The UW’s supercomputer is updated for more speed, GPU capability

By Gretchen Konrady

UW-IT continues to enhance Hyak, the University’s supercomputer, to ensure it best serves ever-growing faculty and student research needs. Upgrades include a new hardware infrastructure now underway, and recent expansions in GPU (Graphics Processing Unit) capacity.

Hyak supports a broad range of workloads — from traditional high-performance computing to data science approaches such as machine learning and other computation-heavy Big Data problems.

An infrastructure update underlies Hyak’s third-generation build, named “Klone.” The update will improve performance even while leveraging existing standard hardware, and includes a new CPU cluster with more data storage overall, faster read/write capability and a more performant (twice as fast), cost-effective and scalable inter-networking architecture among Hyak’s processors.

Other cost-saving efficiencies in the update include migrating to standardized data center components — such as racks and electrical connections — in partnership with the UW-IT Data Center Services and Operations (DCSO) team.

“The hardware refresh will give us more flexibility and allow us to be hardware- and vendor-agnostic because we’re using standard parts,” said Nam Pho, UW-IT Director of Cyberinfrastructure. “And that will make our operations and maintenance for the Data Center more agile.”

Although GPUs have been a part of Hyak since its inception, providing specialized ability to perform certain computations much faster than a traditional computer’s central processing unit, its GPU capacity has grown tenfold over the past three years to meet the growing demand.

The expansion has happened  largely through funding and a partnership with the UW’s Paul G. Allen School of Computer Science & Engineering (CSE).

“Though machine learning has historically been the biggest use case for GPUs on Hyak, researchers are beginning to see other GPU-accelerated applications, such as for molecular dynamics. Any process that relies on a lot of matrix multiplication is a candidate for GPU acceleration,” Nam said.

A new GPUs for Machine Learning page has been added to IT Connect to help guide researchers and educators toward the most appropriate place for doing their research and teaching, in Hyak, Cloud, or smaller jobs on platforms such as Google Colab, and even special-purpose, GPU-based, machine-learning laptops.

Because students are relying more on GPUs for their research, a GPU-based supercomputing cluster prototype was developed last year by the student Research Computing Club and deployed with the help of UW-IT and funding from the Student Technology Fee. Access to Hyak is available to all students through the UW Research Computing Club.

Hyak means fast in the Chinook Wawa language. In Chinook Jargon “Ikt,” meaning “first,” was the name of the first Hyak supercomputer. Hyak is currently on “Mox” for “two,” as its second-generation build. “Klone” means “three,” and is currently in an early-access testing phase ahead of full launch in April 2021.