Search NASA⌕ Search

SEARCH · Search NASA

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 1,477 records · Page 82

SU-E-T-348: Verification MU Calculation for Conformal Radiotherapy with Multileaf Collimator Using Report AAPM TG 114

Purpose: The purpose of this work was to develop and validate an open source independent MU dose calculation software for 3D conformal radiotherapy with multileaf high and low resolution according to the report of AAPM TG 11 Methods: Treatment plans were done using Iplan v4.5 BrainLAB TPS. A 6MV photon beam produced by Primus and Novalis linear accelerators equipped with an Optifocus MLC and HDMLC, respectively. TPS dose calculation algorithms were pencil beam and Monte Carlo. 1082 treatments plans were selected for the study. The algorithm was written in free and open source CodeBlocks C++ platform. Treatment plans were imported by the software using RTP format. Equivalent size field is obtained from the positions of the leaves; the effective depth of calculation can be introduced by TPS's dosimetry report or automatically calculated starting from SSD. The inverse square law is calculated by the 3D coordinates of the isocenter and normalization point of the treatment plan. The dosimetric parameters TPR, Sc, Sp and WF are linearly interpolated. Results: 1082 plans of both machines were analyzed. The average uncertainty between the TPS and the independent calculation was −0.43% ± 2.42% [−7.90%, 7.50%]. Specifically for the Primus the variation obtained was −0.85% ± 2.53% and for the Novalis 0.00% ± 2.23%. Data show that 94.8% of the cases the uncertainty was less than or equal to 5%, while 98.9% is less than or equal to 6%. Conclusion: The developed software is appropriate for use in calculation of UM. This software can be obtained upon request.

07 ISOTOPE AND RADIATION SOURCES↗

MemHC: An Optimized GPU Memory Management Framework for Accelerating Many-body Correlation

The many-body correlation function is a fundamental computation kernel in modern physics computing applications, e.g., Hadron Contractions in Lattice quantum chromodynamics (QCD). This kernel is both computation and memory intensive, involving a series of tensor contractions, and thus usually runs on accelerators like GPUs. Existing optimizations on many-body correlation mainly focus on individual tensor contractions (e.g., cuBLAS libraries and others). In contrast, this work discovers a new optimization dimension for many-body correlation by exploring the optimization opportunities among tensor contractions. More specifically, it targets general GPU architectures (both NVIDIA and AMD) and optimizes many-body correlation’s memory management by exploiting a set of memory allocation and communication redundancy elimination opportunities: first, GPU memory allocation redundancy: the intermediate output frequently occurs as input in the subsequent calculations; second, CPU-GPU communication redundancy: although all tensors are allocated on both CPU and GPU, many of them are used (and reused) on the GPU side only, and thus, many CPU/GPU communications (like that in existing Unified Memory designs) are unnecessary; third, GPU oversubscription: limited GPU memory size causes oversubscription issues, and existing memory management usually results in near-reuse data eviction, thus incurring extra CPU/GPU memory communications.

97 MATHEMATICS AND COMPUTING↗

SU-E-T-96: An Analysis of VMAT SBRT Lung Treatment Plans

Purpose: To present an analysis of 218 lung sbrt treatment plans delivered using Varian Eclipse RapidArc software and delivered with Varian linear accelerators. Methods: A retrospective analysis of treatment plans generated using the Varian Eclipse PRO RapidArc VMAT optimization and AAA photon beam calculation algorithms was done for 218 plans delivered using Varian Trilogy and TrueBeam linear accelerators. Some of the patients were enrolled in the RTOG 0813 and RTOG 0915 lung protocols. All of the patients’ plans were optimized using the guidelines outlined by these protocols. The plans presented were normalized so that 95% of the PTV volume received 100% of the prescribed dose. Results: The dose co formality constraints from the RTOG protocols and the number of plans within those constraints were:Tumor dose between 60% of maximum dose and 90% of maximum: 218/218 = 100%.Volume of 105% tumor dose outside put less than 15% of the PTV volume: 218/218 = 100%.Maximum dose 2 cm away from PTV within protocol table guidelines: No minor deviation: 188/218 = 86.2%. No major deviation: 216/218 = 99.1%.Ratio of 50% dose volume to PTV volume less than R50 in protocol table: No minor deviation 148/218 = 67.9%. No major deviation: 213/218 = 97.7%.Percent of lung receiving 20 Gy less than 10% (no minor violation): 218/218 = 100%.99% of PTV receiving at least 90% of the tumor dose: 214/218 = 98.2%. Conclusion: Varian RapidArc VMAT can successfully deliver lung SBRT treatments as outlined in the RTOG 0813 and 0915 protocols. The largest deviations were seen in the R50 constraints for the larger PTV volumes.

60 APPLIED LIFE SCIENCES↗

Using Directional Diffusion Coefficients for Nonlinear Diffusion Acceleration of the First Order SN Equations in Near-Void Regions

The common definition of the diffusion coeffcient as the inverse of three times the transport cross section is not compat- ible with voids. Morel introduced a non-local tensor diffusion coeffcient that remains finite in voids[1]. It can be obtained by solving an auxiliary transport problem without scattering or fission. Larsen and Trahan successfully applied this diffusion coeffcient for enhancing the accuracy of diffusion solutions of very high temperature reactor (VHTR) problems that feature large, optically thin channels in the z-direction [2]. It is demonstrated that a significant reduction of error can be achieved in particular in the optically thin region. Along the same line of thought, non-local diffusion tensors are applied modeling the TREAT reactor confirming the findings of Larsen and Trahan [3]. Previous work of the authors have introduced a flexible Nonlinear-Diffusion Acceleration (NDA) method for the first order S N equations discretized with the discontinuous finite element method (DFEM), [4], [5], [6]. This NDA method uses a scalar diffusion coeffcient in the low-order system that is obtained as the flux weighted average of the inverse transport cross section. Hence, it su?ers from very large and potentially unbounded diffusion coeffcients in the low order problem. However, it was noted that the choice of the diffusion coeffcient does not influence consistency of the method at convergence and hence the di?usion coeffcient is essentially a free parameter. The choice of the di?usion coeffcient does, however, affect the convergence behavior of the nonlinear di?usion iterations. Within this work we use Morel’s non-local di?usion coef- ficient in the aforementioned NDA formulation in lieu of the flux weighted inverse of three times the transport cross section. The goal of this paper is to demonstrate that significant en- hancement of the spectral properties of NDA can be achieved in near void regions. For testing the spectral properties of the NDA with non-local diffusion coeffcients, the periodical horizontal interface problem is used [7]. This problem consists of alternating stripes of optically thin and thick materials both of which feature scattering ratios close to unity.

97 MATHEMATICS AND COMPUTING↗

High-reliability computing for the smarter planet

The geometric rate of improvement of transistor size and integrated circuit performance, known as Moore's Law, has been an engine of growth for our economy, enabling new products and services, creating new value and wealth, increasing safety, and removing menial tasks from our daily lives. Affordable, highly integrated components have enabled both life-saving technologies and rich entertainment applications. Anti-lock brakes, insulin monitors, and GPS-enabled emergency response systems save lives. Cell phones, internet appliances, virtual worlds, realistic video games, and mp3 players enrich our lives and connect us together. Over the past 40 years of silicon scaling, the increasing capabilities of inexpensive computation have transformed our society through automation and ubiquitous communications. In this paper, we will present the concept of the smarter planet, how reliability failures affect current systems, and methods that can be used to increase the reliable adoption of new automation in the future. We will illustrate these issues using a number of different electronic devices in a couple of different scenarios. Recently IBM has been presenting the idea of a 'smarter planet.' In smarter planet documents, IBM discusses increased computer automation of roadways, banking, healthcare, and infrastructure, as automation could create more efficient systems. A necessary component of the smarter planet concept is to ensure that these new systems have very high reliability. Even extremely rare reliability problems can easily escalate to problematic scenarios when implemented at very large scales. For life-critical systems, such as automobiles, infrastructure, medical implantables, and avionic systems, unmitigated failures could be dangerous. As more automation moves into these types of critical systems, reliability failures will need to be managed. As computer automation continues to increase in our society, the need for greater radiation reliability is necessary. Already critical infrastructure is failing too frequently. In this paper, we will introduce the Cross-Layer Reliability concept for designing more reliable computer systems.

43 PARTICLE ACCELERATORS↗

SU-E-T-03: 3D GPU-Accelerated Secondary Checks of Radiation Therapy Treatment Plans

Purpose: Redundant treatment verifications in conformal and intensity-modulated radiation therapy techniques are traditionally performed with single point calculations. New solutions can replace these checks with 3D treatment plan verifications. This work describes a software tool (Mobius3D, Mobius Medical Systems) that uses a GPU-accelerated collapsed cone algorithm to perform 3D independent verifications of TPS calculations. Methods: Mobius3D comes with reference beam models for common linear accelerators. The system uses an independently developed collapsed cone algorithm updated with recent enhancements. 144 isotropically-spaced cones are used for each voxel for calculations. These complex calculations can be sped up by using GPUs. Mobius3D calculate dose using DICOM information coming from TPS (CT, RT Struct, RT Plan RT Dose). DVH-metrics and 3D gamma tests can be used to compare both TPS and secondary calculations. 170 patients treated with all common techniques as 3DCFRT (including wedged), static and dynamic IMRT and VMAT have been successfully verified with this solution. Results: Calculation times are between 3–5 minutes for 3DCFRT treatments and 15–20 for most complex dMLC and VMAT plans. For all PTVs mean dose and 90% coverage differences are (1.12±0.97)% and (0.68±1.19)%, respectively. Mean dose discrepancies for all OARs is (0.64±1.00)%. 3D gamma (global, 3%/3 mm) analysis shows a mean passing rate of (97.8 ± 3.0)% for PTVs and (99.0±3.0)% for OARs. 3D gamma pasing rate for all voxels in CT has a mean value of (98.5±1.6)%. Conclusion: Mobius3D is a powerful tool to verify all modalities of radiation therapy treatments. Dose discrepancies calculated by this system are in good agreement with TPS. The use of reference beam data results in time savings and can be used to avoid the propagation of errors in original beam data into our QA system. GPU calculations permit enhanced collapsed cone calculations with reasonable calculation times.

60 APPLIED LIFE SCIENCES↗

Effect of altered gravity on temperature regulation in mammals: Investigation of gravity effect on temperature regulation in mammals

Male, Long-Evans hooded rats were instrumented for monitoring core and hypothalamic temperatures as well as shivering and nonshivering thermogenesis in response to decreased ambient temperature in order to characterize the nature of the neural controller of temperature in rats at 1G and evaluate chronic implantation techniques for the monitoring of appropriate parameters at hypergravic fields. The thermoregulatory responses of cold-exposed rats at 2G were compared to those at 1G. A computer model was developed to simulate the thermoregulatory system in the rat. Observations at 1 and 2G were extended to acceleration fields of 1.5, 3.0 and 4.0G and the computer model was modified for application to altered gravity conditions. Changes in the acceleration field resulted in inadequate heat generation rather than increased heat loss. Acceleration appears to impair the ability of the neurocontroller to appropriately integrate input signals for body temperature maintenance.

Horwitz, B. A.↗

SU-E-T-210: Independent MU Dose Calculation Software for S and S IMRT Using Modified Clarkson Integration Sector

Purpose: The purpose of this work was to develop and validate an open source independent MU dose calculation software for S and S IMRT based in the algorithm proposed by Kung et.al. Methods: Treatment plans were done using Iplan v4.5 BrainLAB TPS and S and S IMRT modality. A 6MV photon beam produced by a Primus linear accelerator equipped with an Optifocus MLC was used. TPS dose calculation algorithms were pencil beam and Monte Carlo. 230 IMRT treatments plans were selected for the study. The software was written under MALTLAB environment. Treatment plans were imported by the software using RTP format. Field fluences were reconstructed adding all segments.The algorithm implemented in the software calculates the dose at a reference point as the sum of primary and scatter dose. The primary dose is obtained by masking the fluence map with a circle of radius 1cm. The scatter dose is obtained through a shaped ring mask around the previous circle with a thickness of 0.5cm; the rings are increased one after another with constant thickness until cover the entire map of influence. The dosimetric parameters Sc, Sp and TPR vary depending on radio, the transmission effect of the MLC, inverse square law and dose profile are used for the calculation. Results: The average difference between measured and independent calculated dose was 0.4% ± 2.2% [−6.8%, 6.4%]. For 91% of the studied plans the difference was less than 3%. The difference between the measured and TPS dose with pencilbeam algorithm was 2.6% ± 1.41% [−2.0%, 5.6%] and Monte Carlo algorithm was 0.4% ± 1.5% [−4.9%, 3.7%]. The differences obtained are comparable to that obtained with the ionization chamber and TPS. Conclusion: The developed software is suitable for use in S and S IMRT dose calculation. This application is open and can be downloading under request.

07 ISOTOPE AND RADIATION SOURCES↗

Modeling laser wakefield accelerator experiments with ultrafast particle-in-cell simulations in boosted frames

The development of new laser systems at the 10 Petawatt range will push laser wakefield accelerators to novel regimes, for which theoretical scalings predict the possibility to accelerate electron bunches up to tens of GeVs in meter-scale plasmas. Numerical simulations will play a crucial role in testing, probing, and optimizing the physical parameters and the setup of future experiments. Fully kinetic simulations are computationally very demanding, pushing the limits of today's supercomputers. In this paper, the recent developments in the OSIRIS framework [R. A. Fonseca et al., Lect. Notes Comput. Sci. 2331, 342 (2002)] are described, in particular the boosted frame scheme, which leads to a dramatic change in the computational resources required to model laser wakefield accelerators. Results from one-to-one modeling of the next generation of laser systems are discussed, including the confirmation of electron bunch acceleration to the energy frontier.

43 PARTICLE ACCELERATORS↗

COMPILE: a GWAS computational pipeline for gene discovery in complex genomes

Abstract Background Genome-Wide Association Studies (GWAS) are used to identify genes and alleles that contribute to quantitative traits in large and genetically diverse populations. However, traits with complex genetic architectures create an enormous computational load for discovery of candidate genes with acceptable statistical certainty. We developed a streamlined computational pipeline for GWAS (COMPILE) to accelerate identification and annotation of candidate maize genes associated with a quantitative trait, and then matches maize genes to their closest rice and Arabidopsis homologs by sequence similarity. Results COMPILE executed GWAS using a Mixed Linear Model that incorporated, without compression, recent advancements in population structure control, then linked significant Quantitative Trait Loci (QTL) to candidate genes and RNA regulatory elements contained in any genome. COMPILE was validated using published data to identify QTL associated with the traits of α-tocopherol biosynthesis and flowering time, and identified published candidate genes as well as additional genes and non-coding RNAs. We then applied COMPILE to 274 genotypes of the maize Goodman Association Panel to identify candidate loci contributing to resistance of maize stems to penetration by larvae of the European Corn Borer ( Ostrinia nubilalis ). Candidate genes included those that encode a gene of unknown function, WRKY and MYB-like transcriptional factors, receptor-kinase signaling, riboflavin synthesis, nucleotide-sugar interconversion, and prolyl hydroxylation. Expression of the gene of unknown function has been associated with pathogen stress in maize and in rice homologs closest in sequence identity. Conclusions The relative speed of data analysis using COMPILE allowed comparison of population size and compression. Limitations in population size and diversity are major constraints for a trait and are not overcome by increasing marker density. COMPILE is customizable and is readily adaptable for application to species with robust genomic and proteome databases.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Optimizing superconducting Nb film cavities by mitigating medium-field Q -slope through annealing

Niobium films are of interest in applications in various superconducting devices, such as superconducting radiofrequency cavities for particle accelerators and superconducting qubits for quantum computing. In this study, we address the persistent medium-field Q-slope issue in Nb film cavities, which, despite their high-quality factor at low RF fields, exhibit a significant Q-slope at medium RF fields compared to bulk Nb cavities. Traditional heat treatments, effective in reducing surface resistance and mitigating the Q-slope in bulk Nb cavities, are challenging for Nb-coated copper cavities. To overcome this challenge, we employed DC bias high-power impulse magnetron sputtering to deposit Nb film onto a 1.3 GHz single-cell elliptical bulk Nb cavity, followed by annealing treatments aimed at modifying the properties of the Nb film. In-situ annealing at 340 °C increased the quench field from 10.0 to 12.5 MV m −1 . Vacuum furnace annealing at 600 °C and 800 °C for 3 h resulted in a quench field increase of 13.5 and 15.3 MV m −1 , respectively. Further annealing at 800 °C for 6 h boosted the quench field to 17.5 MV m −1 . Additionally, the annealing treatments significantly reduced the field dependence of the surface resistance. However, increasing the annealing temperature to 900 °C induced a Q-switch phenomenon in the cavity. The analysis of RF performance and material characterization before and after annealing has provided critical insights into how the microstructure and impurity levels in Nb films influence the evolution of the Q-slope in Nb film cavities. Our findings highlight the significant roles of hydrides, high local misorientation, and lattice and surface defects in driving field-dependent losses. By strategically optimizing film properties and controlling impurity levels, we demonstrate a promising pathway to mitigate the medium-field Q-slope, paving the way for more efficient superconducting RF technologies.

Nb film↗

Bridging the Transition from Mesoscale to Microscale Turbulence in Numerical Weather Prediction Models

With a focus towards developing multiscale capabilities in numerical weather prediction models, the specific problem of the transition from the mesoscale to the microscale is investigated. For that purpose, idealized one-way nested mesoscale to large-eddy simulation (LES) experiments were carried out using the Weather Research and Forecasting model framework. It is demonstrated that switching from one-dimensional turbulent diffusion in the mesoscale model to three-dimensional LES mixing does not necessarily result in an instantaneous development of turbulence in the LES domain. However, very large fetches are needed for the natural transition to turbulence to occur. The computational burden imposed by these long fetches necessitates the development of methods to accelerate the generation of turbulence on a nested LES domain forced by a smooth mesoscale inflow. To that end, four new methods based upon finite amplitude perturbations of the potential temperature field along the LES inflow boundaries are developed, and investigated under convective conditions. Each method accelerated the development of turbulence within the LES domain, with two of the methods resulting in a rapid generation of production and inertial range energy content associated to microscales that is consistent with non-nested simulations using periodic boundary conditions. The cell perturbation approach, the simplest and most efficient of the best performing methods, was investigated further under neutral and stable conditions. Successful results were obtained in all the regimes, where satisfactory agreement of mean velocity, variances and turbulent fluxes, as well as velocity and temperature spectra, was achieved with reference non-nested simulations. In contrast, the non-perturbed LES solution exhibited important energy deficits associated to a delayed establishment of fully-developed turbulence. The cell perturbation method has negligible computational cost, significantly accelerates the generation of realistic turbulence, and requires minimal parameter tuning, with the necessary information relatable to mean inflow conditions provided by the mesoscale solution.

17 WIND ENERGY↗

Pandemic drugs at pandemic speed: infrastructure for accelerating COVID-19 drug discovery with hybrid machine learning- and physics-based simulations on high-performance computers

The race to meet the challenges of the global pandemic has served as a reminder that the existing drug discovery process is expensive, inefficient and slow. There is a major bottleneck screening the vast number of potential small molecules to shortlist lead compounds for antiviral drug development. New opportunities to accelerate drug discovery lie at the interface between machine learning methods, in this case, developed for linear accelerators, and physics-based methods. The two in silico methods, each have their own advantages and limitations which, interestingly, complement each other. Here, we present an innovative infrastructural development that combines both approaches to accelerate drug discovery. The scale of the potential resulting workflow is such that it is dependent on supercomputing to achieve extremely high throughput. We have demonstrated the viability of this workflow for the study of inhibitors for four COVID-19 target proteins and our ability to perform the required large-scale calculations to identify lead antiviral compounds through repurposing on a variety of supercomputers.

97 MATHEMATICS AND COMPUTING↗

Automating linear accelerator quality assurance

Purpose: The purpose of this study was 2-fold. One purpose was to develop an automated, streamlined quality assurance (QA) program for use by multiple centers. The second purpose was to evaluate machine performance over time for multiple centers using linear accelerator (Linac) log files and electronic portal images. The authors sought to evaluate variations in Linac performance to establish as a reference for other centers. Methods: The authors developed analytical software tools for a QA program using both log files and electronic portal imaging device (EPID) measurements. The first tool is a general analysis tool which can read and visually represent data in the log file. This tool, which can be used to automatically analyze patient treatment or QA log files, examines the files for Linac deviations which exceed thresholds. The second set of tools consists of a test suite of QA fields, a standard phantom, and software to collect information from the log files on deviations from the expected values. The test suite was designed to focus on the mechanical tests of the Linac to include jaw, MLC, and collimator positions during static, IMRT, and volumetric modulated arc therapy delivery. A consortium of eight institutions delivered the test suite at monthly or weekly intervals on each Linac using a standard phantom. The behavior of various components was analyzed for eight TrueBeam Linacs. Results: For the EPID and trajectory log file analysis, all observed deviations which exceeded established thresholds for Linac behavior resulted in a beam hold off. In the absence of an interlock-triggering event, the maximum observed log file deviations between the expected and actual component positions (such as MLC leaves) varied from less than 1% to 26% of published tolerance thresholds. The maximum and standard deviations of the variations due to gantry sag, collimator angle, jaw position, and MLC positions are presented. Gantry sag among Linacs was 0.336 ± 0.072 mm. The standard deviation in MLC position, as determined by EPID measurements, across the consortium was 0.33 mm for IMRT fields. With respect to the log files, the deviations between expected and actual positions for parameters were small (<0.12 mm) for all Linacs. Considering both log files and EPID measurements, all parameters were well within published tolerance values. Variations in collimator angle, MLC position, and gantry sag were also evaluated for all Linacs. Conclusions: The performance of the TrueBeam Linac model was shown to be consistent based on automated analysis of trajectory log files and EPID images acquired during delivery of a standardized test suite. The results can be compared directly to tolerance thresholds. In addition, sharing of results from standard tests across institutions can facilitate the identification of QA process and Linac changes. These reference values are presented along with the standard deviation for common tests so that the test suite can be used by other centers to evaluate their Linac performance against those in this consortium.

60 APPLIED LIFE SCIENCES↗

Pipelining RDP Data to the "Taxomatic" (Final Report)

This project builds on the results of previously funded research by integrating data and software that had been previously used in building resources used in the preparation of Bergey's Manual of Systematic Bacteriology, 2 nd Edition (Volumes 1 & 2A-C) and the Ribosomal Database Project-II (RDP-II) so as to both enhance the value of the data and create a pipeline approach to keeping the data current. Earlier, we demonstrated the value of using exploratory data analysis (EDA) to visualize large sets of sequence data (notably SSU rRNA gene sequences used in constructing a comprehensive phylogeny of prokaryotes. While the Self- Organizing Self-Correcting Classification (SOSCC) algorithms we developed were computationally efficient and useful for unraveling problems within the underlying data (e.g., identification of annotation errors, detection of unresolved synonymies, taxonomic and nomenclatural errors), bottlenecks at the preprocessing stage limited deployment of our applications as tools for end-users. To overcome the bottlenecks (which included hand alignment and computation of large matrices of pair-wise evolutionary distances), we proposed building a data pipeline between the "Taxomatic" application and RDP-II. The objectives were to accelerate the production and distribution of the updated versions of the prokaryotic taxonomy in lock-step with publication of new taxa and rearrangement of existing taxa, and to distribute these data more readily with the RDP-II and and other stakeholders in the community. A related goal of the current project is to deploy our visualization techniques as an interactive web application by which end-users can view manipulate, and select datasets of particular interest based upon phylogenetic and genomic information, access sequence data, and ultimately the scientific literature where the original observations were made and those that build on the original observations. The Taxomatic is a web-based tool to visualize distance matrices. The tool accepts raw distance matrices or aligned sequence information as data sources. When sequence information is provided the distance matrix is computed using the uncorrected distance model. Users can upload files to the Taxomatic website or sequences can be submitted by a SOAP service. This SOAP service is used by RDP to streamline Taxomatic use with RDP data. In addition to supplying source information, users can either supply their own taxonomic information by uploading it in XML, retrieve data taxonomic information from the RDP using either RDP or Genbank identifiers as source data, with or without classification by the RDP Classifier web service, or completely omit taxonomic data. In the latter case, the input distance matrix can be viewed in the order in which it was loaded.

59 BASIC BIOLOGICAL SCIENCES↗

Magnetohydrodynamics with physics informed neural operators

Abstract The modeling of multi-scale and multi-physics complex systems typically involves the use of scientific software that can optimally leverage extreme scale computing. Despite major developments in recent years, these simulations continue to be computationally intensive and time consuming. Here we explore the use of AI to accelerate the modeling of complex systems at a fraction of the computational cost of classical methods, and present the first application of physics informed neural operators (NOs) (PINOs) to model 2D incompressible magnetohydrodynamics (MHD) simulations. Our AI models incorporate tensor Fourier NOs as their backbone, which we implemented with the TensorLY package. Our results indicate that PINOs can accurately capture the physics of MHD simulations that describe laminar flows with Reynolds numbers Re ⩽ 250 . We also explore the applicability of our AI surrogates for turbulent flows, and discuss a variety of methodologies that may be incorporated in future work to create AI models that provide a computationally efficient and high fidelity description of MHD simulations for a broad range of Reynolds numbers. The scientific software developed in this project is released with this manuscript.

97 MATHEMATICS AND COMPUTING↗

Achieving performance portability in Gaussian basis set density functional theory on accelerator based architectures in NWChemEx

The numerical integration of the exchange–correlation (XC) potential is one of the primary computational bottlenecks in Gaussian basis set Kohn–Sham density functional theory (KS-DFT). To achieve optimal performance and accuracy, care must be taken in this numerical integration to preserve local sparsity as to allow for near linear weak scaling with system size. This leads to an integration scheme with several performance critical kernels which must be hand optimized for each architecture of interest. As the set of available accelerator hardware goes more diverse, a key challenge for developers of KS-DFT software is to maintain performance portability across a wide range of computational architectures. In this article, we examine a modular software design pattern which decouples the implementation details of performance critical kernels from the expression of high-level algorithmic workflows in a device-agnostic language such as C++; thus allowing for developers to target existing and emerging accelerator hardware within a single code base. We consider the efficacy of such a design pattern in the numerical integration of the XC potential by demonstrating its ability to achieve performance portability across a set of accelerator architectures which are representative of those on current and future U.S. Department of Energy Leadership Computing Facilities.

97 MATHEMATICS AND COMPUTING↗

Comparing current cluster, massively parallel, and accelerated systems

Currently there is large architectural diversity in high perfonnance computing systems. They include 'commodity' cluster systems that optimize per-node performance for small jobs, massively parallel processors (MPPs) that optimize aggregate perfonnance for large jobs, and accelerated systems that optimize both per-node and aggregate performance but only for applications custom-designed to take advantage of such systems. Because of these dissimilarities, meaningful comparisons of achievable performance are not straightforward. In this work we utilize a methodology that combines both empirical analysis and performance modeling to compare clusters (represented by a 4,352-core IB cluster), MPPs (represented by a 147,456-core BG/P), and accelerated systems (represented by the 129,600-core Roadrunner) across a workload of four applications. Strengths of our approach include the ability to compare architectures - as opposed to specific implementations of an architecture - attribute each application's performance bottlenecks to characteristics unique to each system, and to explore performance scenarios in advance of their availability for measurement. Our analysis illustrates that application performance is essentially unrelated to relative peak performance but that application performance can be both predicted and explained using modeling.

97 MATHEMATICS AND COMPUTING↗