USGS retires first supercomputer to make way for successor

Supercomputers, also known as high-performance computing (HPC) systems, are powerful machines that can quickly perform large-scale calculations. They enable scientists to collect and analyze data more quickly and simulate complex experiments that might not be practical to complete in a reasonable amount of time on smaller machines.

Before the USGS had supercomputers, scientists built their own supercomputers by networking various components into computing clusters, sometimes called “Frankenstein machines.” One such machine at the newly formed Science Analysis and Synthesis Group demonstrated the need to invest in larger-scale high-performance computing systems to accelerate and expand scientific discovery. This ultimately led to the purchase of the first supercomputer and the creation of the Office of Advanced Research Computing.

This HPC, called Yeti, is installed in the Bureau of Land Management data center at the Denver Federal Center. Yeti was named in honor of the contributions of its early “Frankenstein machine” predecessors.

Yeti has 3,728 central processing unit (CPU) cores and 35,072 CUDA GPU cores, as well as 32,128 GB of random access memory (RAM). To put this in perspective, a typical laptop usually has 4 cores and 8 GB of RAM.

A decade of increasing computer power

Over the years, the Advanced Research Computing team has grown to eight employees and now hosts more than 20 user trainings per year for any interested USGS scientist or researcher to come and learn more about the applications of using supercomputers as research tools.

Training is designed to help researchers gain the experience they need to start working with HPC resources as quickly as possible. To achieve this, we offer a wide range of courses, including introductory courses such as HPC 101 as well as advanced topics such as artificial intelligence (AI)/deep learning.

The staff also offers weekly office hours, giving users the opportunity to ask specific questions and get resources to solve any problems they encounter in their HPC work.

These training and technical support sessions have been well received by scientists, who rely on ARC staff to assist researchers as they use the HPC machines. Throughout Yeti’s tenure, it has been the primary machine used by staff in training sessions.

From the numbers

All in all, Yeti has helped facilitate a large number of jobs, including more than 6.2 million computational tasks. It has helped more than 1,038 users and has been used to train more than 600 users, including 75 users from federal agencies outside the USGS.

“To help illustrate the magnitude of Yeti’s five years of work, imagine that the same workload would take about 14,500 years to run on a single CPU,” said Jeff Falgout, technical lead of the ARC team.

On August 16, 2023, a ribbon-cutting ceremony was held at the Earth Resources Observation and Science Center (EROS) in Sioux Falls, South Dakota, to celebrate the installation of Yeti’s replacement and the newest USGS supercomputer. Hovenweep After the national monument on the border of Utah and Colorado, this machine is currently the most powerful computer in the United States. Ministry of the Interior. It joins the already operational supercomputers Tallgrass and Denali as the next generation of high-performance computers for the USGS.

Hovenweep is the third HPC system installed at EROS and will continue to advance the scientific understanding and management of environmental events such as wildfire risk to homes, extreme weather events (e.g., emergency flood mapping), and volcanic, landslide, and tsunami hazards.

Named after the Tallgrass Prairie National Preserve in Kansas, Tallgrass joined in 2020. It is designed for AI and analytics workloads. Tallgrass is equipped with hardware and software to meet the demands of modern AI, deep learning, and machine learning.

Denali is named after Denali National Park and Preserve in Alaska and was commissioned on January 15, 2020, at EROS in Sioux Falls, South Dakota. Denali is designed for large-scale simulations and analysis.

“It is important to note that the future of USGS HPC projects relies in part on researchers citing or acknowledging these resources in their publications to demonstrate their value to their work,” said Janice Gordon, leader of the Advanced Research Computing Team. “USGS supercomputers have become indispensable tools for researchers, and their importance as an asset to future science only seems to grow.”