team will continue to test the cluster with JetCool’s SmartPlate Systems. This collaboration has set the stage for the University to optimize its entire data center, overcoming the costly and environmental challenges typically associated with facility-level liquid cooling. In addition, this versatile solution enables deployment in both their hybrid and traditional air-cooled sections of the center.
With these advancements, the University of Pisa is not only meeting current AI demands but also paving the way for future innovations in data center efficiency and sustainability.
