Purdue University

Purdue’s Rosen Center for Advanced Computing (RCAC) uses liquid cooling of the servers in its data center to save 40% of the power compared to traditional cooling methods, increasing the sustainability of the data center and reducing the carbon footprint of the university.

RCAC first began using liquid cooling in 2009 and continues to work on bringing in new, advanced cooling techniques to enhance its efficiency and capabilities. In 2020, RCAC deployed its first system using direct-to-chip liquid cooling (DLC), well in advance of its peers, enabling more research to be done for less overall investment from both the researchers and the university.

DLC allows the processors to run at a higher frequency with less variability from different wattages, and saves dramatically more power compared to traditional methods of cooling data centers, which involve air-cooled computers and then using other liquid to cool the air the way an air conditioner does.

“We’re actually bringing the water directly to the CPU and the other hot parts that are within the server that use up all the power, which is how we get the power savings of up to 40%,” explains Patrick Finnegan, lead systems engineer for RCAC.

“Because we’re putting the cooling directly where it’s needed, we can get more stability out of the systems and run the systems that we have at a higher speed so we can actually provide more computing power with the same computers because we’re doing liquid cooling,” adds Finnegan, explaining how the new cooling system not only saves power but also increases the computing capabilities available to faculty partners.

“The direct-to-chip liquid cooling allows the faculty to really see that peak performance that they’re paying for,” adds Alex Younts, principal research engineer for RCAC. “Without it, the clusters would run slower and they would do less science.”

With the ever-expanding use of AI, data center energy efficiency and sustainability is a bigger issue than ever.

“Current AI workloads are highly energy-intensive,” says Yi Ding, an assistant professor of electrical and computer engineering, who researches sustainable computing and tries to improve the energy efficiency and carbon emissions of computer systems, especially AI workloads.

“A single prompt processed by a large language model uses more than 10 times as much energy as a traditional search on a search engine, and because of the widespread adoption of AI, this energy issue has arisen and we have to improve the energy efficiency of data centers,” adds Ding.

RCAC is far ahead of the curve in terms of using liquid cooling to save power and increase data center sustainability, which is part of a broader Purdue push towards renewable energy that includes Purdue operating its own utility plant that is compatible with the use of alternative fuels, a solar power plant adjacent to campus, and wind power farms in neighboring counties.

“A lot of commercial entities are just now looking to do water-cooled data centers,” says Younts. “We began in 2009 doing what they hope to do today, and where they hope to go tomorrow is where we were in 2020.”

“In the history of computing at Purdue, we’ve been able to take leaps into new technologies and be a leader in things like liquid cooling,” adds Finnegan. “We’ve been able to use new and upcoming technologies and push the technology along to provide a much better experience to our faculty partners because of that.”

To learn more about RCAC’s computing resources, please contact rcac-help@purdue.edu or visit the RCAC home page.