Reliability and Availability of Novel Data Center Cooling Systems

8 views
Download
  • Share
Create Account or Sign In to post comments
#Reliability #Failure Modes #Testing #Electronics #Photonics #SiPho #data centers #AI #energy

(29:22 + Q&A) Prof. Patrick McCluskey, University of Maryland, College Park
From the IEEE Symposium on Reliability for Electronics and Photonics Packaging
Summary: It is often stated that artificial intelligence (AI) will have profound impacts on our lives. One often overlooked impact of AI, however, is how it will cause drastic increases in power consumption. A ChatGPT query needs nearly 10 times as much electricity to process as a Google search. As a result, Goldman Sachs Research estimates that data center power demand will grow 160% worldwide by 2030. In the United States, data centers are expected to consume 8% of US power by 2030 compared with 3% in 2022. It has been estimated that 40% of the total energy consumed by data centers is used for cooling of the servers. This has created a powerful incentive for the development and commercialization of more efficient cooling solutions for data centers.
It is critically important that the insertion of these new cooling solutions do not adversely impact the availability of the data center, which needs to exceed 99.875% for Tier 3 data centers. This means assessing and ensuring the reliability of data center cooling systems is of paramount importance when developing new approaches to cooling. This keynote address will present the challenges in developing reliable and efficient cooling technologies and methods to mitigate those concerns. It will cover physics-of-failure models and approaches for assessing the degradation of cooling system components; address Markov chain analysis for assessing system reliability, including the impact of redundancy on availability and efficiency; and discuss the current and emerging methods for ensuring Tier 3 levels of availability.
Bio: Dr. Patrick McCluskey is a Professor of Mechanical Engineering at the University of Maryland, College Park and the Department’s Director of Undergraduate Studies. He has over 25 years of research experience in the areas of thermal management, reliability, and packaging of electronic systems for use in extreme temperature environments and power applications. Dr. McCluskey has co-authored three books, 5 US Patents, and over 200 peer-reviewed technical articles with over 4000 citations. He is an associate editor of the IEEE Transactions on Components, Packaging, and Manufacturing Technology, a member of the board of governors of the IEEE Electronic Packaging Society, a fellow and member of the Executive Council of IMAPS, and a member of ASME and ASEE.

For additional talks from this REPP, or earlier ones, please visit https://attend.ieee.org/repp

(29:22 + Q&A) Prof. Patrick McCluskey, University of Maryland, College Park
From the 2024 IEEE Symposium on Reliability for Electronics and Photonics Packaging
Summary: It is often stated that artificial intelligence (AI) will have profound impacts on our lives. One often overlooked impact of AI, however, is how it will cause drastic increases in power consumption. A ChatGPT query needs nearly 10 times as much electricity to process as a Google search. As a result, Goldman Sachs Research estimates that data center power demand will grow 160% worldwide by 2030...

Speakers in this video

Advertisment

Advertisment