USGS Tallgrass Supercomputer

By Advanced Research Computing (ARC)

Tallgrass is a Cray Urika-CS prototype system designed for Artificial Intelligence (AI) and analytics workloads. Tallgrass is equipped with both the hardware and software to address modern AI, deep learning and machine learning demands.

Sunset falls across beautiful Flint Hills as shadows cascade across the differential erosion terraces formed millions of years — Credit: National Park Service. (Public domain.)

Tallgrass Specifications

792 CPU cores
122,880 CUDA cores
15,360 Tensor cores
100 Gb/s EDR InfiniBand Network
Theoretical Peak Performance:
- 187.2 Tflop/s GPUs
- 3000.0 Tflop/s Tensor
- ~49 Tflop/s CPUs
- ~236.2 Tflop/s total (CPU and GPU combined)

Tallgrass was designed around data analytics, machine learning, and deep learning applications, to enable analysis of large volumes of data.

Machine Learning/Deep Learning focused machine

Architected to speed up AI/ML/DL model training and inference
Jointly funded by CSS (SAS) and OEI

Wanted to be able to use Tallgrass to answer:

How can AI/ML/DL methods be effective for science?
What specific architectures will be optimal to support deep learning and machine learning?
What will our future AI and analytics workloads look like?
Will specialized architectures for AI be needed on future machines?

Suggested Citation:

Falgout, Jeff T, Janice Gordon, Matthew J. Davis, USGS Advanced Research Computing, USGS Tallgrass Supercomputer: U.S. Geological Survey, https://doi.org/10.5066/P9XE7ROJ