Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Accelerators, Computing”

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 1,315 records · Page 73

Understanding GPU errors on large-scale HPC systems and the implications for system design and operation

Increase in graphics hardware performance and improvements in programmability has enabled GPUs to evolve from a graphics-specific accelerator to a general-purpose computing device. Titan, the world's second fastest supercomputer for open science in 2014, consists of more dum 18,000 GPUs that scientists from various domains such as astrophysics, fusion, climate, and combustion use routinely to run large-scale simulations. Unfortunately, while the performance efficiency of GPUs is well understood, their resilience characteristics in a large-scale computing system have not been fully evaluated. We present a detailed study to provide a thorough understanding of GPU errors on a large-scale GPU-enabled system. Our data was collected from the Titan supercomputer at the Oak Ridge Leadership Computing Facility and a GPU cluster at the Los Alamos National Laboratory. We also present results from our extensive neutron-beam tests, conducted at Los Alamos Neutron Science Center (LANSCE) and at ISIS (Rutherford Appleron Laboratories, UK), to measure the resilience of different generations of GPUs. We present several findings from our field data and neutron-beam experiments, and discuss the implications of our results for future GPU architects, current and future HPC computing facilities, and researchers focusing on GPU resilience.

Tiwari, Devesh↗

AHEAD: A Tool for Projecting Next-Generation Hardware Enhancements on GPU-Accelerated Systems

Starting with the Titan supercomputer (at the Oak Ridge Leadership Computing Facility, OLCF) in 2012, top supercomputers have Increasingly leveraged the performance of GPUs to support large-scale computational science. The current No. 1 machine, the 200 petaflop Summit system at OLCF, is a GPU-based machine. Accelerator-based architectures, however, add additional complexity due to node heterogeneity. To inform procurement decisions, supercomputing centers need the tools to quickly model the impact of changes of the node architectures on application performance. We present AHEAD, a profiling and modeling tool to quantify the impact of intra-node communication mechanism (e.g., PCI or NVLink) on application performance. Our experiments show average weighted relative errors of ~19% and ~23% for five CORAL-2 (a collaboration between multiple US Department of Energy, DOE, labs to procure Exascale systems) and 12 Rodinia benchmarks respectively, without running the applications on the target future node.

Abdelhafez, Hazem A.↗

Power distribution studies for CMS forward tracker

The Electronic Systems Engineering Department of the Computing Division at the Fermi National Accelerator Laboratory is carrying out R&D investigations for the upgrade of the power distribution system of the Compact Muon Solenoid (CMS) Pixel Tracker at the Large Hadron Collider (LHC). Among the goals of this effort is that of analyzing the feasibility of alternative powering schemes for the forward tracker, including DC to DC voltage conversion techniques using commercially available and custom switching regulator circuits. Tests of these approaches are performed using the PSI46 pixel readout chip currently in use at the CMS Tracker. Performance measures of the detector electronics will include pixel noise and threshold dispersion results. Issues related to susceptibility to switching noise will be studied and presented. In this paper, we describe the current power distribution network of the CMS Tracker, study the implications of the proposed upgrade with DC-DC converters powering scheme and perform noise susceptibility analysis.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Feature-Based Analysis of Plasma-Based Particle Acceleration Data

Plasma-based particle accelerators can produce and sustain thousands of times stronger acceleration fields than conventional particle accelerators, providing a potential solution to the problem of the growing size and cost of conventional particle accelerators. To facilitate scientific knowledge discovery from the ever growing collections of accelerator simulation data generated by accelerator physicists to investigate next-generation plasma-based particle accelerator designs, we describe a novel approach for automatic detection and classification of particle beams and beam substructures due to temporal differences in the acceleration process, here called acceleration features. The automatic feature detection in combination with a novel visualization tool for fast, intuitive, query-based exploration of acceleration features enables an effective top-down data exploration process, starting from a high-level, feature-based view down to the level of individual particles. We describe the application of our analysis in practice to analyze simulations of single pulse and dual and triple colliding pulse accelerator designs, and to study the formation and evolution of particle beams, to compare substructures of a beam and to investigate transverse particle loss.

43 PARTICLE ACCELERATORS↗

Validation of a cyclic plasticity computational method using fatigue full-field deformation measurements

The evolution of crack tip displacement and strain fields during uniaxial, room temperature, low-cycle fatigue experiments of Nickel superalloy compact tension specimens was measured by a digital image correlation approach and was further used to validate a cyclic plasticity model and corresponding deformation calculations made by a finite elements methodology. The experimental results provided data trends for the opening displacements and near crack tip strains as function of cycles. A finite element model was developed to capture test conditions for a measured crack size. The model captures crack tip plasticity by using a constitutive model calibrated against stress-strain measurements performed on a round bar. Similar quantities were extracted from the model predictions to compare with the digital image correlation measurements for model validation purposes. This type of direct comparison demonstrated that the computational model was capable to adequately capture the crack opening displacements at various stages of the specimen's fatigue life, providing in this way a tool for quantitative cyclic plasticity model validation. In addition, this integrated experimental-computational approach provides a framework to accelerate our understanding related to interactions of fatigue test data and models, as well as ways to inform one another.

36 MATERIALS SCIENCE↗

SU-E-T-125: Dosimetric Comparison of Intensity Modulated Radiation Therapy Using Robotic Versus Traditional Linac Platform in Prostate Cancer

Purpose: Recent development of an MLC for robotic external beam radiotherapy has the potential of new clinical application in conventionally fractionated radiation therapy. This study offers a dosimetric comparison of IMRT plans using Cyberknife with MLC versus conventional linac plans. Methods: Ten prostate cancer patients treated on a traditional linac with IMRT to 7920cGy at 180cGy/fraction were randomly selected. GTVs were defined as prostate plus proximal seminal vesicles. PTVs were defined as GTV+8mm in all directions except 5mm posteriorly. Conventional IMRT planning was performed on Philips Pinnacle and delivered on a standard linac with CBCT and 10mm collimator leaf width. For each case a Cyberknife plan was created using Accuray Multiplan with same CT data set, contours, and dose constraints. All dosimetric data was transferred to third party software for independent computation of contour volumes and DVH. Delivery efficiency was evaluated using total MU, treatment time, number of beams, and number of segments. Results: Evaluation criteria including percent target coverage, homogeneity index, and conformity index were found to be comparable. All dose constraints from QUANTEC were found to be statistically similar except rectum V50Gy and bladder V65Gy. Average rectum V50Gy was lower for robotic IMRT (30.07%±6.57) versus traditional (34.73%±3.62, p=0.0130). Average bladder V65Gy was lower for robotic (17.87%±12.74) versus traditional (21.03%±11.93, p=0.0405). Linac plans utilized 9 coplanar beams, 48.9±3.8 segments, and 19381±2399MU. Robotic plans utilized 38.4±9.0 non-coplanar beams, 85.5±21.0 segments and 42554.71±16381.54 MU. The average treatment was 15.02±0.60 minutes for traditional versus 20.90±2.51 for robotic. Conclusion: The robotic IMRT plans were comparable to the traditional IMRT plans in meeting the target volume dose objectives. Critical structure dose constraints were largely comparable although statistically significant differences were found in favor of the robotic platform in terms of rectum V50Gy and bladder V65Gy at a cost of 25% longer treatment time.

60 APPLIED LIFE SCIENCES↗

Technical Note: Development and performance of a software tool for quality assurance of online replanning with a conventional Linac or MR-Linac

Purpose: To develop an integrated quality assurance (QA) software tool for online replanning capable of efficiently and automatically checking radiation treatment (RT) planning parameters and gross plan quality, verifying treatment plan data transfer from treatment planning system (TPS) to record and verify (R&V) system, performing a secondary monitor unit (MU) calculation with or without a presence of a magnetic field from MR-Linac, and validating the delivery record consistency with the plan. Methods: The software tool, named ArtQA, was developed to obtain and compare plan and treatment parameters from both the TPS and the R&V system database. The TPS data are accessed via direct file reading and the R&V data are retrieved via open database connectivity and structured query language. Plan quality is evaluated with both the logical consistency of planning parameters and the achieved dose–volume histograms. Beams in between the TPS and R&V system are matched based on geometry configurations. To consider the effect of a 1.5 T transverse magnetic field from MR-Linac in the secondary MU calculation, a method based on modified Clarkson integration algorithm was developed and tested for a series of clinical situations. Results: ArtQA has been used in their clinic and can quickly detect inconsistencies and deviations in the entire RT planning process. With the use of the ArtQA tool, the efficiency for plan check including plan quality, data transfer, and delivery check can be improved by at least 60%. The newly developed independent MU calculation tool for MR-Linac reduces the difference between the plan and calculated MUs by 10%. Conclusions: The software tool ArtQA can be used to perform a comprehensive QA check from planning to delivery with conventional Linac or MR-Linac and is an essential tool for online replanning where the QA check needs to be performed rapidly.

60 APPLIED LIFE SCIENCES↗

Leading-order electroweak corrections to the Drell-Yan process in the CMS experiment at LHC

The leading part of electroweak radiative corrections to the Drell-Yan process in the CMS (compact muon solenoid detector) experiment at LHC were calculated. Analytic and numerical tests indicate that the results are independent of all nonphysical parameters. A numerical analysis of radiative effects was performed by using the FORTRAN code READY with allowance for experimental constraints at the CMS detector. The results presented for the relative correction to the fully differential cross section for the Drell-Yan process over the entire kinematical region of the CMS detector would hopefully be helpful in handling experimental data.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

XploreNAS : Explore Adversarially Robust and Hardware-efficient Neural Architectures for Non-ideal Xbars

Compute In-Memory platforms such as memristive crossbars are gaining focus as they facilitate acceleration of Deep Neural Networks (DNNs) with high area and compute efficiencies. However, the intrinsic non-idealities associated with the analog nature of computing in crossbars limits the performance of the deployed DNNs. Furthermore, DNNs are shown to be vulnerable to adversarial attacks leading to severe security threats in their large-scale deployment. Thus, finding adversarially robust DNN architectures for non-ideal crossbars is critical to the safe and secure deployment of DNNs on the edge. This work proposes a two-phase algorithm-hardware co-optimization approach called XploreNAS that searches for hardware efficient and adversarially robust neural architectures for non-ideal crossbar platforms. We use the one-shot Neural Architecture Search approach to train a large Supernet with crossbar-awareness and sample adversarially robust Subnets therefrom, maintaining competitive hardware efficiency. Our experiments on crossbars with benchmark datasets (SVHN, CIFAR10, CIFAR100) show up to ~8–16% improvement in the adversarial robustness of the searched Subnets against a baseline ResNet-18 model subjected to crossbar-aware adversarial training. We benchmark our robust Subnets for Energy-Delay-Area-Products (EDAPs) using the Neurosim tool and find that with additional hardware efficiency–driven optimizations, the Subnets attain ~1.5–1.6× lower EDAPs than ResNet-18 baseline.

97 MATHEMATICS AND COMPUTING↗

The Study on the Shock Wave Propagation Rule of a Gas Explosion in a Gas Compartment

Combined with the k-ε turbulence model of general application, a refined finite element model of a utility tunnel’s gas compartment filled with the methane/air mixture is developed. A series of analyses are made by using the powerful industry-leading computational fluid dynamics (CFD) software flame acceleration simulator (FLACS) to study the shock wave propagation rule in the gas compartment. The longitudinal and transversal distribution laws of the explosion shock wave are gained taking into consideration the spatial characteristics of the gas compartment. The influences of a few parameters, such as initial conditions and section size of the gas compartment, on the shock wave propagation rule are further discussed. The basic procedure for predicting the peak pressure of the blast wave is provided by considering the initial conditions and the gas compartment, and the corresponding injury effect of the explosion wave on the living beings is assessed. The investigation demonstrates that the peak pressure by the coupled effect between the initial conditions is significantly influenced, especially at the upper and lower gas explosion limits. The peak pressure increases gradually as the width or height increases, and both basically meet the linear relation. The proposed method can forecast the peak pressure of the explosion shock wave in the gas compartment accurately. According to the peak pressure longitudinal and transversal distributions of the blast wave, the peak pressure is far greater than the killing pressure threshold in the underground and closed space; consequently, it is not safe for the living beings in the gas compartment.

Lin, Shu-Chao↗

Dax v. 1.0

The Dax Toolkit enables the fine-grained concurrency for data analysis and visualization required to drive multi-core processors, many-core GPU accelerators, and emerging architectures for exascale computing. Dax simplifies the design of finely threaded algorithms by characterizing them in per element operations, which we call worklets. Worklets simplify design because they can behave as serial processes without the complications of memory clashes or other race conditions. Yet worklets can be scheduled on a practically unlimited number of threads, thereby making effective utilization of many processing cores. Operations coordinated across elements are encapsulated in a few versatile communicative operations that are adapted to the specifics of computing devices.

SciDAC↗

ML-Shock-Time-Series-Synthesis

Open-source machine learning tools for GPU-batched synthetic shock time-series generation, GPU-accelerated batched Shock Response Spectrum (SRS) computation, and standardized benchmark datasets.

Watts, Adam↗

Computational and empirical studies predict Mycobacterium tuberculosis -specific T cells as a biomarker for infection outcome

Identifying biomarkers for tuberculosis (TB) is an ongoing challenge in developing immunological correlates of infection outcome and protection. Biomarker discovery is also necessary for aiding design and testing of new treatments and vaccines. To effectively predict biomarkers for infection progression in any disease, including TB, large amounts of experimental data are required to reach statistical power and make accurate predictions. We took a two-pronged approach using both experimental and computational modeling to address this problem. We first collected 200 blood samples over a 2-year period from 28 non-human primates (NHP) infected with a low dose of Mycobacterium tuberculosis . We identified T cells and the cytokines that they were producing (single and multiple) from each sample along with monkey status and infection progression data. Machine learning techniques were used to interrogate the experimental NHP datasets without identifying any potential TB biomarker. In parallel, we used our extensive novel NHP datasets to build and calibrate a multi-organ computational model that combines what is occurring at the site of infection (e.g., lung) at a single granuloma scale with blood level readouts that can be tracked in monkeys and humans. We then generated a large in silico repository of in silico granulomas coupled to lymph node and blood dynamics and developed an in silico tool to scale granuloma level results to a full host scale to identify what best predicts Mycobacterium tuberculosis (Mtb) infection outcomes. The analysis of in silico blood measures identifies Mtb -specific frequencies of effector T cell phenotypes at various time points post infection as promising indicators of infection outcome. As a result, we emphasize that pairing wetlab and computational approaches holds great promise to accelerate TB biomarker discovery.

59 BASIC BIOLOGICAL SCIENCES↗

Small-angle X-ray scattering characteristics of mouse brain: Planar imaging measurements and tomographic imaging simulations

Small-angle x-ray scattering (SAXS) imaging can differentiate tissue types based on their nanoscale molecular structure. However, characterization of the coherent scattering cross-section profile of relevant tissues is needed to optimally design SAXS imaging techniques for a variety of biomedical applications. Reported measured nervous tissue x-ray scattering cross sections under a synchrotron source have had limited agreement. We report a set of x-ray cross-section measurements obtained from planar SAXS imaging of 1 mm thick mouse brain (APP/PS1 wild-type) coronal slices using an 8 keV laboratory x-ray source. Two characteristic peaks were found at 0.96 and 1.60 nm -1 attributed to myelin. The peak intensities varied by location in the slice. We found that regions of gray matter, white matter, and corpus callosum could be segmented by their increasing intensities of myelin peaks respectively. Measured small-angle x-ray scattering cross sections were then used to define brain tissue scattering properties in a GPU-accelerated Monte Carlo simulation of SAXS computed tomography (CT) using a higher monochromatic x-ray energy (20 keV) to study design trade-offs for noninvasive in vivo SAXS imaging on a small-animal head including radiation dose, signal-to-noise ratio (SNR), and the effect of skull presence on the previous two metrics. Simulation results show the estimated total dose to the mouse head for a single SAXS-CT slice was 149.4 mGy. The pixel SNR was approximately 30.8 for white matter material whether or not a skull was present. In this early-stage proof-of-principle work, we have demonstrated our brain cross-section data and simulation tools can be used to assess optimal instrument parameters for dedicated small-animal SAXS-CT prototypes.

59 BASIC BIOLOGICAL SCIENCES↗

Parallelization and checkpointing of GPU applications through program transformation

GPUs have emerged as a powerful tool for accelerating general-purpose applications. The availability of programming languages that makes writing general-purpose applications for running on GPUs tractable have consolidated GPUs as an alternative for accelerating general purpose applications. Among the areas that have benefited from GPU acceleration are: signal and image processing, computational fluid dynamics, quantum chemistry, and, in general, the High Performance Computing (HPC) Industry. In order to continue to exploit higher levels of parallelism with GPUs, multi-GPU systems are gaining popularity. In this context, single-GPU applications are parallelized for running in multi-GPU systems. Furthermore, multi-GPU systems help to solve the GPU memory limitation for applications with large application memory footprint. Parallelizing single-GPU applications has been approached by libraries that distribute the workload at runtime, however, they impose execution overhead and are not portable. On the other hand, on traditional CPU systems, parallelization has been approached through application transformation at pre-compile time, which enhances the application to distribute the workload at application level and does not have the issues of library-based approaches. Hence, a parallelization scheme for GPU systems based on application transformation is needed. Like any computing engine of today, reliability is also a concern in GPUs. GPUs are vulnerable to transient and permanent failures. Current checkpoint/restart techniques are not suitable for systems with GPUs. Checkpointing for GPU systems present new and interesting challenges, primarily due to the natural differences imposed by the hardware design, the memory subsystem architecture, the massive number of threads, and the limited amount of synchronization among threads. Therefore, a checkpoint/restart technique suitable for GPU systems is needed. The goal of this work is to exploit higher levels of parallelism and to develop support for application-level fault tolerance in applications using multiple GPUs. Our techniques reduce the burden of enhancing single-GPU applications to support these features. To achieve our goal, this work designs and implements a framework for enhancing a single-GPU OpenCL application through application transformation.

97 MATHEMATICS AND COMPUTING↗