WEST LAFAYETTE, Ind. – Purdue University’s computing power has grown in strength and speed with its new Anvil supercomputer. The new capacity will allow Purdue to contribute to the 21st research program of the century by simultaneously feeding computational and data-driven science tools across multiple universities.
Having started operations in November, Anvil is now at full capacity, funded by a $22 million grant from the National Science Foundation. In its mission to serve as a resource for Purdue and outside institutions across the United States, it has already facilitated the work of more than 20 institutions in its first four months of operation.
“Anvil is not only the largest capacity system Purdue has ever built, it’s also the most diverse,” said Carol Song, principal researcher for ITaP Research Computing and principal researcher and project manager for Anvil. “Its various components are all integrated in one place with GPUs and large memory nodes complementing a 1,000 compute node cluster. As Anvil grows, it is also capable of supporting streams much more heterogeneous work patterns that are more common in the research taking place today at Purdue.
As evidenced by its use so far, Anvil’s next-level strength and versatility promise to be a boon to Purdue researchers who have identified myriad potential uses for its computing power.
For example, Jonathan Poggie, professor of aeronautics and astronautics, seeks to use Anvil to accelerate his study on the prediction and possible control of the aerodynamic heating produced in flight at high Mach speeds. Also, Daniel Aliaga, an associate professor of computer science, and his research group are exploring the use of Anvil’s GPU capability to help facilitate the creation of “what-if” design tools in digital urban planning. These futuristic tools will enable urban planners around the world to automatically integrate, process, analyze and visualize the complex interdependencies between urban form, function and the natural environment.
Additionally, new treatments for heart disease are the promise of a project that came to Anvil from out of state. Yinglong Miao, assistant professor in the Department of Molecular Biosciences and Center for Computational Biology at the University of Kansas, works in advanced biomolecular modeling and drug discovery. He uses the power of Anvil to speed up his biomolecular simulations to unlock cures for heart disease. “Using processors, this work could take months or even years,” Miao said. “With GPUs on Anvil, we can run these simulations much faster. Instead of needing months, we just need weeks.
Anvil’s current strength is now in a category that defies the imagination. At peak performance, it can now operate at an astonishing 5.1 petaFLOPS (floating point operations per second). A computer system with one petaFLOPS is capable of performing a quadrillion [1015] FLOPS. A petaFLOPS computing system can handle in just one second what would take an individual one calculation every second for 31,688,765 years. Multiply those individual calculations by a factor of 5.1, and the product is Anvil’s single-second peak capacity.
Anvil’s new power will dramatically increase the computing capacity available to users of NSF’s Extreme Science and Engineering Discovery Environment (XSEDE), in which Purdue has been a partner for 10 years. It will now be able to provide more than a billion hours of computation to XSEDE each year. Additionally, 10% of Anvil’s computing power will be made available for high-impact initiatives, including research partnerships with industry.
Researchers can request access to Anvil through the XSEDE attribution process. New assignments can be requested at any time by completing a request form through the XSEDE User Portal. Allocation on Anvil for classroom use and training can be claimed through the Year-Round Education Allocation process. Substantial research stipends can be requested four times a year (see calendar) and are peer-reviewed quarterly.
Detailed Anvil Details
Anvil consists of 1,000 nodes, which are individual computers consisting of one or more central processing units (CPUs) with memory. Each node features two 64-core, third-generation AMD EPYC processors and will deliver over one billion processor core hours to XSEDE each year. Anvil’s nodes are interconnected with 100 Gbps Mellanox HDR InfiniBand. The supercomputer ecosystem also includes 32 large memory nodes, each with 1TB of RAM, and 16 nodes each with four NVIDIA A100 Tensor Core GPUs (graphics processing units) delivering 1.5 petaFLOPS of single-precision performance to support supports machine learning and artificial intelligence applications. Research on Anvil will be able to take advantage of a diverse set of storage technologies, anchored by a parallel file system of more than 10 PB, reinforced by more than 3 PB of flash disk. New workflows will benefit from the block and object storage systems also supported by Anvil.
More information about Anvil is available on Purdue’s Anvil website. Anyone with questions should contact [email protected] Anvil is funded by NSF Award #2005632.
About Purdue University
Purdue University is a leading public research institution that develops practical solutions to today’s toughest challenges. Ranked in each of the past four years as one of the 10 most innovative universities in the United States by US News & World Report, Purdue delivers groundbreaking research and groundbreaking discoveries. Committed to hands-on, online, real-world learning, Purdue provides transformative education for all. Committed to affordability and accessibility, Purdue has frozen tuition and most fees at 2012-13 levels, allowing more students than ever to graduate debt-free. Learn how Purdue never stops in the persistent pursuit of the next giant leap at https://stories.purdue.edu.
Campus visitors must follow the standards set out in the Protect Purdue guidelines.
Writer: Amy Raley, [email protected]
Media contact: Brian Huchel, [email protected]
Sources: Carol Song, Principal Investigator for ITaP Research Computing
James Johnson, Senior Director of Operations, Research Computing, [email protected]
