Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “GPU parallelization”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 55 records · Page 3
All sourcesNASA NTRSDOE OSTI
DOE OSTI2025

Asynchronous-many-task systems: Challenges and opportunities - Scaling an AMR astrophysics code on exascale machines using Kokkos and HPX

Dynamic and adaptive mesh refinement is pivotal in high-resolution, multi-physics, multi-model simulations, necessitating precise physics resolution in localized areas across expansive domains. Today’s supercomputers’ extreme heterogeneity presents a significant challenge for dynamically adaptive codes, highlighting the importance of achieving performance portability at scale. Our research focuses on astrophysical simulations, particularly stellar mergers, to elucidate early universe dynamics. Here, we present Octo-Tiger, leveraging Kokkos, HPX, and SIMD for portable performance at scale in complex, massively parallel adaptive multi-physics simulations. Octo-Tiger supports diverse processors, accelerators, and network backends. Experiments demonstrate exceptional scalability across several heterogeneous supercomputers including Perlmutter, Frontier, and Fugaku, encompassing major GPU architectures and x86, ARM, and RISC-V CPUs. Parallel efficiency of 47.59% (110,080 cores and 6880 hybrid A100 GPUs) on a full-system run on Perlmutter (26% HPCG peak performance) and 51.37% (using 32,768 cores and 2048 MI250X) on Frontier are achieved.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

Using Apptainer in a Pilot-based Distributed Workload

GlideinWMS is a pilot and pressure-based workload manager for distributed scientific computing. Many experiments like CMS and Fermilab’s Neutrino experiments use it to provision elastic clusters for their analysis and simulations, split into close to a million concurrent jobs. Most user jobs require containers, and the pilots use Apptainer to set up the desired platform. For the pilots that run as regular batch jobs, Apptainer is safer, lighter, and easier to use than other containerization solutions. Many images used by the pilots are expanded SIF images distributed via the CernVM-FS: this combination is very efficient. At Fermilab, for example, we store on GitHub Dockerfiles that mimic the platform in the worker nodes of local clusters. GitHub workflows build and push the images to Docker Hub, and a service periodically pulls and converts them to the expanded SIF images in the CernVM-FS, so the scientists can find a familiar environment everywhere. Apptainer has also been used to run services inside the pilot jobs, like benchmarks that characterize the worker node being used, or a Triton Inference Server that allows sharing a GPU with all the jobs that run in parallel on a node.

Mambelli, Marco [Fermilab] (ORCID:0000000294892681↗
DOE OSTI2025

PCMS: Parallel Coupler For Multimodel Simulations

This paper presents the Parallel Coupler for Multimodel Simulations (PCMS), a new GPU accelerated generalized coupling framework for coupling simulation codes on leadership class supercomputers. PCMS includes distributed control and field mapping methods for up to five dimensions. For field mapping PCMS can utilize discretization and field information to accommodate physics constraints. PCMS is demonstrated with a coupling of the gyrokinetic microturbulence code XGC with a Monte Carlo neutral transport code DEGAS2 and with a 5D distribution function coupling of an energetic particle transport code (GNET) to a gyrokinetic microturbulence code (GTC). Weak scaling is also demonstrated on up to 2,080 GPUs of Frontier with a weak scaling efficiency of 85%.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2024

Efficient distributed continual learning for steering experiments in real-time

Deep learning has emerged as a powerful method for extracting valuable information from large volumes of data. However, when new training data arrives continuously (i.e., is not fully available from the beginning), incremental training suffers from catastrophic forgetting (i.e., new patterns are reinforced at the expense of previously acquired knowledge). Training from scratch each time new training data becomes available would result in extremely long training times and massive data accumulation. Rehearsal-based continual learning has shown promise for addressing the catastrophic forgetting challenge, but research to date has not addressed performance and scalability. To fill this gap, we propose an approach based on a distributed rehearsal buffer that efficiently complements data-parallel training on multiple GPUs to achieve high accuracy, short runtime, and scalability. It leverages a set of buffers (local to each GPU) and uses several asynchronous techniques for updating these local buffers in an embarrassingly parallel fashion, all while handling the communication overheads necessary to augment input minibatches using unbiased, global sampling. We further propose a generalization of rehearsal buffers to support both classification and generative learning tasks, as well as more advanced rehearsal strategies (notably Dark Experience Replay, leveraging knowledge distillation). We illustrate this approach with a real-life HPC streaming application from the domain of ptychographic image reconstruction. Furthermore, we run extensive experiments on up to 128 GPUs of the ThetaGPU supercomputer to compare our approach with baselines representative of training-from-scratch (the upper bound in terms of accuracy) and incremental training (the lower bound). Results show that rehearsal-based continual learning achieves a top-5 validation accuracy close to the upper bound, while simultaneously exhibiting a runtime close to the lower bound.

Asynchronous data management↗
DOE OSTI2025

A two-level GPU-accelerated incomplete LU preconditioner for general sparse linear systems

This paper presents a parallel preconditioning approach based on incomplete LU (ILU) factorizations in the framework of Domain Decomposition (DD) for general sparse linear systems. We focus on distributed memory parallel architectures, specifically, those that are equipped with graphic processing units (GPUs). In addition to block-Jacobi, we present general purpose two-level ILU Schur complement-based approaches, where different strategies are presented to solve the coarse-level reduced system. These strategies are combined with modified ILU methods in the construction of the coarse-level operator, in order to effectively remove smooth errors by targeting an algebraically smooth vector. We leverage available GPU-based sparse matrix kernels to accelerate the setup and the solve phases of the proposed ILU preconditioner. We evaluate the efficiency of the proposed methods as a smoother for algebraic multigrid (AMG) and as a preconditioner for Krylov subspace methods on challenging anisotropic diffusion problems and a collection of general sparse matrices.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Evolution of the SLATE linear algebra library

SLATE (Software for Linear Algebra Targeting Exascale) is a distributed, dense linear algebra library targeting both CPU-only and GPU-accelerated systems, developed over the course of the Exascale Computing Project (ECP). While it began with several documents setting out its initial design, significant design changes occurred throughout its development. In some cases, these were anticipated: an early version used a simple consistency flag that was later replaced with a full-featured consistency protocol. In other cases, performance limitations and software and hardware changes prompted a redesign. Sequential communication tasks were parallelized; host-to-host MPI calls were replaced with GPU device-to-device MPI calls; more advanced algorithms such as Communication Avoiding LU and the Random Butterfly Transform (RBT) were introduced. Early choices that turned out to be cumbersome, error prone, or inflexible have been replaced with simpler, more intuitive, or more flexible designs. Applications have been a driving force, prompting a lighter weight queue class, nonuniform tile sizes, and more flexible MPI process grids. Of paramount importance has been building a portable library that works across several different GPU architectures – AMD, Intel, and NVIDIA – while keeping a clean and maintainable codebase. Here we explore the evolving design choices and their effects, both in terms of performance and software sustainability.

Gates, Mark↗
DOE OSTI2025

A Block-Structured Adaptive Mesh Framework to Solve Radiation Transfer Equation in Irregular Embedded Geometries

Radiation transport arises in various scientific, industrial, and medical fields, and understanding its effect in applications is needed to make accurate predictions, safety assessments and performance optimizations. Solving the Radiation Transport Equation (RTE) is challenging due to its integro-differential nature, which involves both differential and integral terms. The differential term describes the change in radiation intensity due to absorption and emission, while the integral term accounts for scattering. The accurate modeling of radiation is further complicated in many applications due to the complex, irregular geometries. Various methods exist for solving the RTE, including the zonal, Monte Carlo, spherical harmonics, discrete ordinates, and finite volume methods. Traditional mesh-based approaches, which rely on structured or unstructured meshes, struggle with irregular geometries due to: a) the difficulty of conforming structured grids to irregular domains, b) challenges in enforcing boundary conditions correctly, and c) the additional computational cost of unstructured mesh methods. This work presents a second-order accurate method for solving the RTE in irregular geometries. The radiation intensity is discretized using the finite-volume method in both spatial and angular directions on regular Cartesian grid blocks. Leveraging the block-structured adaptive mesh refinement (AMR) framework provided by AMReX, our method refines the grid locally to reduce spatial discretization error, ensuring a converged numerical solution while minimizing computational costs elsewhere. A two-stage deferred correction approach is employed: First, a first-order discretization on grid blocks is solved using an algebraic multigrid method in HYPRE. Second, a correction term is applied explicitly to achieve second-order accuracy. The correction term is calculated by approximating the radiation flux on cell faces using a Total Variation Diminishing (TVD) scheme. This approach ensures quick convergence of the multigrid method while preserving higher-order accuracy of the numerical solution. Irregular geometries are resolved as embedded boundaries (EB), resulting in both cut cells and regular cells. In cut cells, we modify the fluxes using face fractions and incorporate additional contributions from EB boundary conditions. To ensure higher-order convergence near the EB interface, the correction term is modified by interpolating the radiation intensity to fictitious ghost points. The implementation takes advantage of modern supercomputers by leveraging AMReX’sMPI/X parallelization strategy where X can be MPI or a GPU accelerator including CUDA, HIP and DPC++. We validate our solver using classical test cases, both with and without EB, demonstrating accuracy and efficiency. Additionally, we analyze the impact of adaptive mesh refinement on solution accuracy and computational cost, highlighting the advantages of our approach for high-resolution radiation transport simulations.

computational fluid dynamics (CFD)↗
DOE OSTI2026

Multi-task Parallelism for Robust Pre-training of Graph Foundation Models on Multi-source, Multi-fidelity Atomistic Modeling Data

Graph foundation models using graph neural networks promise sustainable, efficient atomistic modeling. To tackle challenges of processing multi-source, multi-fidelity data during pre-training, recent studies employ multi-task learning, in which shared message passing layers initially process input atomistic structures regardless of source, then route them to multiple decoding heads that predict data-specific outputs. This approach stabilizes pre-training and enhances a model’s transferability to unexplored chemical regions. Preliminary results on approximately four million structures are encouraging, yet questions remain about generalizability to larger, more diverse datasets and scalability on supercomputers. We propose a multi-task parallelism method that distributes each head across computing resources with GPU acceleration. Implemented in the open-source HydraGNN architecture, our method was trained on over 24 million structures from five datasets and tested on the Perlmutter, Aurora, and Frontier supercomputers, demonstrating efficient scaling on all three highly heterogeneous super-computing architectures.

Lupo Pasini, Massimiliano [ORNL] (ORCID:0000000249↗
DOE OSTI2025

Real-Time Bayesian Inference at Extreme Scale: A Digital Twin for Tsunami Early Warning Applied to the Cascadia Subduction Zone

We present a Bayesian inversion-based digital twin that employs acoustic pressure data from seafloor sensors, along with 3D coupled acoustic–gravity wave equations, to infer earthquake-induced spatiotemporal seafloor motion in real time and forecast tsunami propagation toward coastlines for early warning with quantified uncertainties. Our target is the Cascadia subduction zone, with one billion parameters. Computing the posterior mean alone would require 50 years on a 512 GPU machine. Instead, exploiting the shift invariance of the parameter-to-observable map and devising novel parallel algorithms, we induce a fast offline–online decomposition. The offline component requires just one adjoint wave propagation per sensor; using MFEM, we scale this part of the computation to the full El Capitan system (43,520 GPUs) with 92% weak parallel efficiency. Moreover, given real-time data, the online component exactly solves the Bayesian inverse and forecasting problems in 0.2 seconds on a modest GPU system, a ten-billion-fold speedup.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

SPADES (Scalable Parallel Discrete Events Simulation) [SWR-24-99]

SPADES (Solver for PArallel Discrete Event Simulation) is an open-source parallel discrete event simulation (PDES) package built on the AMReX library. Targeted at solving discrete event systems in parallel, this software package aims to be performance portable and scalable on heterogeneous computing architectures, e.g., graphic processing units (GPU). SPADES implements optimistic synchronization with rollback through an implementation of the Time Warp algorithm. An alternative conservative synchronization approach is also implemented using the Lower Bound on Incoming Time Stamp. In our implementation, logical processes are represented as cells in a grid and event messages are represented as particles. SPADES supports various parallel decomposition strategies, including the use of the Message Passing Interface (MPI) and OpenMP threading. All major GPU architectures (e.g., Intel, AMD, NVIDIA) are supported through the use of performance portability functionalities implemented in AMReX. The SPADES software is released in NREL Software Record SWR-24-99 “SPADES (Scalable Parallel Discrete Events Simulation)”.

Henry de Frahan, Marc [National Renewable Energy L↗
DOE OSTI2026

Rasterization with Data-Parallel Primitives

Parallel rasterization can suffer from race conditions during fragment generation, which is traditionally addressed by using specialized hardware accessible via vendor graphics APIs. Unfortunately, graphics APIs are increasingly problematic on high-performance computers, either because they are not provided or because of concerns about dependencies with in situ visualization. In response, we present a hardware-agnostic rasterization algorithm that handles race conditions using only data-parallel primitives (DPPs), enabling efficient rendering on HPC systems without graphics API dependencies and aligning with recent efforts to deliver visualization software with DPPs. Our evaluation consists of three phases: (1) evaluating portability across different CPU and GPU architectures, (2) evaluating competitiveness with a community standard, and (3) evaluating performance across varying workloads and available parallelism. The supporting experiments run on both AMD and NVIDIA GPUs, considering data sets as large as 460 million triangles and 160 million pixels. While performance generally falls short of graphics API baselines, it achieves interactive frame rates on most workloads. As a result, we conclude our approach is a viable solution for rasterization on high-performance computers since our approach is portably performant across different architectures without the need for specialized vendor support.

Buckley, Makani [University of Oregon] (ORCID:0009↗
DOE OSTI2026

CORE-BFS: Communication-Optimized REctangular-partitioned BFS Achieving 160.845 TeraTEPS on Frontier Supercomputer

Distributed Breadth-First Search (BFS) is fundamental to many large-scale graph applications, but its performance on parallel systems is often limited by high communication overhead. This paper presents CORE-BFS, an extremely scalable GPU-based BFS implementation that introduces a unique rectangular 2D partitioning-based design for Frontier supercomputer. To further improve performance, we propose four key optimizations: (1) Rectangular 2D-partition specific data formats that use two compressed row and one compressed column status array bitmaps combined with a Double Compressed Sparse Row (DCSR) format per partition, reducing memory footprint and inter-rank traffic; (2) Adaptive frontier & communication strategy that unifies top-down and bottom-up traversal on the rectangular layout, uses lazy synchronization in top-down levels, and switches variants based on frontier size to minimize communication overhead; (3) Frontier-split degree-aware update that maps frontier vertices to thread-centric, wavefront-centric, and block-centric kernels based on their degree to improve GPU utilization and memory coalescing; (4) Row-reduction pipeline that overlaps bottom-up adjacency list processing with row-wise bitmap reduction to hide inter-rank latency. Together, these techniques increase parallelism while reducing memory and communication overhead. On the Graph500 benchmark, CORE - BFS scales up to 9,248 Frontier nodes with scale-42 graphs and reaches 160.845 TTEPS, delivering a 5.42 × speedup over our previous Frontier implementation.

Yang, Haoshen [Rutgers University]↗
DOE OSTI2025

Thermonuclear Burn in a Multiphysics Code on GPUs

Multiphysics codes links to a library called SINGE for the calculation of thermonuclear (TN) burn rates, but some current multiphysics codes do not attempt to leverage the support for parallel operation that SINGE provides. Our goal is to investigate implementations of the SINGE workflow and analyze how the use of a performance portability layer could reduce run time on CPU archi tectures while also supporting GPU architectures without requiring code modifications. We looked to the Kokkos C++ Performance Portability Ecosystem to implement hardware agnostic parallel patterns.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
DOE OSTI2024

Picasso: Memory-Efficient Graph Coloring Using Palettes With Applications in Quantum Computing

A coloring of a graph is an assignment of colors to vertices such that no two neighboring vertices have the same color. The need for memory-efficient coloring algorithms is motivated by their application in computing clique partitions of graphs arising in quantum computations where the objective is to map a large set of Pauli strings into a compact set of unitaries. We present Picasso, a randomized memory-efficient iterative parallel graph coloring algorithm with theoretical sublinear space guarantees under practical assumptions. The parameters of our algorithm provide a trade-off between coloring quality and resource consumption. To assist the user, we also propose a machine learning model to predict the coloring algorithm’s parameters considering these trade-offs. We provide a sequential and a parallel implementation of the proposed algorithm. We perform an experimental evaluation on a 64-core AMD CPU equipped with 512 GB of memory and an Nvidia A100 GPU with 40GB of memory. For a small dataset where existing coloring algorithms can be executed within the 512 GB memory budget, we show up to 68× memory savings. On massive datasets we demonstrate that GPU-accelerated Picasso can process inputs with 49.5× more Pauli strings (vertex set in our graph) and 2,478× more edges than state-of-the-art parallel approaches.

artificial intelligence, quantum computing↗
DOE OSTI2026

Benchmarking of massively parallel phase-field codes for directional solidification

We present a detailed benchmark comparing two state-of-the-art phase-field implementations for simulating alloy solidification under experimentally relevant conditions. The study investigates the directional solidification of Al-3wt%Cu under high-velocity solidification conditions and SCN-0.46wt% camphor under microgravity conditions from National Aeronautics and Space Administration (NASA) DECLIC-DSI-R experiments. Both codes, one employing finite-difference discretization with uniform mesh and GPU-acceleration (GPU-PF) and the other one employing finite-element discretization with adaptive-mesh and CPU-parallelization (PRISMS-PF), solve the same quantitative phase-field formulation that incorporates an anti-trapping current for the solidification of dilute alloys. We evaluate the predictions of each code for dendritic morphology, primary spacing, and tip dynamics in both 2D and 3D, as well as their numerical convergence and computational performance. While existing benchmark problems have primarily focused on simplified or small-scale simulations, they do not reflect the computational and modeling challenges posed by employing experimentally relevant time and length scales. Our results provide a practical framework for assessing phase-field code performance as well as validating and facilitating their application in integrated computational materials engineering (ICME) workflows that require integration with realistic experimental data.

36 MATERIALS SCIENCE↗
DOE OSTI2024

Portable HCAL reconstruction in the CMS detector using the Alpaka library

CMS has deployed a number of different GPU algorithms at the High-Level Trigger (HLT) in Run 3. As the code base for GPU algorithms continues to grow, the burden for developing and maintaining separate implementations for GPU and CPU becomes increasingly challenging. To mitigate this, CMS has adopted the Alpaka (Abstraction Library for Parallel Kernel Acceleration) library as the performance portability solution to provide a single-code base for parallel execution on both GPUs and CPUs in CMS software (CMSSW). A direct CUDA version of HCAL energy reconstruction, called Minimization At Hcal, Iteratively (MAHI), has been deployed at the HLT in the 2022-2023 data taking period. This contribution will describe how the CUDA version is converted into a portable implementation using the Alpaka library. We will discuss the porting experience from CUDA to Alpaka, the validation process and the performance of the Alpaka version in CPU and GPU.

Kwok, Martin↗
DOE OSTI2025

Geometric GNNs for charged particle tracking at GlueX

Nuclear physics experiments are aimed at uncovering the fundamental building blocks of matter. The experiments involve high-energy collisions that produce complex events with many particle trajectories. Tracking charged particles resulting from collisions in the presence of a strong magnetic field is critical to enable the reconstruction of particle trajectories and precise determination of interactions. It is traditionally achieved through combinatorial approaches that scale worse than linearly as the number of hits grows. Since particle hit data naturally form a point cloud and can be structured as graphs, graph neural networks (GNNs) emerge as an intuitive and effective choice for this task. In this study, we evaluate the GNN model for track finding on the data from the GlueX experiment at Jefferson Lab. We use simulation data to train the model and test on both simulation and real GlueX measurements. We demonstrate that GNN-based track finding outperforms the currently used traditional method at GlueX in terms of segment-based efficiency at a fixed purity while providing faster inferences. We show that the GNN model can achieve significant speedup by processing multiple events in batches, which exploits the parallel computation capability of graphical processing units (GPUs). Finally, we compare the GNN implementation on GPU and field-programmable gate array and describe the trade-off.

batched GNN pipeline↗