Search NASA⌕ Search

SEARCH · Search NASA

Results for “hardware 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 325 records · Page 18

X-ray scatter correction method for dedicated breast computed tomography

Purpose: To improve image quality and accuracy in dedicated breast computed tomography (BCT) by removing the x-ray scatter signal included in the BCT projections. Methods: The previously characterized magnitude and distribution of x-ray scatter in BCT results in both cupping artifacts and reduction of contrast and accuracy in the reconstructions. In this study, an image processing method is proposed that estimates and subtracts the low-frequency x-ray scatter signal included in each BCT projection postacquisition and prereconstruction. The estimation of this signal is performed using simple additional hardware, one additional BCT projection acquisition with negligible radiation dose, and simple image processing software algorithms. The high frequency quantum noise due to the scatter signal is reduced using a noise filter postreconstruction. The dosimetric consequences and validity of the assumptions of this algorithm were determined using Monte Carlo simulations. The feasibility of this method was determined by imaging a breast phantom on a BCT clinical prototype and comparing the corrected reconstructions to the unprocessed reconstructions and to reconstructions obtained from fan-beam acquisitions as a reference standard. One-dimensional profiles of the reconstructions and objective image quality metrics were used to determine the impact of the algorithm. Results: The proposed additional acquisition results in negligible additional radiation dose to the imaged breast ({approx}0.4% of the standard BCT acquisition). The processed phantom reconstruction showed substantially reduced cupping artifacts, increased contrast between adipose and glandular tissue equivalents, higher voxel value accuracy, and no discernible blurring of high frequency features. Conclusions: The proposed scatter correction method for dedicated breast CT is feasible and can result in highly improved image quality. Further optimization and testing, especially with patient images, is necessary to characterize its impact on clinical performance.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

NDCX-II project commencing at LBNL

Earlier this year, the U.S. Department of Energy Office of Fusion Energy Sciences approved the NDCX-II project, a second-generation Neutralized Drift Compression eXperiment. NDCX-II is a collaborative effort of scientists and engineers from Lawrence Berkeley National Laboratory (LBNL), Lawrence Livermore National Laboratory (LLNL), and the Princeton Plasma Physics Laboratory (PPPL), in a formal collaboration known as the Virtual National Laboratory for Heavy Ion Fusion Science (HIFS-VNL). Supported by $11 M of funding from the American Recovery and Reinvestment Act, construction at LBNL commenced in July of 2009, with completion anticipated in March of 2012. Applications of this facility will include studies of: the basic physics of the poorly understood 'warm dense matter' regime of temperatures around 1 eV and densities near solid, using uniform, volumetric ion heating of thin foil targets; ion energy coupling into an ablating plasma (such as that which occurs in an inertial fusion target) using beams with time-varying kinetic energy; space-charge-dominated ion beam dynamics; and beam focusing and pulse compression in neutralizing plasma. The machine will complement facilities at GSI in Darmstadt, Germany, but will employ lower ion kinetic energies and commensurately shorter stopping ranges in matter. Much of this research will contribute directly toward the collaboration's ultimate goal of electric power production via heavy-ion beam-driven inertial confinement fusion ('Heavy-Ion Fusion', or HIF). In inertial fusion, a target containing fusion fuel is heated by energetic 'driver' beams, and undergoes a miniature thermonuclear explosion. Currently the largest U.S. research program in inertial confinement is at Livermore's National Ignition Facility (NIF), a multibillion-dollar, stadium-sized laser facility optimized for studying physics issues relevant to nuclear stockpile stewardship. Nonetheless, NIF is expected to establish the fundamental feasibility of fusion ignition on the laboratory scale, and thus advance this approach to fusion energy. Heavy ion accelerators have a number of attributes (such as efficiency, longevity, and use of magnetic fields for final focusing) that make them attractive candidates as Inertial Fusion energy (IFE) drivers As with LBNL's existing NDCX-I, the new machine will produce short ion pulses using the technique of neutralized drift compression. A head-to-tail velocity gradient is imparted to the beam, which then shortens as it drifts in neutralizing plasma that suppresses space-charge forces. NDCX-II will make extensive use of induction cells and other hardware from the decommissioned ATA facility at LLNL. Figure (1) shows the layout of the facility, to be sited in LBNL's Building 58 alongside the existing NDCX-I apparatus. This second-generation facility represents a significant upgrade from the existing NDCX-I. It will be extensible and reconfigurable; in the configuration that has received the most emphasis, each NDCX-II pulse will deliver 30 nC of ions at 3 MeV into a mm-scale spot onto a thin-foil target. Pulse compression to {approx} 1 ns occurs in the accelerator as well as in the drift compression line; the beam is manipulated using suitably tailored voltage waveforms in the accelerating gaps. NDCX-II employs novel beam dynamics. To use the 200 kV Blumlein power supplies from ATA (blue cylinders in the figure), the pulse duration must first be reduced to less than 70 ns. This shortening is accomplished in an initial stage of non-neutral drift compression, downstream of the injector and the first few induction cells. The compression is sufficiently rapid that fewer than ten long-pulse waveform generators are needed, with Blumleins powering the rest of the acceleration. Extensive simulation studies have enabled an attractive physics design; these employ both a new 1-D code (ASP) and the VNL's workhorse 2-D/3-D code Warp. Snapshots from a simulation movie (available online) appear in Fig. 2. Studies on a dedicated test stand are quantifying the performance of the ATA hardware and of pulsed solenoids that will provide transverse beam confinement (ions require much stronger fields than the electrons accelerated by ATA). For more information, see the recent article in the Berkeley Lab News and references therein. Joe Kwan is the NDCX-II project manager and Alex Friedman is the leader for the physics design.

43 PARTICLE ACCELERATORS↗

Record acceleration of the two-dimensional Ising model using a high-performance wafer-scale engine

The versatility and wide-ranging applicability of the Ising model, originally introduced to study phase transitions in magnetic materials, have made it a cornerstone in statistical physics and a valuable tool for evaluating the performance of emerging computer hardware. Here, we present a novel implementation of the two-dimensional Ising model on Cerebras Wafer-Scale Engine (WSE) – a revolutionary processor that is opening new frontiers in computing. In our deployment of the checkerboard algorithm, we optimized the Ising model to take advantage of the unique WSE architecture. Specifically, we employed a compressed bit representation storing 16 spins on each int16 word, and efficiently distributed the spins over the processing units enabling seamless weak scaling and limiting communications to only immediate neighboring units. Our implementation can handle up to 754 simulations in parallel, achieving an aggregate of over 61.8 trillion flip attempts per second for Ising models with up to 200 million spins. This represents a gain of up to 148 times over previously reported single-devices with a highly optimized implementation on NVIDIA V100 and up to 88 times in productivity compared to NVIDIA H100. Our findings highlight the significant potential of the WSE in scientific computing, particularly in the field of materials modeling.

Ising model↗

Hardware MPI message matching: Insights into MPI matching behavior to inform design: Hardware MPI message matching

Here, this paper explores key differences of MPI match lists for several important United States Department of Energy (DOE) applications and proxy applications. This understanding is critical in determining the most promising hardware matching design for any given high-speed network. The results of MPI match list studies for the major open-source MPI implementations, MPICH and Open MPI, are presented, and we modify an MPI simulator, LogGOPSim, to provide match list statistics. These results are discussed in the context of several different potential design approaches to MPI matching–capable hardware. The data illustrate the requirements for different hardware designs in terms of performance and memory capacity. Finally, this paper's contributions are the collection and analysis of data to help inform hardware designers of common MPI requirements and highlight the difficulties in determining these requirements by only examining a single MPI implementation.

97 MATHEMATICS AND COMPUTING↗

Exploiting Performance Portability in Search Algorithms for Autotuning

Autotuning seeks the best configuration of an application by orchestrating hardware and software knobs that affect performance on a given machine. Autotuners adopt various search techniques to efficiently find the best configuration, but they often ignore lessons learned on one machine when tuning for another machine. We demonstrate that a surrogate model built from performance results on one machine can speedup the autotuning search by 1.6X to 130X on a variety of modern architectures.

Performance↗

Hierarchical Task-Data Parallelism using Kokkos and Qthreads

This report describes a new capability for hierarchical task-data parallelism using Sandia's Kokkos and Qthreads, and evaluation of this capability with sparse matrix Cholesky factorization and social network triangle enumeration mini-applications. Hierarchical task-data parallelism consists of a collection of tasks with executes-after dependences where each task contains data parallel operations performed on a team of hardware threads. The collection of tasks and dependences form a directed acyclic graph of tasks - a task DAG. Major challenges of this research and development effort include: portability and performance across multicore CPU; manycore Intel Xeon Phi, and NVIDIA GPU architectures; scalability with respect to hardware concurrency and size of the task DAG; and usability of the application programmer interface (API).

97 MATHEMATICS AND COMPUTING↗

Pulse discrimination with a Gaussian mixture model on an FPGA

A Gaussian Mixture Model (GMM) based machine learning algorithm has been applied to the problem of gamma/neutron pulse shape discrimination (PSD). The algorithm has been successfully implemented on a standard PC as well as a field programmable gate array (FPGA). We describe the GMM classifier and its implementation on these two different types of hardware. We compare the performance of the algorithm on these two platforms against each other, along with other standard techniques applied in PSD. Our results show that the FPGA-based GMM classifier outperforms the standard PSD techniques in terms of classification accuracy at low particle energy and executes more quickly than its CPU-based counterpart.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

JDFTx: Software for joint density-functional theory

Density-functional theory (DFT) has revolutionized computational prediction of atomic-scale properties from first principles in physics, chemistry and materials science. Continuing development of new methods is necessary for accurate predictions of new classes of materials and properties, and for connecting to nano- and mesoscale properties using coarse-grained theories. JDFTx is a fully-featured open-source electronic DFT software designed specifically to facilitate rapid development of new theories, models and algorithms. Using an algebraic formulation as an abstraction layer, compact C++11 code automatically performs well on diverse hardware including GPUs (Graphics Processing Units). This code hosts the development of joint density-functional theory (JDFT) that combines electronic DFT with classical DFT and continuum models of liquids for first-principles calculations of solvated and electrochemical systems. In addition, the modular nature of the code makes it easy to extend and interface with, facilitating the development of multi-scale toolkits that connect to ab initio calculations, e.g. photo-excited carrier dynamics combining electron and phonon calculations with electromagnetic simulations.

97 MATHEMATICS AND COMPUTING↗

Accelerating End-to-End Deep Learning for Particle Reconstruction using CMS open data

Machine learning algorithms are gaining ground in high energy physics for applications in particle and event identification, physics analysis, detector reconstruction, simulation and trigger. Currently, most data-analysis tasks at LHC experiments benefit from the use of machine learning. Incorporating these computational tools in the experimental framework presents new challenges. This paper reports on the implementation of the end-to-end deep learning with the CMS software framework and the scaling of the end-to-end deep learning with multiple GPUs. The end-to-end deep learning technique combines deep learning algorithms and low-level detector representation for particle and event identification. We demonstrate the end-to-end implementation on a top quark benchmark and perform studies with various hardware architectures including single and multiple GPUs and Google TPU.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Qu8its for quantum simulations of lattice quantum chromodynamics

We explore the utility of d = 8 qudits, qu8its, for quantum simulations of the dynamics of 1+1⁢D SU(3) lattice quantum chromodynamics, including a mapping for arbitrary number of flavors and lattice size and a reorganization of the Hamiltonian for efficient time evolution. Recent advances in parallel gate applications, along with the shorter application times of single-qudit operations compared with two-qudit operations, lead to significant projected advantages in quantum simulation fidelities and circuit depths using qu8its rather than qubits. The number of two-qudit entangling gates required for time evolution using qu8its is found to be more than a factor of 5 fewer than for qubits. Here, we anticipate that the developments presented in this work will enable improved quantum simulations to be performed using emerging quantum hardware.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

A Novel Protection Scheme for Unbalanced Faults in Inverter Dominated Networks: A Computationally Efficient Algorithm for Entry-Level Relays

Microgrids are now a common practice in distribution systems to increase resilience and reliability. However, microgrid protection remains a critical challenge, considering its requirement to operate in both grid connected and islanded, and the variability in fault characteristics under each mode of operation. This paper presents unbalanced power (S unb ) based fault detection algorithm, which considers local voltage and current unbalances to determine faults in the system. S unb is a computationally efficient fault detection algorithm that is suitable for implementation in the programmable logic of entry level protective relays. In addition, the difference in current and voltage unbalance (D n ) is used to determine the fault type. The proposed method demonstrates high sensitivity and selectivity for line-to-ground (LG), line-to-line (LL), and double line-to-ground (LLG) faults, representing the most common faults in distribution systems. It also allows relay coordination with upstream and downstream protection devices in both island and grid connected operation, while preserving grading margins. The same pickup and time multiplier settings of a particular relay for both modes of operation eliminates the need for adaptive settings, which rely on communication networks. Validation was performed with a hardware-in-the-loop (HIL) setup using Typhoon HIL real time simulator interfaced with three entry-level, SEL 751 relays. Results confirmed the algorithm’s ability to discriminate fault conditions, and determine the fault type under both operating modes, maintain fast detection times, and ensure proper protection coordination.

fault classification↗

Machine Learning-based False Data Injection Attack Detection and Localization in Power Grids

Cyberattacks on critical infrastructures can be catastrophic and bring nations to their knees. Therefore, detecting these attacks is crucial and challenging. This paper presents a novel approach for detecting and locating cyberattacks affecting an electrical power system. The adversary employs a man-in-the-middle technique to inject false data into the communication between distributed energy resources (DER) and Microgrid Controller (MGC) with the goal of disrupting power delivery. The approach for detection and localization is based on integrating multiple machine learning-based anomaly detection models that combine network traffic data and grid measurements. Experiments are performed to assess the method's performance using a hardware-in-the-loop real-time simulation testbed which includes Modbus TCP/IP communication. Power system topology and operating conditions are based on actual topology and real-world data provided by the Holy Cross Energy utility network. Results confirm that the method can be successfully employed for detecting and localizing cyberattacks.

Leao, Bruno P.↗

Performance-Aligned LLMs for Generating Fast HPC Code

Optimizing scientific software is a difficult task because codebases are often large and complex, and performance can depend upon several factors including the algorithm, its implementation, and hardware among others. Causes of poor performance can originate from disparate sources and be difficult to diagnose. Recent years have seen a multitude of work that use large language models (LLMs) to assist in software development tasks. However, these tools are trained to model the distribution of code as text, and are not specifically designed to understand performance aspects of code. In this work, we introduce a reinforcement learning based methodology to align the outputs of code LLMs with performance. This allows us to build upon the current code modeling capabilities of LLMs and extend them to generate better performing code. Here, we demonstrate that our fine-tuned model improves the expected speedup of generated code over base models for a set of benchmark tasks from 0.9 to 1.6 for serial code and 1.9 to 4.5 for OpenMP parallel code.

Computer science↗

IGFC response to initial fuel cell load for various syngas compositions

The system response to an initial electric load of the fuel cell during the startup of a direct-fired fuel cell turbine power system was studied using the Hybrid Performance (Hyper) project hardware-based simulation facility at the U.S. Department of Energy, National Energy Technology Laboratory for a range of input fuel compositions. The facility was brought to a steady condition at a temperature deemed adequate to minimize stress on the fuel cell during the initial load transient. A 1D distributed fuel cell model operating in real-time was used to produce individual cell transient temperature profiles during the course of the load change. The process was conducted with humidified hydrogen, and then repeated with various syngas compositions representative of different gasifier technologies. The results provide insight into control strategy requirements for mitigation of expected fuel cell failure modes relevant to available gasifier technology.

30 DIRECT ENERGY CONVERSION↗

Multi-GPU implementation of a VMAT treatment plan optimization algorithm

Purpose: Volumetric modulated arc therapy (VMAT) optimization is a computationally challenging problem due to its large data size, high degrees of freedom, and many hardware constraints. High-performance graphics processing units (GPUs) have been used to speed up the computations. However, GPU’s relatively small memory size cannot handle cases with a large dose-deposition coefficient (DDC) matrix in cases of, e.g., those with a large target size, multiple targets, multiple arcs, and/or small beamlet size. The main purpose of this paper is to report an implementation of a column-generation-based VMAT algorithm, previously developed in the authors’ group, on a multi-GPU platform to solve the memory limitation problem. While the column-generation-based VMAT algorithm has been previously developed, the GPU implementation details have not been reported. Hence, another purpose is to present detailed techniques employed for GPU implementation. The authors also would like to utilize this particular problem as an example problem to study the feasibility of using a multi-GPU platform to solve large-scale problems in medical physics. Methods: The column-generation approach generates VMAT apertures sequentially by solving a pricing problem (PP) and a master problem (MP) iteratively. In the authors’ method, the sparse DDC matrix is first stored on a CPU in coordinate list format (COO). On the GPU side, this matrix is split into four submatrices according to beam angles, which are stored on four GPUs in compressed sparse row format. Computation of beamlet price, the first step in PP, is accomplished using multi-GPUs. A fast inter-GPU data transfer scheme is accomplished using peer-to-peer access. The remaining steps of PP and MP problems are implemented on CPU or a single GPU due to their modest problem scale and computational loads. Barzilai and Borwein algorithm with a subspace step scheme is adopted here to solve the MP problem. A head and neck (H and N) cancer case is then used to validate the authors’ method. The authors also compare their multi-GPU implementation with three different single GPU implementation strategies, i.e., truncating DDC matrix (S1), repeatedly transferring DDC matrix between CPU and GPU (S2), and porting computations involving DDC matrix to CPU (S3), in terms of both plan quality and computational efficiency. Two more H and N patient cases and three prostate cases are used to demonstrate the advantages of the authors’ method. Results: The authors’ multi-GPU implementation can finish the optimization process within ∼1 min for the H and N patient case. S1 leads to an inferior plan quality although its total time was 10 s shorter than the multi-GPU implementation due to the reduced matrix size. S2 and S3 yield the same plan quality as the multi-GPU implementation but take ∼4 and ∼6 min, respectively. High computational efficiency was consistently achieved for the other five patient cases tested, with VMAT plans of clinically acceptable quality obtained within 23–46 s. Conversely, to obtain clinically comparable or acceptable plans for all six of these VMAT cases that the authors have tested in this paper, the optimization time needed in a commercial TPS system on CPU was found to be in an order of several minutes. Conclusions: The results demonstrate that the multi-GPU implementation of the authors’ column-generation-based VMAT optimization can handle the large-scale VMAT optimization problem efficiently without sacrificing plan quality. The authors’ study may serve as an example to shed some light on other large-scale medical physics problems that require multi-GPU techniques.

60 APPLIED LIFE SCIENCES↗

BitGNN: Unlocking the Performance Potential of Binary Graph Neural Networks on GPUs

Graph Neural Networks (GNNs) have shown compelling results in many graph-based learning tasks. They are, however, time-consuming. Recent work has shown a promising direction in improving GNN speed and shrinking the size — network binarization, which binarizes network values and operations. Prior work, however, mainly focused on algorithm designs, leaving it open on how to fully materialize the performance potential. This work fills the gap by proposing techniques to best map binary GNNs and their computations to fit the nature of bit manipulations, optimizations and algorithms to maximize BSpMM kernel efficiency, and solutions to other factors influencing the end-to-end time on GPUs. Results on real-world graphs show that the proposed techniques outperform state of-the-art binary GNN implementations by 21-67× with little accuracy loss.

Chen, Jou-An↗

Pressio

Reduced-order Models (ROMs). ROMs are "compressed" approximations of high-fidelity computational models. Implementing ROM methods in production codes is typically intrusive, i.e. the native codes must be adapted and modified, with individual ROM methods reimplemented in every application code. This clearly constitutes an obstacle to users who want to use ROMs. To address this, we have developed Pressio, which is a C++ header-only ROM library that can be used by an application code in a non-intrusive fashion. The main objectives behind the design of Pressio are: (a) support for general computational models (data structures, PDE types); (b) heterogeneous hardware (CPU, GPU); (c) performance; and (d) extensibility. Using metaprogramming techniques, Pressio detects and leverages the application?s native linear algebra data structures (e.g. vector and matrix) to run the ROM methods. The application only needs to expose the velocity vector and Jacobian matrix via an adapter class. Pressio checks at compile time that the target application meets the correct API. The library is structured in several modular packages including (a) non-linear least-square solvers (e.g. Gauss-Newton with and without line-search); (b) singular value decomposition (for the computation of the ROM generalized coordinates) and (c) algebraic hyper-reduction.SAND2019-5878 M

Blonigan, Patrick↗