Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “parallel performance”

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 1,351 records · Page 75
All sourcesNASA NTRSDOE OSTI
DOE OSTI2024

Large language model evaluation for high–performance computing software development

We apply AI-assisted large language model (LLM) capabilities of GPT-3 targeting high-performance computing (HPC) kernels for (i) code generation, and (ii) auto-parallelization of serial code in C ++, Fortran, Python and Julia. Our scope includes the following fundamental numerical kernels: AXPY, GEMV, GEMM, SpMV, Jacobi Stencil, and CG, and language/programming models: (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). Kernel implementations are generated using GitHub Copilot capabilities powered by the GPT-based OpenAI Codex available in Visual Studio Code given simple + + prompt variants. To quantify and compare the generated results, we propose a proficiency metric around the initial 10 suggestions given for each prompt. For auto-parallelization, we use ChatGPT interactively giving simple prompts as in a dialogue with another human including simple “prompt engineering” follow ups. Results suggest that correct 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 language keywords, while Julia prompts perform acceptably well for its Threads and CUDA.jl programming models. Finally, we expect to provide an initial quantifiable point of reference for code generation in each programming model using a state-of-the-art LLM. Overall, understanding the convergence of LLMs, AI, and HPC is crucial due to its rapidly evolving nature and how it is redefining human-computer interactions.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Impacts of floating-point non-associativity on reproducibility for HPC and deep learning applications

Run to run variability in parallel programs caused by floating-point non-associativity has been known to significantly affect reproducibility in iterative algorithms, due to accumulating errors. Non-reproducibility can critically affect the efficiency and effectiveness of correctness testing for stochastic programs. Recently, the sensitivity of deep learning training and inference pipelines to floating-point non-associativity has been found to sometimes be extreme. It can prevent certification for commercial applications, accurate assessment of robustness and sensitivity, and bug detection. New approaches in scientific computing applications have coupled deep learning models with high-performance computing, leading to an aggravation of debugging and testing challenges. Here we perform an investigation of the statistical properties of floating-point non-associativity within modern parallel programming models, and analyze performance and productivity impacts of replacing atomic operations with deterministic alternatives on GPUs. We examine the recently-added deterministic options in PyTorch within the context of GPU deployment for deep learning, uncovering and quantifying the impacts of input parameters triggering run to run variability and reporting on the reliability and completeness of the documentation. Finally, we evaluate the strategy of exploiting automatic determinism that could be provided by deterministic hardware, using the Groq LPUTM accelerator for inference portions of the deep learning pipeline. We demonstrate the benefits that a hardware-based strategy can provide within reproducibility and correctness efforts.

Shanmugavelu, Sanjif↗
DOE OSTI2020

Fast and Accurate Core Analysis by the Full-Immersion Pressure-Pulse Decay: Part 2—Practice and Demonstration

Summary This study continues the work of presenting a novel approach for making petrophysical assessments of tight core samples. This method, the full-immersion pressure-pulse decay (Hannon 2019), involves applying a rapid increase in pressure in a chamber surrounding the entire outer surface area of a cylindrical sample, shutting the system in, and monitoring the pressure decay in the chamber as it reaches a new equilibrium. A precursor article covered the numerical simulator designed to model flow through the sample, demonstrating its performance and accuracy in addition to providing a first-order comparison between the speed and shape of the pressure-decay responses of the full-immersion method with those of other similar transient methods. This study covers the parameter-estimation procedure and experimental verification through a proof-of-concept laboratory investigation. The investigations provided here demonstrate that under appropriate, achievable experimental conditions, the pressure data can be analyzed in such a way that returns an estimate of the porosity and apparent permeabilities both parallel and perpendicular to bedding from a single test performed on a single cylindrical sample. After determining these experimental conditions (the uniqueness window), this report outlines a data-inversion strategy to estimate the petrophysical properties (porosity, horizontal permeability, and vertical permeability) from each test. This strategy is put to the test through comparisons with measurements performed by a commercial core laboratory. A common set of samples recovered from an outcrop of a tight-gas sandstone formation were investigated using the full-immersion method, and their results are compared with those from conventional steady-state measurement procedures performed by the commercial laboratory. Comparisons between petrophysical characterizations of these samples, which had permeabilities between 25 nd and 2.3 μd, demonstrated close agreement in most cases. However, whereas steady-state measurements performed at the professional laboratory required 4 to 5 hours of testing time per measurement of a single permeability, similar assessments using the full-immersion technique, requiring approximately 5 to 10 minutes to complete, returned estimates of the horizontal and vertical permeability simultaneously. Additional analyses are provided to determine principal reasons of discrepancies in instances where agreement was not as strong. Based on lessons learned from these experiences, the report closes with suggestions on areas of improvement in the experimental approach. Once complete, these developments should propel this technology to fill a critical need to determine petrophysical properties (porosity and permeability) of tight rocks in a time-efficient manner and in a way that does not compromise their accuracy.

Energy & Fuels↗
NASA NTRS1978

The current status of super computers

In this paper, commercially available super computers are surveyed. Computer performance in general is limited by circuit speeds and physical size. Assuming the use of the fastest technology, super computers typically use parallelism in the form of either vector processing or array processing to obtain performance. The Burroughs Scientific Processor is an array computer with 16 separate processors, the Cray-1 and CDC STAR-100 are vector processors, the Goodyear Aerospace STARAN is an array processor with up to 8192 single bit processors, and the Systems Development Corporation PEPE is a collection of up to 288 separate processors.

Knight, J. C.↗
NASA NTRS1998

Synchronous Parallel Emulation and Discrete Event Simulation System with Self-Contained Simulation Objects and Active Event Objects

The present invention is embodied in a method of performing object-oriented simulation and a system having inter-connected processor nodes operating in parallel to simulate mutual interactions of a set of discrete simulation objects distributed among the nodes as a sequence of discrete events changing state variables of respective simulation objects so as to generate new event-defining messages addressed to respective ones of the nodes. The object-oriented simulation is performed at each one of the nodes by assigning passive self-contained simulation objects to each one of the nodes, responding to messages received at one node by generating corresponding active event objects having user-defined inherent capabilities and individual time stamps and corresponding to respective events affecting one of the passive self-contained simulation objects of the one node, restricting the respective passive self-contained simulation objects to only providing and receiving information from die respective active event objects, requesting information and changing variables within a passive self-contained simulation object by the active event object, and producing corresponding messages specifying events resulting therefrom by the active event objects.

Steinman, Jeffrey S.↗
NASA NTRS2000

Impact of Parallel Computing on Large Scale Aeroelastic Computations

Aeroelasticity is computationally one of the most intensive fields in aerospace engineering. Though over the last three decades the computational speed of supercomputers have substantially increased, they are still inadequate for large scale aeroelastic computations using high fidelity flow and structural equations. In addition to reaching a saturation in computational speed because of changes in economics, computer manufactures are stopping the manufacturing of mainframe type supercomputers. This has led computational aeroelasticians to face the gigantic task of finding alternate approaches for fulfilling their needs. The alternate path to over come speed and availability limitations of mainframe type supercomputers is to use parallel computers. During this decade several different architectures have evolved. In FY92 the US Government started the High Performance Computing and Communication (HPCC) program. As a participant in this program NASA developed several parallel computational tools for aeroelastic applications. This talk describes the impact of those application tools on high fidelity based multidisciplinary analysis.

Guruswamy, Guru P.↗
DOE OSTI2025

Thermal Performance of Triply Periodic Minimal Surface Lattice Structures in Single-Phase Dielectric Fluid Cooling of Power Electronics

Additive manufacturing has transformed thermal management by enabling the production of complex, optimized geometries that conventional manufacturing methods cannot achieve. This study investigates the single-phase convective heat transfer performance of gyroid triply periodic minimal surface (TPMS) lattice structures with functional porosity. TPMS structures provide high surface area to volume ratios and are amenable to 3D printing. A gyroid numerical model was created and validated against an existing experimental study with a similar feature size to the investigated geometries. The TPMS structure has a periodic width of 1.6 mm, a length of 10 mm, and a height of 4 mm, with a functional porosity ranging from 0.5 to 0.8, decreasing with distance from the heated surface. Three different flow configurations were examined for an inlet fluid temperature of 70 °C. The inlet velocities range from 0.01 to 1.2 m/s, corresponding to a Reynolds number range of 10–900 with a heat flux of 50 W/cm 2 applied at the base. AmpCool ® AC-110 dielectric fluid (Prandtl number 59.5) was used as the coolant. Thermal performance and friction characteristics were studied for the three flow orientations. The parallel flow configuration was identified as the most efficient for heat removal. A detailed analysis of the numerical results highlights the underlying physics behind the thermal performance differences among the flow configurations.

33 ADVANCED PROPULSION SYSTEMS↗
NASA NTRS1991

Transient, hypervelocity flow in an axisymmetric nozzle

The performance of an axisymmetric nozzle designed to produce uniform parallel flow with a nominal Mach number of 8 is examined. The nozzle employs a flow decoupling approach and is calibrated by measuring Pitot pressures near the exit plane for a number of operating conditions. The detailed starting process of the nozzle is simulated numerically, and a quasi-one-dimensional method is used for investigating the transient wave structure and the nozzle starting time over a number of tunnel operating conditions. An axisymmetric Navier-Stokes code is utilized in order to illustrate the multidimensional and viscous aspects of the starting process. Although the steady performance observed is found adequate, settling times are much longer than predicted by the classical quasi-one-dimensional models.

Jacobs, P. A.↗
DOE OSTI2023

On the use of a multigrid-reduction-in-time algorithm for multiscale convergence of turbulence simulations

Simulations of turbulent flow present challenges in terms of accuracy and affordability on modern highly-parallel computer architectures. A multigrid-reduction-in-time algorithm is used to provide a framework for separately evolving different scales of turbulence and for parallelizing the temporal domain, thereby increasing the concurrency. It is hypothesized that the space–time locality of the small scales of turbulence can be used to circumvent difficulties in applying temporal multigrid to flows dominated by inertial physics. For algorithms that fall well short of spectral accuracy (fourth-order is used in this work) attention must be paid to the accuracy of features on scales transferred between multigrid levels. Numerical experiments were performed using implicit large-eddy simulation. Results from applying the approach to an infinite-Reynolds number Taylor–Green flow and a double-shear flow at a Reynolds number of 11650 provide strong evidence that the approach has merit. The multigrid-reduction-in-time framework can be used to parallelize the temporal domain of a high-Reynolds-number turbulent flow and permit independent convergence of different scales. Establishing this foundation allows for future research in reducing the wall-clock time to solve turbulent flows while retaining the same accuracy as sequential solvers. In conclusion, current performance results from parallelizing the temporal domain are not competitive with those from sequential-in-time methods.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS1988

Scan line graphics generation on the massively parallel processor

Described here is how researchers implemented a scan line graphics generation algorithm on the Massively Parallel Processor (MPP). Pixels are computed in parallel and their results are applied to the Z buffer in large groups. To perform pixel value calculations, facilitate load balancing across the processors and apply the results to the Z buffer efficiently in parallel requires special virtual routing (sort computation) techniques developed by the author especially for use on single-instruction multiple-data (SIMD) architectures.

Dorband, John E.↗
DOE OSTI2025

Scale-up Unlearnable Examples Learning with High-performance Computing

Recent advancements in AI models, like ChatGPT, are structured to retain user interactions, which could inadvertently include sensitive healthcare data. In the healthcare field, particularly when radiologists use AI-driven diagnostic tools hosted on online platforms, there is a risk that medical imaging data may be repurposed for future AI training without explicit consent, spotlighting critical privacy and intellectual property concerns around healthcare data usage. Addressing these privacy challenges, a novel approach known as Unlearnable Examples (UEs) has been introduced, aiming to make data unlearnable to deep learning models. A prominent method within this area, called Unlearnable Clustering (UC), has shown improved UE performance with larger batch sizes but was previously limited by computational resources (e.g., a single workstation). To push the boundaries of UE performance with theoretically unlimited resources, we scaled up UC learning across various datasets using Distributed Data Parallel (DDP) training on the Summit supercomputer. Our goal was to examine UE efficacy at high-performance computing (HPC) levels to prevent unauthorized learning and enhance data security, particularly exploring the impact of batch size on UE’s unlearnability. Utilizing the robust computational capabilities of the Summit, extensive experiments were conducted on diverse datasets such as Pets, MedMNist, Flowers, and Flowers102. Our findings reveal that both overly large and overly small batch sizes can lead to performance instability and affect accuracy. However, the relationship between batch size and unlearnability varied across datasets, highlighting the necessity for tailored batch size strategies to achieve optimal data protection. The use of Summit’s high-performance GPUs, along with the efficiency of the DDP framework, facilitated rapid updates of model parameters and consistent training across nodes. Our results underscore the critical role of selecting appropriate batch sizes based on the specific characteristics of each dataset to prevent learning and ensure data security in deep learning applications. The source code is publicly available at https: // github. com/ hrlblab/ UE_ HPC .

Zhu, Yanfan [Vanderbilt University, Nashville, TN,↗
NASA NTRS1986

Two-dimensional nonsteady viscous flow simulation on the Navier-Stokes computer miniNode

The needs of large-scale scientific computation are outpacing the growth in performance of mainframe supercomputers. In particular, problems in fluid mechanics involving complex flow simulations require far more speed and capacity than that provided by current and proposed Class VI supercomputers. To address this concern, the Navier-Stokes Computer (NSC) was developed. The NSC is a parallel-processing machine, comprised of individual Nodes, each comparable in performance to current supercomputers. The global architecture is that of a hypercube, and a 128-Node NSC has been designed. New architectural features, such as a reconfigurable many-function ALU pipeline and a multifunction memory-ALU switch, have provided the capability to efficiently implement a wide range of algorithms. Efficient algorithms typically involve numerically intensive tasks, which often include conditional operations. These operations may be efficiently implemented on the NSC without, in general, sacrificing vector-processing speed. To illustrate the architecture, programming, and several of the capabilities of the NSC, the simulation of two-dimensional, nonsteady viscous flows on a prototype Node, called the miniNode, is presented.

Nosenchuck, Daniel M.↗
NASA NTRS2003

Quantum Search in Hilbert Space

A proposed quantum-computing algorithm would perform a search for an item of information in a database stored in a Hilbert-space memory structure. The algorithm is intended to make it possible to search relatively quickly through a large database under conditions in which available computing resources would otherwise be considered inadequate to perform such a task. The algorithm would apply, more specifically, to a relational database in which information would be stored in a set of N complex orthonormal vectors, each of N dimensions (where N can be exponentially large). Each vector would constitute one row of a unitary matrix, from which one would derive the Hamiltonian operator (and hence the evolutionary operator) of a quantum system. In other words, all the stored information would be mapped onto a unitary operator acting on a quantum state that would represent the item of information to be retrieved. Then one could exploit quantum parallelism: one could pose all search queries simultaneously by performing a quantum measurement on the system. In so doing, one would effectively solve the search problem in one computational step. One could exploit the direct- and inner-product decomposability of the unitary matrix to make the dimensionality of the memory space exponentially large by use of only linear resources. However, inasmuch as the necessary preprocessing (the mapping of the stored information into a Hilbert space) could be exponentially expensive, the proposed algorithm would likely be most beneficial in applications in which the resources available for preprocessing were much greater than those available for searching.

Zak, Michail↗
DOE OSTI2024

Performance Evaluation of Multi-Vendor Grid-Forming Inverters for Grid-Connected Operation Through Hardware Experimentation: Preprint

Existing real-world projects of GFM inverters that operate in parallel to power grids typically are sized between dozens and a few hundreds Megawatt (MW) scale according to a recent NERC GFM inverter white paper. These large systems are often difficult to evaluate prior to deployments because of their large size. The performance of smaller GFM inverters (dozens to a few hundreds MW) that operate parallel with power grids (distribution systems) is even less understood. There is an opportunity to better understand these systems through hardware testing under controlled laboratory conditions. Therefore, this paper presents the functional performance evaluation tests of multiple (three) commercial GFM inverters when they operate in parallel with the grid through hardware experiments. The goal of these tests is to explore the GFM inverters' functionalities and dynamic response when in parallel with power grids to eventually develop universal specifications for GFM inverters. Both steady state (changing the inverter's frequency and voltage droop) and transient (adding step change in grid's frequency/voltage) tests are performed for each GFM inverter with the same testing circuit and testing protocol. The experimental results indicate the bench-marked performance that: 1) the GFM inverters can be dispatched through frequency and voltage droop intercepts to output the target power when paralleled to the grid; 2) the GFM inverters automatically respond to system frequency and voltage events to output the needed power, however, the GFM inverters all show stability issues when absorbing reactive power from the grid.

grid-forming inverters↗
DOE OSTI2023

High-fidelity parallel entangling gates on a neutral-atom quantum computer

The ability to perform entangling quantum operations with low error rates in a scalable fashion is a central element of useful quantum information processing. Neutral-atom arrays have recently emerged as a promising quantum computing platform, featuring coherent control over hundreds of qubits and any-to-any gate connectivity in a flexible, dynamically reconfigurable architecture. The main outstanding challenge has been to reduce errors in entangling operations mediated through Rydberg interactions. Here we report the realization of two-qubit entangling gates with 99.5% fidelity on up to 60 atoms in parallel, surpassing the surface-code threshold for error correction. Our method uses fast, single-pulse gates based on optimal control, atomic dark states to reduce scattering and improvements to Rydberg excitation and atom cooling. We benchmark fidelity using several methods based on repeated gate applications, characterize the physical error sources and outline future improvements. Finally, we generalize our method to design entangling gates involving a higher number of qubits, which we demonstrate by realizing low-error three-qubit gates. By enabling high-fidelity operation in a scalable, highly connected system, these advances lay the groundwork for large-scale implementation of quantum algorithms, error-corrected circuits and digital simulations.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS1987

Parallel VLSI architecture emulation and the organization of APSA/MPP

The Applicative Programming System Architecture (APSA) combines an applicative language interpreter with a novel parallel computer architecture that is well suited for Very Large Scale Integration (VLSI) implementation. The Massively Parallel Processor (MPP) can simulate VLSI circuits by allocating one processing element in its square array to an area on a square VLSI chip. As long as there are not too many long data paths, the MPP can simulate a VLSI clock cycle very rapidly. The APSA circuit contains a binary tree with a few long paths and many short ones. A skewed H-tree layout allows every processing element to simulate a leaf cell and up to four tree nodes, with no loss in parallelism. Emulation of a key APSA algorithm on the MPP resulted in performance 16,000 times faster than a Vax. This speed will make it possible for the APSA language interpreter to run fast enough to support research in parallel list processing algorithms.

Odonnell, John T.↗
DOE OSTI2021

CUDAMicroBench

CUDAMicroBench is a collection of fourteen microbenchmarks that demonstrate performance challenges in CUDA programming and techniques to optimize the CUDA programs to address these challenges. It also includes examples and techniques for using advanced CUDA features such as data shuffling between threads, dynamic parallelism, etc that can help users optimize the CUDA program for performance.

Liao, Chunhua↗
DOE OSTI2022

ExaSGD: 2022 Kernel Thrust Activities

The Kernel Thrust milestone ADSE22-407 covers the development of device-capable optimization algorithms and solvers technologies required by the ExaSGD project’s software stack in order to solve security-constrained alternating current optimal power flow (SC-ACOPF) problems on emerging exascale architectures. To this extent, in FY22 the main objective of the Kernel Thrust was (i) provide sparse optimization solver that runs efficiently on hardware accelerator devices (i.e., NVIDIA and AMD GPUs) to perform intra-node computations, (ii) strengthen the reliability and increase the performance of the mixed-dense sparse (MDS) solver of HiOp for deployment on the FY22 target architectures, Summit and Crusher, and (iii) increase performance by improving the mathematical algorithm and refining the parallel MPI-based implementation of the coarse-grain parallel solver HiOp-PriDec for capabilities deployment on the FY22 target architectures, Summit and Crusher. This document presents the developments and contributions done by the Kernels Thrust Team in FY22 toward completion of the above-mentioned objectives. These contributions progressed along four main development (sub)thrusts: (1) Design and implementation of a sparse optimization solver for use on hardware accelerators; (2) Improvement of the mathematical algorithm and of the parallel implementation of HiOp-PriDec to ensure readiness and efficient coarse-grain parallelism for FY23 target exascale machine; and (3) Support Software and Application Development Thrusts of the exaSGD project in their deployment of the project’s software stack on AMD- and NVIDIA-based architectures. The development of the sparse optimization solver (thrust 1 above) was new in FY22 and resulted in a new sparse solver in HiOp (available as of version 0.6). The second development thrust was a continuation of the efforts from FY21 and improved the mathematical algorithm and the communication strategy of the HiOp-PriDec solver. The last developement thrust is a large collaborative effort. Namely, the project’s teams from multiple labs (LLNL, PNNL, ORNL, and NREL) performed large-scale demonstration of the ExaSGD software stack, namely the optimization solvers of HiOp interfaced with the modeling front-end ExaGO and the stochastic sampler PowerScenarios. These demonstration efforts solved large-scale instances of the SC-ACOPF challenge problem of medium network sizes (10, 000-bus system) and large number of contingencies on Summit (NVIDIA accelerators) and Crusher (AMD accelerators) systems at ORNL.

97 MATHEMATICS AND COMPUTING↗