NESAP for Doudna

Doudna System poster art 1025 x 685

NESAP for Doudna teams are selected to explore how new HPC technologies or novel combinations of existing ones can accelerate scientific workflows. Awarded through a competitive proposal process, these teams engage in a three-year partnership with NERSC staff and postdocs to optimize their workflows specifically for Doudna.

NESAP for Doudna teams are eligible for a NERSC project allocation of 2,000 GPU node hours in 2025 and 10,000 GPU node hours in 2026. In addition, NESAP teams receive priority for attending NERSC-led hackathons and other engagement events and are given early access to the Doudna system once it is commissioned.

NESAP strategic partners

Project Science area PI Summary
HPC4EIC Nuclear (theory) Felix Ringer, Stony Brook University
 Developing and deploying ML and generative models for physics analyses at the Electron-Ion Collider (EIC), focusing on scalability, I/O performance, and managing complex software stacks.
USQCD HEP (theory), Nuclear (theory)
 Peter Boyle, Brookhaven National Laboratory Optimizing lattice QCD Monte Carlo evaluations on Doudna, through improved communication bandwidth, mixed-precision methods, and accelerated solution of the Dirac equation.
Schrödinger’s Devs Quantum Patrick Diehl, Los Alamos National Laboratory
 Developing and evaluating a distributed quantum simulator leveraging CUDA-Q on Doudna for scalable execution of quantum circuits, and exploring performance portability with iTensors.
SciGPT AI/Applied-Math Dmitriy Morozov, Lawrence Berkeley National Laboratory Training a large autoregressive transformer model (SciGPT) on high-resolution spatio-temporal datasets, including online extreme-scale training with HPC simulations for a foundation model.
Reactantanigans Computer Science William S. Moses, University of Illinois Enhancing Oceananigans for simulating oceanic fluid dynamics by leveraging MLIR compiler technology for performance, automatic differentiation, and optimal floating-point precision.
ML4S2D: Machine Learning for Seasonal to Decadal Predictability (E3SM) Earth Science Mark Taylor, Sandia National Laboratory Optimizing the E3SM project's workflow for seasonal to decadal Earth system prediction, specifically accelerating physics-based simulations and ML training of model emulators.
MLCP Materials/Chemistry
 Tucker Carrington, Queen’s University Developing simulation methods for molecular spectroscopy and dynamics by solving the vibrational Schrödinger equation, focused on load balancing and solver performance.
Deep Underground Neutrino Experiment (DUNE) High-energy physics (experiment) Matt Kramer, Lawrence Berkeley National Laboratory Optimizing computational workflows, including rapid supernova prompt processing, Far Detector simulation/reconstruction with inference-as-a-service, and CPU/GPU co-scheduling.
HACC/OpenCosmo Cosmology Salman Habib, Argonne National Laboratory
 Optimizing CRK-HACC, an extreme-scale cosmological hydrodynamics code, by investigating mixed precision techniques and integrating AI methods to achieve a 10x throughput gain.
Beam, pLasma & Accelerator Simulation Toolkit (BLAST): WarpX/ImpactX/HiPACE++ Fusion High-energy physics, Fusion
 Axel Huebl, Lawrence Berkeley National Laboratory
NIFTEA: NERSC Integrated Fusion Tokamak Edge Analysis Fusion Robert Hager, Princeton Plasma Physics Laboratory Simulating edge physics in tokamak fusion devices by coupling and optimizing XGC, M3D-C1, and DEGAS2 (with OpenMC), incorporating AI surrogates and automating mesh generation.
DFDM Geo Sciences Barbara Romanowicz, UC Berkeley Implementing and optimizing a novel large-scale solver for elastic wave propagation in the Earth, the Distributional Finite Difference Method (DFDM), for applications in geophysics.
DIII-D CSS Fusion Sterling Smith, General Atomics Porting and optimizing time-sensitive computational workflows like CAKE and IONORB to Doudna, with new features supporting real-time experimental feedback.
HMMER-GPU Biology
 Kjiersten Fagnan, Lawrence Berkeley National Laboratory & DOE Joint Genome Institute
 Accelerating the HMMER software suite, a critical bottleneck in JGI's genome annotation workflows, by porting and optimizing its performance on GPUs.
SeparationML Materials/Chemistry Ping Yang, Los Alamos National Laboratory
 Identifying new f-element selective molecules through hypothesis-driven generative AI and high-throughput multi-fidelity simulations, integrating LLMs and HPC tools.
High-Throughput Design of Materials/Chemistry with Tailored Thermal Properties Materials/Chemistry Anubhav Jain, Lawrence Berkeley National Laboratory Scaling up high-throughput phonon calculations for materials design to 10-100 GPU nodes and overcoming memory and I/O bottlenecks for complex materials on the Doudna system.
RCSB Protein Data Bank Biology Jeremy Henry, Rutgers University & UC San Diego Scale up an ETL workflow for protein sequence, structure, and annotation data; integrate AI/ML for structural and text-based search; and transition to a Kubernetes-native environment on NERSC.
NCEM Materials/Chemistry Peter Ercius, Lawrence Berkeley National Laboratory Enhancing a cross-facility HPC-driven ecosystem for real-time electron microscopy feedback, building on previous work in streaming data, enabling more computationally intensive analysis.
Materials Intelligence Research Materials/Chemistry Chuck Witt, Harvard University Developing and applying methods for ML-accelerated materials science, optimizing iterative workflows that combine electronic structure calculations and ML interatomic potentials.
Representation Learning Biology Petrus Zwart, Lawrence Berkeley National Laboratory Scaling workflows for learning task-specific representations from diverse scientific imaging datasets, leveraging foundation models and addressing challenges across various scales.