USGS Pathfinder Supercomputer

By Advanced Research Computing (ARC)

Pathfinder is the U.S. Geological Survey's newest supercomputer or high-performance computing (HPC) system, built to accelerate scientific discovery at unprecedented scale.

Designed to support advanced modeling, simulation, artificial intelligence, machine learning, and large-scale data analysis, Pathfinder enables researchers to tackle complex Earth science challenges that exceed the capabilities

Pathfinder is online and ready to accelerate your science.

Compute

352 compute nodes

45,056 AMD EPYC Zen 4 cores

5 GPU nodes, 20 NVIDIA H100

2.25 Pflop/s measured cpu performance, 85.8% of theoretical peak

863.6 Tflop/s FP64 Total GPU performance

Up to 4.80 Pflop/s Mixed Precision GPU performance

Storage

WekaFS, all-flash, 5.9 PB usable

21 servers, 504 NVMe drives

≈1.0 TB/s read, 0.7 TB/s write

30.7 million read IOPS total, 115k per node

Network

Two-tier InfiniBand high speed fabric

25 switches, 745 links

100 Gb/s to every node

Non-blocking Fat Tree topology

48 Tb/s full injection bandwidth

The facility

Pathfinder is located at NSF NCAR's Wyoming Supercomputing Center, which opened in 2012 and is purpose-built for supercomputing. LEED Gold — NCAR's published figures — are among the most efficient datacenters in the country. This is the first partner machine in NCAR’s facility, which is set up for the necessary liquid cooling and power needed to make Pathfinder run.

Suggested Citation

If you publish results obtained from using Pathfinder please cite the following in your publications:

MLA 9:

Falgout, J. T., J. R. Laura, R. L. Lee, K. D. Moran, and J. Gordon. USGS Advanced Research Computing, USGS Pathfinder Supercomputer. U.S. Geological Survey, https://doi.org/10.5066/P14EOBYI.

Bibtex:

@misc{falgout_usgs_pathfinder,

author = {Falgout, J. T. and Laura, J. R. and Lee, R. L. and Moran, K. D. and Gordon, J.},

title = {{USGS Advanced Research Computing, USGS Pathfinder Supercomputer}},

publisher = {U.S. Geological Survey},

doi = {10.5066/P14EOBYI},

url = {https://doi.org/10.5066/P14EOBYI}