Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “CUDA”

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 109 records · Page 6
All sourcesNASA NTRSDOE OSTI
DOE OSTI2017

GPU-Centric Communication on NVIDIA GPU Clusters with InfiniBand: A Case Study with OpenSHMEM

GPUs have become an essential component for building compute clusters with high compute density and high performance per watt. As such clusters scale to have 1000s of GPUs, efficiently moving data between the GPUs becomes imperative to get maximum performance. NVSHMEM is an implementation of the OpenSHMEM standard for NVIDIA GPU clusters which allows communication to be issued from inside GPU kernels. In earlier work, we have shown how NVSHMEM can be used to achieve better application performance on GPUs connected through PCIe or NVLink. As part of this effort, we implement IB verbs for Mellanox InfiniBand adapters in CUDA. We evaluate different design alternatives, taking into consideration the relaxed memory model, automatic memory access coalescing and thread hierarchy on the GPU. We also consider correctness issues that arise in these designs. We take advantage of these designs transparently or through API extensions in NVSHMEM. With micro-benchmarks, we show that a Nvidia Pascal P100 GPU is able saturate the network bandwidth using only one or two of its 56 available streaming multiprocessors (SM). On a single GPU using a single IB EDR adapter, we achieve a throughput of around 90 million messages per second. In addition, we implement a 2dstencil application kernel using NVSHMEM and compare its performance with a CUDA-aware MPI-based implementation that uses GPUDirect RDMA. Speedups in the range of 23% to 42% are seen for input sizes large enough to fill the occupancy of Nvidia Pascal P100 GPUs on 2 to 4 nodes indicating that there are gains to be had by eliminating the CPU from the communication path when all computation runs on the GPU.

Potluri, Sreeram↗
DOE OSTI2014

A Step towards Energy Efficient Computing: Redesigning a Hydrodynamic Application on CPU-GPU, In: 2014 IEEE 28th International Parallel and Distributed Processing Symposium

Power and energy consumption are becoming an increasing concern in high performance computing. Compared to multi-core CPUs, GPUs have a much better performance per watt. In this paper we discuss efforts to redesign the most computation intensive parts of BLAST, an application that solves the equations for compressible hydrodynamics with high order finite elements, using GPUs BLAST, Dobrev. In order to exploit the hardware parallelism of GPUs and achieve high performance, we implemented custom linear algebra kernels. We intensively optimized our CUDA kernels by exploiting the memory hierarchy, which exceed the vendor's library routines substantially in performance. We proposed an auto tuning technique to adapt our CUDA kernels to the orders of the finite element method. Compared to a previous base implementation, our redesign and optimization lowered the energy consumption of the GPU in two aspects: 60% less time to solution and 10% less power required. Compared to the CPU-only solution, our GPU accelerated BLAST obtained a 2.5× overall speedup and 1.42× energy efficiency (green up) using 4th order (Q_4) finite elements, and a 1.9× speedup and 1.27× green up using 2nd order (Q2) finite elements.

Computer Science↗
DOE OSTI2014

A Framework for Lattice QCD Calculations on GPUs

Efficient calculation of the greatest common divisor (GCD) for big integers each whose number of bits is greater than or equal to 1024 has drawn a considerable amount of attention because it can be used to detect a weakness of the RSA security infrastructure. This paper presents a parallel binary GCD algorithm and its implementation for big integers on the Intel Xeon Phi coprocessor. This algorithm is capable of computing GCDs efficiently on many pairs of big integers in parallel by utilizing all cores on a Xeon Phi coprocessor as well as taking advantage of all vector processing units of the coprocessor to speed up critical integer operations within the algorithm. Using 240 threads on a Xeon Phi coprocessor to carry out GCD calculations for a large amount of 2048-bit integers, the implementation achieves the speedup of 30 times over a sequential binary GCD algorithm implementation on a single CPU core, and it delivers twice amount of performance in comparison to the same sequential binary GCD implementation running on 240 threads of the Xeon Phi.

Winter, F. T.↗
DOE OSTI2023

Understanding Performance Portability of SYCL Kernels: A Case Study with the All-Pairs Distance Calculation in Bioinformatics on GPUs

SYCL is a portable programming model. Toward the goal of a better understanding of performance portability of SYCL kernels on GPUs, we select a bioinformatics kernel for computing the all-pairs distance as a case study. After migrating the kernel from CUDA to HIP and SYCL, we evaluate the performance of the CUDA, HIP, and SYCL kernels on NVIDIA V100 and AMD MI210 GPUs. We analyze the GPU instructions from the kernels to explain performance gaps between SYCL and CUDA/HIP. We hope that the findings are valuable for improving performance portability of SYCL.

Jin, Zheming↗
DOE OSTI2023

Evaluating performance and portability of high-level programming models: Julia, Python/Numba, and Kokkos on exascale nodes

We explore the performance and portability of the high-level programming models: the LLVM-based Julia and Python/Numba, and Kokkos on high-performance computing (HPC) nodes: AMD Epyc CPUs and MI250X graphical processing units (GPUs) on Frontier’s test bed Crusher system and Ampere’s Arm-based CPUs and NVIDIA’s A100 GPUs on the Wombat system at the Oak Ridge Leadership Computing Facilities. We compare the default performance of a hand-rolled dense matrix multiplication algorithm on CPUs against vendor-compiled C/OpenMP implementations, and on each GPU against CUDA and HIP. Rather than focusing on the kernel optimization per-se, we select this naive approach to resemble exploratory work in science and as a lower-bound for performance to isolate the effect of each programming model. Julia and Kokkos perform comparably with C/OpenMP on CPUs, while Julia implementations are competitive with CUDA and HIP on GPUs. Performance gaps are identified on NVIDIA A100 GPUs for Julia’s single precision and Kokkos, and for Python/Numba in all scenarios. We also comment on half-precision support, productivity, performance portability metrics, and platform readiness. We expect to contribute to the understanding and direction for high-level, high-productivity languages in HPC as the first-generation exascale systems are deployed.

Godoy, William↗
DOE OSTI2018

Graphic Processor Realization of the Formal Solution of the Problem of Radiation Transfer in the Neutron Star Atmosphere

The realization (on the CUDA platform) of the search for the formal solution to the problem of radiation transfer in the strongly magnetized plasma in the atmosphere of a neutron star on graphic processors with the CUDA architecture. The solution is obtained using the Galerkin method with finite elements. The realization of graphic processors makes it possible to substantially accelerate computations in constructing models for the atmospheres of neutron stars.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2018

Abstractions and Directives for Adapting Wavefront Algorithms to Future Architectures

Architectures are rapidly evolving, and exascale machines are expected to offer billion-way concurrency. We need to rethink algorithms, languages and programming models among other components in order to migrate large scale applications and explore parallelism on these machines. Although directive-based programming models allow programmers to worry less about programming and more about science, expressing complex parallel patterns in these models can be a daunting task especially when the goal is to match the performance that the hardware platforms can offer. One such pattern is wavefront. This paper extensively studies a wavefront-based miniapplication for Denovo, a production code for nuclear reactor modeling. We parallelize the Koch-Baker-Alcouffe (KBA) parallel-wavefront sweep algorithm in the main kernel of Minisweep (the miniapplication) using CUDA, OpenMP and OpenACC. Our OpenACC implementation running on NVIDIA's next-generation Volta GPU boasts an 85.06x speedup over serial code, which is larger than CUDA's 83.72x speedup over the same serial implementation. Our experimental platform includes SummitDev, an ORNL representative architecture of the upcoming Summit supercomputer. Our parallelization effort across platforms also motivated us to define an abstract parallelism model that is architecture independent, with a goal of creating software abstractions that can be used by applications employing the wavefront sweep motif.

Searles, Robert↗
DOE OSTI2020

Performance Evaluation of a Two-Dimensional Flood Model on Heterogeneous High-Performance Computing Architectures

This paper describes the implementation of a two-dimensional hydrodynamic flood model with two different numerical schemes on heterogeneous high-performance computing architectures. Both schemes were able to solve the nonlinear hyperbolic shallow water equations using an explicit upwind first-order approach on finite differences and finite volumes, respectively, and were conducted using MPI and CUDA. Four different test cases were simulated on the Summit supercomputer at Oak Ridge National Laboratory. Both numerical schemes scaled up to 128 nodes (768 GPUs) with a maximum 98.2x speedup of over 1 GPU. The lowest run time for the 10 day Hurricane Harvey event simulation at 5 meter resolution (272 million grid cells) was 50 minutes. GPUDirect communication proved to be more convenient than the standard communication strategy. Both strong and weak scaling are shown.

Sharif, Md Bulbul↗
DOE OSTI2023

Evaluation of OpenAI Codex for HPC Parallel Programming Models Kernel Generation

We evaluate AI-assisted generative capabilities on fundamental numerical kernels in high-performance computing (HPC), including AXPY, GEMV, GEMM, SpMV, Jacobi Stencil, and CG. We test the generated kernel codes for a variety of language-supported programming models, including (1) C++ (e.g., OpenMP [including offload], OpenACC, Kokkos, SyCL, CUDA, and HIP), (2) Fortran (e.g., OpenMP [including offload] and OpenACC), (3) Python (e.g., numpy, Numba, cuPy, and pyCUDA), and (4) Julia (e.g., Threads, CUDA.jl, AMDGPU.jl, and KernelAbstractions.jl). We use the GitHub Copilot capabilities powered by the GPT-based OpenAI Codex available in Visual Studio Code as of April 2023 to generate a vast amount of implementations given simple + + prompt variants. To quantify and compare the results, we propose a proficiency metric around the initial 10 suggestions given for each prompt. Results suggest that the OpenAI Codex outputs for C++ correlate with the adoption and maturity of programming models. For example, OpenMP and CUDA score really high, whereas HIP is still lacking. We found that prompts from either a targeted language such as Fortran or the more general purpose Python can benefit from adding code keywords, while Julia prompts perform acceptably well for its mature programming models (e.g., Threads and CUDA.jl). We expect for these benchmarks to provide a point of reference for each programming model's community. Overall, understanding the convergence of large language models, AI, and HPC is crucial due to its rapidly evolving nature and how it is redefining human-computer interactions.

Godoy, William↗
DOE OSTI2025

Mojo: MLIR-based Performance-Portable HPC Science Kernels on GPUs for the Python Ecosystem

We explore the performance and portability of the novel Mojo language for scientific computing workloads on GPUs. As the first language based on the LLVM’s Multi-Level Intermediate Representation (MLIR) compiler infrastructure, Mojo aims to close performance and productivity gaps by combining Python’s interoperability and CUDA-like syntax for compile-time portable GPU programming. We target four scientific workloads: a seven-point stencil (memory-bound), BabelStream (memory-bound), miniBUDE (compute-bound), and Hartree–Fock (compute-bound with atomic operations); and compare their performance against vendor baselines on NVIDIA H100 and AMD MI300A GPUs. We show that Mojo’s performance is competitive with CUDA and HIP for memory-bound kernels, whereas gaps exist on AMD GPUs for atomic operations and for fast-math compute-bound kernels on both AMD and NVIDIA GPUs. Although the learning curve and programming requirements are still fairly low-level, Mojo can close significant gaps in the fragmented Python ecosystem in the convergence of scientific computing and AI.

Godoy, William [ORNL] (ORCID:0000000225905178)↗
DOE OSTI2018

StCUDA

StCUDA is a software package for VisualWorks Smalltalk that contains classes and methods that interfaces Smalltalk to CUDA API's. Supporting class StCUDA contains the low level calls to the CUDA API's

Koh, Aik-Siong↗
DOE OSTI2019

FPChecker

CUDA is the programming model used to program NVIDIA GPUs, which are widely used in DOE supercomputers. FPChecker allows programmers to detect and fix hidden exceptions in the floating-point computations that CUDA executes. When a floating-point exception occurs (e.g., division by zero), GPUs cannot catch these errors. FPChecker intercepts each floating-point operations (addition, subtraction, multiplication and division) and reports to users where exceptions occur (file and line of code).

Laguna Peralta, Ignacio↗
DOE OSTI2019

NAMSA

In NAMSA, we introduce two main solutions to address the limitations of MSA-type algorithms for simulation of electron scattering/microscopy data: just-in-time compilation and work partitioning via MPI and GPU-to-GPU communication. First, the just-in-time (JIT) compilation strategy produces optimally tuned CUDA C++ kernels due to the data layout being perfectly known at runtime and compiled via calls from the Python interpreter. NAMSA automatically optimizes the 2-D FFT planning, and CUDA multi-streams scheduling/launching of probe simulations. With NAMSA, we can simulate electron scattering/microscopy on hundreds to thousands of GPUs, while removing the major limitations of memory and bandwidth from MSA.

Laanait, Nouamane↗
DOE OSTI2026

matsim-agents v1.0

matsim-agents is a multi-agent AI framework for atomistic materials simulation and discovery. It orchestrates large language models (LLMs), machine-learned interatomic potentials (MLIPs), and DFT codes into a single agentic loop running on laptops and DOE leadership-class supercomputers. MULTI-AGENT ORCHESTRATION A LangGraph state machine with three nodes: a Planner that converts a natural-language research objective into structured tasks; an Executor that dispatches atomistic tools and loops until the queue is empty; and an Analyst that summarizes results into a human-readable report. State is checkpointed after every step and human-in-the-loop gates can be inserted at any edge. HYPOTHESIS-DRIVEN DISCOVERY CHAT An interactive REPL (matsim-agents chat) that couples LLM dialogue with atomistic simulation. Chemical formulas are automatically detected in conversation turns and trigger a full crystal-phase exploration: structure generation → relaxation → stability scoring → result injection back into the conversation, creating a closed hypothesis-refinement loop. CRYSTAL PHASE ENUMERATION Given a composition, the phase explorer enumerates prototypes by stoichiometry: elemental (fcc/bcc/hcp/sc/diamond), binary 1:1 (rocksalt/CsCl/zincblende/ wurtzite/fluorite/rutile), ternary 1:1:3 (cubic perovskite), ternary 1:2:4 (perovskite + spinel), quaternary 1:1:2:6 (Fm-3m double perovskite). 2-D prototypes (graphene, h-BN, MoS2 2H/1T) and multilayer stacking are also supported via --include-2d and --num-layers. SUPERCELL GENERATION AND SITE DECORATION Auto-tiling to a minimum atom count (--min-atoms), explicit NxNxN tiling (--supercell), symmetry-distinct site decorations (--n-orderings), and isotropic lattice-scale sweeps (--lattice-scales) for volume bracketing. MLFF RELAXATION AND STABILITY SCORING HydraGNN (multi-headed GNN) drives structure relaxation via ASE with FIRE, BFGS, or BFGSLineSearch. Stability output: delta-E/atom ranking across phases and a max-residual-force dynamical-stability proxy. Other MLIPs (MACE, NequIP, Orb) can be plugged in through the same interface. DFT BACKENDS Quantum ESPRESSO pw.x and VASP 6.6 are first-class labellers. Both have validated GPU builds and SLURM/PBS launchers for three DOE platforms: Frontier (AMD MI250X, ROCm), Aurora (Intel PVC, oneAPI), Perlmutter (NVIDIA A100, CUDA). QE produces ~100 binaries (pw.x, ph.x, epw.x, ...). VASP supports scf, relax, vc-relax, and vc-relax-shape run types. ACTIVE-LEARNING LOOP matsim-agents al run CONFIG.yaml drives an iterative HydraGNN-DFT loop: MD generates candidates → ensemble/MC-dropout uncertainty selects the most informative → DFT labels them in parallel inside one allocation → dataset grows → HydraGNN retrains → repeat. DFT backend is a single YAML toggle (dft.backend: vasp | qe). LLM-generated seed structures are supported (no curated POSCAR library needed). Config uses ${VAR}, ${VAR:-default}, ${VAR:?msg} shell-style substitution for cross-user/cross-site portability. LLM BACKENDS Ollama (local, default), vLLM (HPC multi-GPU serving), OpenAI, Anthropic, HuggingFace Transformers+Accelerate. Selected at runtime via flag or env var with no code changes. HPC PORTABILITY Same Python entry points run on Frontier (ROCm 7.2), Aurora (oneAPI), and Perlmutter (CUDA 12). DFT and ML stacks are never co-loaded in the same shell; they couple through the scheduler and filesystem. Advanced multi-node launchers (serve, discovery-chat, single-relaxation, active-learning, QE warm-start) are provided for all three platforms. CODABENCH COMPETITION BUNDLE A self-contained benchmark: 159 atomistic test structures across 11 material classes, 5 tasks (formation energy, forces, ML relaxation, AI-DFT relaxation, phase stability ranking), public/private leaderboard split (30/70), and four ready-to-run baselines: MACE-MP-0, HydraGNN, UMA, AllScAIP.

Lupo Pasini, Massimiliano [Oak Ridge National Labo↗
DOE OSTI2016

Reaching bandwidth saturation using transparent injection parallelization

Although logically available, applications may not exploit enough instantaneous communication concurrency to maximize network utilization on HPC systems. This is exacerbated in hybrid programming models that combine single program multiple data with OpenMP or CUDA. We present the design of a “multi-threaded” runtime able to transparently increase the instantaneous network concurrency and to provide near saturation bandwidth, independent of the application configuration and dynamic behavior. The runtime offloads communication requests from application level tasks to multiple communication servers. The servers use system specific performance models to attain network saturation. Our techniques alleviate the need for spatial and temporal application level message concurrency optimizations. Experimental results show improved message throughput and bandwidth by as much as 150% for 4 KB messages on InfiniBand and by as much as 120% for 4 KB messages on Cray Aries. For more complex operations such as all-to-all collectives, we observe as much as 30% speedup. This translates into 23% speedup on 12,288 cores for a NAS FT implemented using FFTW. We observe as much as 76% speedup on 1500 cores for an already optimized UPC+OpenMP geometric multigrid application using hybrid parallelism. For the geometric multigrid GPU implementation, we observe as much as 44% speedup on 512 GPUs.

Chaimov, Nicholas↗
DOE OSTI2013

In-Situ Statistical Analysis of Autotune Simulation Data using Graphical Processing Units

Developing accurate building energy simulation models to assist energy efficiency at speed and scale is one of the research goals of the Whole-Building and Community Integration group, which is a part of Building Technologies Research and Integration Center (BTRIC) at Oak Ridge National Laboratory (ORNL). The aim of the Autotune project is to speed up the automated calibration of building energy models to match measured utility or sensor data. The workflow of this project takes input parameters and runs EnergyPlus simulations on Oak Ridge Leadership Computing Facility’s (OLCF) computing resources such as Titan, the world’s second fastest supercomputer. Multiple simulations run in parallel on nodes having 16 processors each and a Graphics Processing Unit (GPU). Each node produces a 5.7 GB output file comprising 256 files from 64 simulations. Four types of output data covering monthly, daily, hourly, and 15- minute time steps for each annual simulation is produced. A total of 270TB+ of data has been produced. In this project, the simulation data is statistically analyzed in-situ using GPUs while annual simulations are being computed on the traditional processors. Titan, with its recent addition of 18,688 Compute Unified Device Architecture (CUDA) capable NVIDIA GPUs, has greatly extended its capability for massively parallel data processing. CUDA is used along with C/MPI to calculate statistical metrics such as sum, mean, variance, and standard deviation leveraging GPU acceleration. The workflow developed in this project produces statistical summaries of the data which reduces by multiple orders of magnitude the time and amount of data that needs to be stored. These statistical capabilities are anticipated to be useful for sensitivity analysis of EnergyPlus simulations

accelerators↗
DOE OSTI2017

Multi-scale calculation based on dual domain material point method combined with molecular dynamics

This dissertation combines the dual domain material point method (DDMP) with molecular dynamics (MD) in an attempt to create a multi-scale numerical method to simulate materials undergoing large deformations with high strain rates. In these types of problems, the material is often in a thermodynamically non-equilibrium state, and conventional constitutive relations are often not available. In this method, the closure quantities, such as stress, at each material point are calculated from a MD simulation of a group of atoms surrounding the material point. Rather than restricting the multi-scale simulation in a small spatial region, such as phase interfaces, or crack tips, this multi-scale method can be used to consider non-equilibrium thermodynamic e ects in a macroscopic domain. This method takes advantage that the material points only communicate with mesh nodes, not among themselves; therefore MD simulations for material points can be performed independently in parallel. First, using a one-dimensional shock problem as an example, the numerical properties of the original material point method (MPM), the generalized interpolation material point (GIMP) method, the convected particle domain interpolation (CPDI) method, and the DDMP method are investigated. Among these methods, only the DDMP method converges as the number of particles increases, but the large number of particles needed for convergence makes the method very expensive especially in our multi-scale method where we calculate stress in each material point using MD simulation. To improve DDMP, the sub-point method is introduced in this dissertation, which provides high quality numerical solutions with a very small number of particles. The multi-scale method based on DDMP with sub-points is successfully implemented for a one dimensional problem of shock wave propagation in a cerium crystal. The MD simulation to calculate stress in each material point is performed in GPU using CUDA to accelerate the computation. The numerical properties of the multiscale method are investigated as well as the results from this multi-scale calculation are compared of particles needed for convergence makes the method very expensive especially in our multi-scale method where we calculate stress in each material point using MD simulation. To improve DDMP, the sub-point method is introduced in this dissertation, which provides high quality numerical solutions with a very small number of particles. The multi-scale method based on DDMP with sub-points is successfully implemented for a one dimensional problem of shock wave propagation in a cerium crystal. The MD simulation to calculate stress in each material point is performed in GPU using CUDA to accelerate the computation. The numerical properties of the multiscale method are investigated as well as the results from this multi-scale calculation are compared with direct MD simulation results to demonstrate the feasibility of the method. Also, the multi-scale method is applied for a two dimensional problem of jet formation around copper notch under a strong impact.

74 ATOMIC AND MOLECULAR PHYSICS↗
DOE OSTI2017

Milestone Completion Report WBS 1.3.5.05 ECP/VTK-m FY17Q2 [MS-17/01] Better Dynamic Types Design SDA05-1

The FY17Q2 milestone of the ECP/VTK-m project, which is the first milestone, includes the completion of design documents for the introduction of virtual methods into the VTK-m framework. Specifically, the ability from within the code of a device (e.g. GPU or Xeon Phi) to jump to a virtual method specified at run time. This change will enable us to drastically reduce the compile time and the executable code size for the VTK-m library. Our first design introduced the idea of adding virtual functions to classes that are used during algorithm execution. (Virtual methods were previously banned from the so called execution environment.) The design was straightforward. VTK-m already has the generic concepts of an “array handle” that provides a uniform interface to memory of different structures and an “array portal” that provides generic access to said memory. These array handles and portals use C++ templating to adjust them to different memory structures. This composition provides a powerful ability to adapt to data sources, but requires knowing static types. The proposed design creates a template specialization of an array portal that decorates another array handle while hiding its type. In this way we can wrap any type of static array handle and then feed it to a single compiled instance of a function. The second design focused on the mechanics of implementing virtual methods on parallel devices with a focus on CUDA. Our initial experiments on CUDA showed a very large overhead for using virtual C++ classes with virtual methods, the standard approach. Instead, we are using an alternate method provided by C that uses function pointers. With the completion of this milestone, we are able to move to the implementation of objects with virtual (like) methods. The upshot will be much faster compile times and much smaller library/executable sizes.

97 MATHEMATICS AND COMPUTING↗