Yan receives NSF grant to advance multi-GPU supercomputer abilities

The grant will accelerate graph processing in supercomputers, providing software support for the fast deployment of GPU supercomputers nationwide.
Written by: Tehreem Khan
Media contact: Alicia Rohan

Stream DaYan 2RT scrDa Yan, Ph.D., assistant professor at the University of Alabama at Birmingham College of Arts and Sciences’ Department of Computer Science, received an RII Track-4 grant from the National Science Foundation to improve the efficiency of large-scale processing in Graphics Processing Unit supercomputers.

The $275,573 grant will be used to support Yan’s project, titled “Massively Parallel Graph Processing on Next-Generation Multi-GPU Supercomputers.” The impact of this project will be significant, considering that the nation is replacing Central Processing Unit supercomputers with GPU supercomputers faster than ever before to benefit from significant performance improvement and energy efficiency. But this trend creates challenges for large-scale graph processing since users must develop tailor-made GPU programs for each individual graph problem. This project can potentially address this limitation by providing a unified programming framework with optimized system support.  

CPUs act as the brain of a computer and provide instructions to other components of a computer telling them what to do, while GPUs were originally designed to accelerate the creation of images in a frame buffer intended for output to a display device. General-purpose computing on graphics processing units is the use of a GPU beyond computations for computer graphics, to perform computation in applications traditionally handled by the CPU with massive parallelism, including problems in graph theory.

“Built on my success in developing the T-thinker framework for speeding up fundamental graph operations in a CPU-rich environment, this project aims to develop a new graph-parallel framework called T-thinker GPU, which will effectively speed up graph computations in a GPU-rich environment,” Yan said. “At least three fundamental graph operations will be implemented on top of T-thinker GPU: subgraph matching, dense subgraph mining and frequent subgraph pattern mining.”

The grant will also support two collaborative visits for Yan and a Ph.D. student at the Argonne National Laboratory in summer 2023 and summer 2024. The principal investigator will develop and test these GPU tools on ANL’s exascale supercomputer, Aurora. Once developed, they will work with domain scientists in ANL to deploy and use these GPU tools in scientific applications such as bioinformatics and searching knowledge graphs. These efforts will significantly improve how the computer science curriculum is developed and taught.

“By bringing up-to-date GPU programming technology learned at ANL back to the UAB Department of Computer Science, we will contribute to parallel computing curriculum, especially on modern GPU technology,” Yan said. “The courses will be able to train more students to become GPU programming professionals who are in urgent demand not only at UAB but also across the state of Alabama.”

The impact does not stop there. According to Yan, the project will bring enough GPU experience back to collaborate with local researchers and to explore the use of the proposed T-thinker GPU framework, which was developed to address intensive big data problems and help divide a larger big data problem into smaller tasks. This framework will be used in applications such as third-generation DNA sequencing, large multi-omics network visualization and biomedical image analysis, benefiting a multitude of industries.