Research & Software

Compute Resources & Allocations

Principal Investigator on compute allocation awards spanning four DOE leadership-class systems at NERSC and OLCF — upcoming Doudna and Perlmutter, upcoming Discovery, Frontier and Summit; obtained through competitive programs (ECP, CAAR, NESAP, ERCAP). These awards have supported real-world applications, programming-system deployment, and LLM-based software development across successive generations of GPU-accelerated (pre-)exascale architectures.

Discovery Computing System — Oak Ridge National Laboratory (ORNL) Arriving 2028

• Principal Investigator, selected as a Discovery CAAR (Center for Accelerated Application Readiness) program recipient

Doudna Compute System — Lawrence Berkeley National Laboratory (LBNL) Late 2026 onwards

• Principal Investigator via NERSC Exascale Science Applications Program (NESAP)
• Preparing scientific workflows for Doudna (NVIDIA Vera-Rubin CPU–GPU platform, deploys late 2026–early 2027)

Perlmutter Compute System — Lawrence Berkeley National Laboratory (LBNL) 2021–Current

• Principal Investigator via NERSC ERCAP
• Running plasma physics applications & exploring LLMs for scientific software development
• Hardware: NVIDIA Ampere 100 GPUs & AMD EPYC CPUs

Frontier Compute System — Oak Ridge National Laboratory (ORNL) 2022–Current

• Principal Investigator via OLCF CAAR & Exascale Computing Project (ECP)
• Preparing scientific workflows for the Frontier exascale supercomputer
• Hardware: AMD 250x GPUs & AMD EPYC CPUs

Summit (Pre-exascale) Compute System — Oak Ridge National Laboratory (ORNL) 2020–2023

• Principal Investigator via DOE Exascale Computing Project (ECP)
• Deployed programming systems on the Summit pre-exascale supercomputer
• Hardware: NVIDIA Volta 100 GPUs & IBM Power9 CPUs

NSF ACCESS Allocation Program 2022–Current

• Principal Investigator for classroom teaching & research via the NSF Advanced Cyberinfrastructure Coordination Ecosystem (ACCESS) program
• Hardware includes NVIDIA A100/H100 GPUs, AMD EPYC and Intel Xeon CPUs

Additional Hardware & Lab Systems 2020–2025

• Multiple hardware devices received from NVIDIA & AMD
• Used to build and maintain lab computing systems

Software Packages Developed

Designing and developing software is an integral part of research. The portfolio below is divided into four complementary domains that collectively advance HPC, GPU computing, and computational science: GPU/hardware characterization tools, validation & verification frameworks for parallel programming models, machine learning for biomedicine, and GPU-accelerated science applications.

GPU / Hardware Characterization

Benchmarking Characterization of Advanced GPUs — execution-centric microarchitectural characterizations of advanced GPU/accelerator hardware through systematic benchmarking, covering AMD MI300A’s FP8 matrix cores, asynchronous execution, and structured sparsity, and NVIDIA Blackwell’s memory hierarchy, SM pipelines, and 5th-gen tensor cores (vs. Hopper).

Validation & Verification of Programming Models

LLM4VV — explores using LLMs-as-judges to validate and verify compiler test suites for parallel programming frameworks, a process that has historically been manual.

OpenMP Validation & Verification Testsuite (SOLLVE) — functional, orphan, and cross test codes for the OpenMP offloading model (v4.0–6.1), validating NVIDIA, Cray, LLVM, Intel, and GNU compilers on Summit, Frontier, Perlmutter, and Aurora. 61 stars · 25 forks · 10 watchers

OpenACC Validation & Verification Testsuite (SOLLVE) — functional, orphan, and cross test codes for OpenACC (v1.0–3.4), validating NVIDIA, Cray, and GNU compilers on Summit, Frontier, and Perlmutter. 32 stars · 2 forks · 26 watchers

Machine Learning & Biomedical Applications

Drug Response Prediction for Cancer Treatment Optimization — with FNLCR and NCI/NIH, applied to the NCI60 dataset; delivered UNNT, a utility comparing neural-network and tree-based models (PLOS Computational Biology, 2024).

Machine Learning for Rare Disease Drug Response Classification — with Nemours Children Health System, UCSF, and Emory; RNA-seq-driven pipeline for predicting patient sensitivity to therapeutic inhibitors, applied to BeatAML (Bioinformatics Advances, 2023).

Reimagining Algorithms: GPU Refactoring for Faster Science

OpenACC-Accelerated PhysiCell — accelerates diffusion computations in this open-source agent-based biological simulator; reduced a 3-D cancer immunology run from 9.3 hours to 1.7 hours (>80% reduction). Subsequently integrated into the main PhysiCell project.

MURaM GPU/CPU Performance Portability — with Max Planck Institute for Solar System Research and NCAR, refactored the MURaM solar-atmosphere simulation code for performance-portable execution across CPUs and GPUs using OpenACC. 14 stars · 8 forks

The Artificial Scientist: In-Transit ML of Plasma Simulations — with ORNL, HZDR, HPE, and AMD; a streaming workflow performing ML directly on simulation data as it is generated, bypassing the filesystem bottleneck (IPDPS 2025, Best Paper Nomination).