Lenovo and Intel provide technology for Harvard University's first liquid-cooled supercomputer

Lenovo and Intel are delivering advanced supercomputing infrastructure to enable discoveries in earthquake forecasting, disease spread, and star formation.
The newest supercomputer cluster of Harvard University's Faculty of Arts and Sciences Research Computing (FASRC) runs 3–4 times faster thanks to an upgrade of Lenovo ThinkSystem SD650 NeXtScale servers using Lenovo Neptune™ liquid cooling technology and 2nd generation Intel Xeon Platinum 8268 processors.

Lenovo announces the formation of the Exascale Technology Visionaries Council, intended to work toward democratizing exascale technology among organizations of all sizes

Extending the reach and capabilities of the most advanced high-performance computing (HPC) technologies — From Exascale to Everyscale™ — is a fundamental element of Lenovo's efforts toward a more open, knowledge-based and eco-friendly digital society in line with the Smarter Technology for All™ motto. For researchers at Harvard University's Faculty of Arts and Sciences Research Computing (FASRC), smarter means energy-efficient technology that cools servers without affecting the planet's temperature rise.

Harvard bets on liquid cooling

FASRC was established in 2007 to provide state-of-the-art computing services for advanced research. Recently, FASRC announced the creation of its largest HPC cluster — Cannon, named after the legendary American astronomer Annie Jump Cannon. The FASRC Cannon cluster is a large-scale HPC system enabling modeling and simulation in the sciences and social sciences, engineering, and the fields of public health and education for more than 600 laboratory groups and over 4,500 Harvard researchers. Faster and more efficient data processing is crucial for thousands of scientists who are improving aftershock forecasting after earthquakes, modeling black holes using data from the Event Horizon Telescope program, mapping invisible ocean pollution, developing new methods for tracking and predicting influenza, and developing a new statistical analysis technique to better understand detailed aspects of star formation.
Seeking to upgrade its previous Odyssey cluster, FASRC leveraged the long-standing collaboration between Lenovo and Intel on HPC and AI development in data centers. FASRC wanted to maintain a high number of processors and increase their individual performance, knowing that 25 percent of all computations take place on a single core. Liquid cooling is essential to ensure higher performance today and greater scalability for the future.

Cannon comprises more than 30,000 processor cores of 2nd generation Intel Xeon Scalable processors and includes Lenovo Neptune™ liquid cooling technology, which uses water's better heat conduction compared with air. As a result, critical server components can operate at lower temperatures, which promotes better performance and energy savings. The significantly better performance achieved by this new system is the result of Lenovo's efforts to democratize exascale solutions among a broad range of users in line with the From Exascale to Everyscale™ motto.

Although Cannon's storage system is distributed across various locations, the main computing resources are located at the Massachusetts Green High Performance Computing Center — a LEED Platinum-certified data center in Holyoke, Massachusetts. The Cannon cluster runs 670 servers — Lenovo ThinkSystem SD650 equipped with Lenovo Neptune™ direct water cooling and Intel Xeon Platinum 8268 processors with 24 cores per socket and 48 cores per node. Each node of the Cannon cluster is now several times faster than any previous node, and tasks such as geophysical models of the Earth are performed 3–4 times faster than in the previous system. During the first four weeks of operation, Cannon completed more than 4.2 million jobs using more than 21 million processor-hours.

"In science, iteration and reproducibility matter. However, iteration is a luxury not always available in university research, because you often have to race against time to meet a deadline" — says Scott Yockel, Director of Research Computing at Harvard University's Faculty of Arts and Sciences. "Thanks to greater computing performance and faster data processing by the Cannon cluster, our researchers can now try new approaches in data experiments without worrying about the time consequences of a possible failure. The option to fail makes our researchers more competitive."

Lenovo and Intel support Harvard: Cannon technology

Additional cores and higher system performance are also attracting researchers from other university departments, including Psychology and Public Health, who can more frequently leverage machine learning capabilities to make better discoveries faster.

Lenovo forms the Exascale Technology Visionaries Council

Project Everyscale for exascale

Intel, Lenovo and several other leading HPC industry players are forming a visionaries council aimed at democratizing the benefits of exascale technology among users of all sizes — not limiting it to top-tier government and university installations. As part of efforts to democratize exascale technology across the HPC community, the council, named Project Everyscale, will address component technologies enabling exascale computers to operate. The council's areas of interest will include all aspects of HPC system design, from alternative cooling technologies to performance, density, racks, storage, HPC and AI convergence, and other issues. The visionaries will share their thoughts as customers on the council's forum to set the direction for exascale innovation that everyone can benefit from and to build a coherent picture of the industry's future.

FASRC is a founding member of Project Everyscale along with the National Computational Infrastructure ("NCI" of Canberra, Australia), the Barcelona Supercomputing Center (Spain), the Simons Foundation's Flatiron Institute (New York), the Inter University Accelerator Centre (New Delhi, India), the Leibniz Supercomputing Centre (LRZ) of the Bavarian Academy of Sciences in Munich ("LRZ", Germany), the Potsdam Institute for Climate Impact Research (Germany), Rutgers University (New Brunswick, New Jersey, USA), Texas A&M University (College Station, Texas, USA), Tsinghua University (Beijing, China), the University of Birmingham (Birmingham, United Kingdom), the University of Chicago Research Computing Center (Chicago, Illinois, USA), and the SciNet Supercomputing Centre at the University of Toronto (Canada). The project's members are leaders in breakthrough research on the world's biggest problems in fields such as computational chemistry, geospatial analysis, climate change, astronomy, healthcare and meteorology.
"In collaboration with Intel, we are bringing together the most important players and the greatest minds in the HPC industry to develop an innovation plan that will stimulate the development and democratization of exascale technology among users of all sizes" — said Scott Tease, General Manager for HPC and AI, Lenovo Data Center Group.

"We are pleased that Intel is a key participant in this important supercomputing initiative together with Lenovo and other HPC industry leaders" — said Trish Damkroger, Vice President and General Manager of Intel's Extreme Computing Organization. "Through Project Everyscale, we want to increase the accessibility of exascale technology and offer top-class Xeon Scalable processors, accelerators, storage, networking, software and other infrastructure components to HPC customers of any scale and for any workloads."

The council is set to begin work in early 2020.

Summary

The Cannon cluster at Harvard University, built on Lenovo ThinkSystem SD650 servers with Neptune™ liquid cooling and Intel Xeon Platinum 8268 processors, performs tasks 3–4 times faster than the previous Odyssey system. During the first four weeks of operation, more than 4.2 million jobs were completed, using more than 21 million processor-hours. The infrastructure serves more than 600 laboratory groups and 4,500 researchers.

At the same time, Lenovo and Intel established Project Everyscale — an exascale technology visionaries council bringing together the world's leading HPC centers. Its goal is to democratize exascale solutions beyond the largest government and university installations, among organizations of all sizes.