Search NASA⌕ Search

SEARCH · Search NASA

Results for “computer system benchmarking”

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 271 records · Page 15

Enhanced geometric capabilities for the transient analysis code T-ReX and its application to simulating TREAT experiments

Here, advances in computational architecture have prompted a resurgence in the simulation of reactor transients from first principles. Most codes are unable to simulate transient events with complex models, and require numerous approximations. The code T-ReX (Transient-Reactor eXperiment simulator), an extensive update to TDKENO, has been developed as a transient analysis tool with few geometric limitations and minimal theoretical approximations. T-ReX achieves this by employing the Improved Quasi-Static (IQS) method to solve the time dependent Boltzmann transport equation with explicit representation of delayed neutrons. The primary change in T-ReX relative to TDKENO is the incorporation of a modified version of the Monte Carlo code KENO-VI to calculate the flux shape and model the geometry of a problem. Using KENO-VI to model systems allows exact representation of the geometry. The changes to T-ReX are verified by comparison of solutions to computational benchmark problems found with a previous version of TDKENO that made use of KENO V.a, and several other codes with time-dependent capabilities. In addition, a three-dimensional model of the Transient Reactor Test Facility (TREAT) core at the Idaho National Laboratory (INL) is constructed with KENO-VI, and used to validate T ReX. T-ReX produces results that agree with benchmark problems, and are in better agreement with TREAT experimental data than TDKENO.

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗

Robust inverse scattering full waveform seismic tomography for imaging complex structure

Seismic tomography becomes important tool recently for imaging complex subsurface. It is well known that imaging complex rich fault zone is difficult. In this paper, The application of time domain inverse scattering wave tomography to image the complex fault zone would be shown on this paper, especially an efficient time domain inverse scattering tomography and their run in cluster parallel computer which has been developed. This algorithm is purely based on scattering theory through solving Lippmann Schwienger integral by using Born's approximation. In this paper, it is shown the robustness of this algorithm especially in avoiding the inversion trapped in local minimum to reach global minimum. A large data are solved by windowing and blocking technique of memory as well as computation. Parameter of windowing computation is based on shot gather's aperture. This windowing technique reduces memory as well as computation significantly. This parallel algorithm is done by means cluster system of 120 processors from 20 nodes of AMD Phenom II. Benchmarking of this algorithm is done by means Marmoussi model which can be representative of complex rich fault area. It is shown that the proposed method can image clearly the rich fault and complex zone in Marmoussi model even though the initial model is quite far from the true model. Therefore, this method can be as one of solution to image the very complex mode.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Taming the virtual space for incremental full configuration interaction

Incremental full configuration interaction (iFCI) closely approximates the FCI limit with polynomial cost through a many-body expansion of the correlation energy, providing highly accurate total energies within a given basis set. To extend iFCI beyond previous basis set limitations, this work introduces a novel natural orbital (NO) screening approach, incremental NO full configuration interaction (iNO-FCI). By consideration of the importance of virtual orbital selection in the convergence of iFCI, iNO-FCI maximizes the consistency between orbitals selected for each correlated body. iNO-FCI employs a principle of cancellation of errors and ensures that the same set of virtual NOs is used for interdependent terms. Here, this strategy significantly reduces computational cost without compromising precision. Computational savings of up to 95% are demonstrated, allowing access to larger basis sets that were previously computationally prohibitive. iNO-FCI is herein introduced and benchmarked for several difficult test cases involving double-bond dissociation, biradical systems, conjugated π systems, and the spin gap of a Cu-based transition metal complex.

Correlation energy↗

Tackling the Challenges in Scene Graph Generation With Local-to-Global Interactions

In this work, we seek new insights into the underlying challenges of the scene graph generation (SGG) task. Quantitative and qualitative analysis of the visual genome (VG) dataset implies: 1) ambiguity: even if interobject relationship contains the same object (or predicate), they may not be visually or semantically similar; 2) asymmetry: despite the nature of the relationship that embodied the direction, it was not well addressed in previous studies; and 3) higher-order contexts: leveraging the identities of certain graph elements can help generate accurate scene graphs. Motivated by the analysis, we design a novel SGG framework, Local-to-global interaction networks (LOGINs). Locally, interactions extract the essence between three instances of subject, object, and background, while baking direction awareness into the network by explicitly constraining the input order of subject and object. Globally, interactions encode the contexts between every graph component (i.e., nodes and edges). Finally, Attract and Repel loss is utilized to fine-tune the distribution of predicate embeddings. By design, our framework enables predicting the scene graph in a bottom-up manner, leveraging the possible complementariness. To quantify how much LOGIN is aware of relational direction, a new diagnostic task called Bidirectional Relationship Classification (BRC) is also proposed. Overall, experimental results demonstrate that LOGIN can successfully distinguish relational direction than existing methods (in BRC task), while showing state-of-the-art results on the VG benchmark (in SGG task).

97 MATHEMATICS AND COMPUTING↗

EqualWrites: Reducing Intra-set Write Variations for Enhancing Lifetime of Non-volatile Caches

Driven by the trends of increasing core-count and bandwidth-wall problem, the size of last level caches (LLCs) has greatly increased and hence, the researchers have explored non-volatile memories (NVMs) which provide high density and consume low-leakage power. Since NVMs have low write-endurance and the existing cache management policies are write variation-unaware, effective wear-leveling techniques are required for achieving reasonable cache lifetimes using NVMs. We present EqualWrites, a technique for mitigating intra-set write variation. In this paper, our technique works by recording the number of writes on a block and changing the cache-block location of a hot data-item to redirect the future writes to a cold block to achieve wear-leveling. Simulation experiments have been performed using an x86-64 simulator and benchmarks from SPEC06 and HPC (high-performance computing) field. The results show that for single, dual and quad-core system configurations, EqualWrites improves cache lifetime by 6.31X, 8.74X and 10.54X, respectively. In addition, its implementation overhead is very small and it provides larger improvement in lifetime than three other intra-set wear-leveling techniques and a cache replacement policy.

97 MATHEMATICS AND COMPUTING↗

Heterogeneous Multi-Domain Dataset Synthesis to Facilitate Privacy and Risk Assessments in Smart City IoT

The emergence of the Smart Cities paradigm and the rapid expansion and integration of Internet of Things (IoT) technologies within this context have created unprecedented opportunities for high-resolution behavioral analytics, urban optimization, and context-aware services. However, this same proliferation intensifies privacy risks, particularly those arising from cross-modal data linkage across heterogeneous sensing platforms. To address these challenges, this paper introduces a comprehensive, statistically grounded framework for generating synthetic, multimodal IoT datasets tailored to Smart City research. The framework produces behaviorally plausible synthetic data suitable for preliminary privacy risk assessment and as a benchmark for future re-identification studies, as well as for evaluating algorithms in mobility modeling, urban informatics, and privacy-enhancing technologies. As part of our approach, we formalize probabilistic methods for synthesizing three heterogeneous and operationally relevant data streams—cellular mobility traces, payment terminal transaction logs, and Smart Retail nutrition records—capturing the behaviors of a large number of synthetically generated urban residents over a 12-week period. The framework integrates spatially explicit merchant selection using K-Dimensional (KD)-tree nearest-neighbor algorithms, temporally correlated anchor-based mobility simulation reflective of daily urban rhythms, and dietary-constraint filtering to preserve ecological validity in consumption patterns. In total, the system generates approximately 116 million mobility pings, 5.4 million transactions, and 1.9 million itemized purchases, yielding a reproducible benchmark for evaluating multimodal analytics, privacy-preserving computation, and secure IoT data-sharing protocols. To show the validity of this dataset, the underlying distributions of these residents were successfully validated against reported distributions in published research. We present preliminary uniqueness and cross-modal linkage indicators; comprehensive re-identification benchmarking against specific attack algorithms is planned as future work. This framework can be easily adapted to various scenarios of interest in Smart Cities and other IoT applications. By aligning methodological rigor with the operational needs of Smart City ecosystems, this work fills critical gaps in synthetic data generation for privacy-sensitive domains, including intelligent transportation systems, urban health informatics, and next-generation digital commerce infrastructures.

IoT↗

Neutron Number Probability Distributions in a Subcritical System Using the Forward Master Equation

The subject of this paper is the determination of the equilibrium probability distribution function (PDF) of the neutron population within a subcritical multiplying system in the presence of an intrinsic random source as a result of spontaneous fission (SF). Such systems are strongly stochastic and low order moments such as the mean and variance of the neutron number are not sufficient to characterize the true state of the neutron population. We base our work on the forward Master equation in a lumped model formulation for P{sub n} (t), n = 0, 1, 2,..., the probability of there existing η neutrons at time t and obtain solutions for the steady state distributions. The key parameters in this model, neutron lifetime and fission probability, are rigorously obtained by numerically solving the k-eigenvalue problem in a one-dimensional spherical system and using standard definitions of lumped parameters in terms of production and loss rates in finite systems. Previous work has examined single-chain PDFs (a single fission and the resulting progeny) for neutrons, uncoupled neutron-photon single-chain PDFs, both using a lumped backward Master equation formulation, as well as neutron PDFs in supercritical systems. In this paper, we analyze strictly subcritical systems, for which we compare analytically derived solutions with numerical computations for different multiplicity distribution models. For subcritical systems where the neutron population remains finite at steady state, we show that the forward Master equation can be very efficiently solved numerically to obtain the equilibrium distributions. Finally, this work is motivated by potential application to passive interrogation for detection of diverted nuclear materials as well to criticality safety. The equilibrium neutron number probability distribution in subcritical systems has been numerically investigated for several models of the induced and spontaneous fission multiplicity distribution. A well known approximate but closed form solution, namely the quadratic model, was shown to be unacceptably accurate for subcritical media. The matrix inversion and the recursion methods agree well and were benchmarked with the binary fission model, but the recursion formulation proves far more computationally efficient. A qualitative change in the shape of the neutron PDF is observed for increasing source strength, as expected, and a transition from a stochastic to a deterministic system is expected for a sufficiently strong source. We are currently extending this approach to obtain the equilibrium joint distribution of neutrons within the system and those that have leaked, to explicitly represent the neutron PDF as observed by an external detector. (authors)

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Radiation Detection Computational Benchmark Scenarios

Modeling forms an important component of radiation detection development, allowing for testing of new detector designs, evaluation of existing equipment against a wide variety of potential threat sources, and assessing operation performance of radiation detection systems. This can, however, result in large and complex scenarios which are time consuming to model. A variety of approaches to radiation transport modeling exist with complementary strengths and weaknesses for different problems. This variety of approaches, and the development of promising new tools (such as ORNL’s ADVANTG) which combine benefits of multiple approaches, illustrates the need for a means of evaluating or comparing different techniques for radiation detection problems. This report presents a set of 9 benchmark problems for comparing different types of radiation transport calculations, identifying appropriate tools for classes of problems, and testing and guiding the development of new methods. The benchmarks were drawn primarily from existing or previous calculations with a preference for scenarios which include experimental data, or otherwise have results with a high level of confidence, are non-sensitive, and represent problem sets of interest to NA-22. From a technical perspective, the benchmarks were chosen to span a range of difficulty and to include gamma transport, neutron transport, or both and represent different important physical processes and a range of sensitivity to angular or energy fidelity. Following benchmark identification, existing information about geometry, measurements, and previous calculations were assembled. Monte Carlo results (MCNP decks) were reviewed or created and re-run in order to attain accurate computational times and to verify agreement with experimental data, when present. Benchmark information was then conveyed to ORNL in order to guide testing and development of hybrid calculations. The results of those ADVANTG calculations were then sent to PNNL for compilation. This is a report describing the details of the selected Benchmarks and results from various transport codes.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

An HPC benchmark survey and taxonomy for characterization

The field of High-Performance Computing (HPC) is defined by providing computing devices with highest performance for a variety of demanding scientific users. The tight co-design relationship between HPC providers and users propels the field forward, paired with technological improvements, achieving continuously higher performance and resource utilization. A key device for system architects, architecture researchers, and scientific users are benchmarks, allowing for well-defined assessment of hardware, software, and algorithms. Many benchmarks exist in the community, from individual niche benchmarks testing specific features, to large-scale benchmark suites for whole procurements. We survey the available HPC benchmarks, summarizing them in table form with key details and concise categorization, also through an interactive website. For categorization, we present a benchmark taxonomy for well-defined characterization of benchmarks.

Benchmarking↗

A comparison of global optimization algorithms with standard benchmark functions and real-world applications using Energy Plus

There is an increasing interest in the use of computer algorithms to identify combinations of parameters which optimise the energy performance of buildings. For such problems, the objective function can be multi-modal and needs to be approximated numerically using building energy simulation programs. As these programs contain iterative solution algorithms, they introduce discontinuities in the numerical approximation to the objective function. Metaheuristics often work well for such problems, but their convergence to a global optimum cannot be established formally. Moreover, different algorithms tend to be suited to particular classes of optimization problems. To shed light on this issue we compared the performance of two metaheuristics, the hybrid CMA-ES/HDE and the hybrid PSO/HJ, in minimizing standard benchmark functions and real-world building energy optimization problems of varying complexity. From this we find that the CMA-ES/HDE performs well on more complex objective functions, but that the PSO/HJ more consistently identifies the global minimum for simpler objective functions. Both identified similar values in the objective functions arising from energy simulations, but with different combinations of model parameters. This may suggest that the objective function is multi-modal. The algorithms also correctly identified some non-intuitive parameter combinations that were caused by a simplified control sequence of the building energy system that does not represent actual practice, further reinforcing their utility.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Criticality Accident Alarm System Shielding Benchmark: Integral Experiment Request 498, Critical Engineering Decision 1 Report

This report supports Critical/Subcritical Experiment Decision 1 (CED-1) for Integral Experiment Request (IER) 498, a shielding benchmark produced using the Godiva IV critical assembly at the National Critical Experiments Research Center (NCERC). Shielding benchmarks are datasets collected to provide a comparison for computer models that simulate the effect of shielding on radiation transmission from a source to a detector. The purpose of a CED-1 report is to produce a conceptual design for an integral experiment and to identify the long lead-time items. The design presented in this report favors certain detailed design options such as material selections. However, alternative options can be selected as the project evolves. Dimensions are specified only as needed to evaluate feasibility and performance. Development of Criticality Accident Alarm System (CAAS) shielding benchmarks supports two main programmatic needs. The first need is to improve confidence in models of facilities that house nuclear material handling operations. Better models can support effective CAAS detector placement when criticality may result in dose, and it enables more accurate analysis of criticality accident impacts at various locations in a facility. A key quantity that can be verified is the secondary gamma production caused by neutron interactions with various materials. The second programmatic need supported by shielding benchmarks is the ongoing development of methods being integrated into nuclear simulation codes such as the Monte Carlo N-Particle (MCNP) transport code and the SCALE code system used in efforts funded by the Nuclear Criticality Safety Program (NCSP). Previous work performed under the International Criticality Safety Benchmark Experimental Program (ICSBEP) to produce shielding benchmark data is primarily limited to a set of measurements at the SILENE critical assembly at the Valduc facility in France. The experiment used an open configuration and a burst mode operation, which limited detector options to thermoluminescent dosimeters (TLDs) and activation foils. Because TLDs do not record gamma spectra, it was difficult to verify the gamma energy spectrum from the experiment. In this work, the Godiva IV assembly will be used because it is a fast system capable of producing a wide range of source strengths, including super-prompt critical bursts. The room return for the facility is expected to be substantial and challenging to characterize, so a more closed, shielded experimental configuration is considered. Furthermore, because the physics of radiation transport outside the assembly does not change between the burst mode and the delayed critical mode, it is anticipated that most measurements will be made with the Godiva IV assembly in delayed critical mode, operating within a range that will allow electronic counters to collect accurate data. Scoping calculations concluded that a workable room return shield is comprised of 40 cm of 5% borated polyethylene backed by 7 cm of lead. This configuration was shown to produce a ratio of signal–to–room return of over 100 for up to 2 mean free paths of shield sample material, and over 10 for up to 4 mean free paths of shield sample material, ensuring that useful results can be obtained. With no shielding sample in the room return shield, it was shown that the neutron and gamma spectra do not appreciably change with a change in room size. This scoping calculation does not preclude the use of tungsten powder in a binder, some other effective gamma shielding material, or a hybrid design. Dose rates for various source intensities at an unshielded detector are listed herein to inform detector selection during CED-2, and calibration chains are identified. Preliminary estimates of costs and schedules are outlined but remain to be verified. Values could double or triple as estimates become more accurate and are principally included here to demonstrate feasibility. Preliminary estimates indicate that the apparatus can be installed within a feasible schedule and budget. Two measurements per day are assumed. Producing a shielding benchmark at NCERC using the Godiva IV assembly has been shown to be feasible, with manageable uncertainties and risks. A clear, continuing need for shielding benchmark data remains present, and this report identifies and evaluates a method for creating integrated shielding benchmark datasets.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Dynamic signatures of electronically nonadiabatic coupling in sodium hydride: a rigorous test for the symmetric quasi-classical model applied to realistic, ab initio electronic states in the adiabatic representation

Sodium hydride (NaH) in the gas phase presents a seemingly simple electronic structure making it a potentially tractable system for the detailed investigation of nonadiabatic molecular dynamics from both computational and experimental standpoints. The single vibrational degree of freedom, as well as the strong nonadiabatic coupling that arises from the excited electronic states taking on considerable ionic character, provides a realistic chemical system to test the accuracy of quasi-classical methods to model population dynamics where the results are directly comparable against quantum mechanical benchmarks. Here, using a simulated pump–probe type experiment, this work presents computational predictions of population transfer through the avoided crossings of NaH via symmetric quasi-classical Meyer–Miller (SQC/MM), Ehrenfest, and exact quantum dynamics on realistic, ab initio potential energy surfaces. The main driving force for population transfer arises from the ground vibrational level of the D 1 Σ + adiabatic state that is embedded in the manifold of near-dissociation C 1 Σ + vibrational states. When coupled through a sharply localized first-order derivative coupling most of the population transfers between t = 15 and t = 30 fs depending on the initially excited vibronic wavepacket. While quantum mechanical effects are expected due to the reduced mass of NaH, predictions of the population dynamics from both the SQC/MM and Ehrenfest models perform remarkably well against the quantum dynamics benchmark. Additionally, an analysis of the vibronic structure in the nonadiabatically coupled regime is presented using a variational eigensolver methodology.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Anderson acceleration with approximate calculations: Applications to scientific computing

Here we provide rigorous theoretical bounds for Anderson acceleration (AA) that allow for approximate calculations when applied to solve linear problems. We show that, when the approximate calculations satisfy the provided error bounds, the convergence of AA is maintained while the computational time could be reduced. We also provide computable heuristic quantities, guided by the theoretical error bounds, which can be used to automate the tuning of accuracy while performing approximate calculations. For linear problems, the use of heuristics to monitor the error introduced by approximate calculations, combined with the check on monotonicity of the residual, ensures the convergence of the numerical scheme within a prescribed residual tolerance. Motivated by the theoretical studies, we propose a reduced variant of AA, which consists in projecting the least-squares used to compute the Anderson mixing onto a subspace of reduced dimension. The dimensionality of this subspace adapts dynamically at each iteration as prescribed by the computable heuristic quantities. We numerically show and assess the performance of AA with approximate calculations on: (i) linear deterministic fixed-point iterations arising from the Richardson's scheme to solve linear systems with open-source benchmark matrices with various preconditioners and (ii) non-linear deterministic fixed-point iterations arising from non-linear time-dependent Boltzmann equations.

97 MATHEMATICS AND COMPUTING↗

Trinity Acceptance Tests Performance Summary.

Ensuring Real Applications perform well on Trinity is key to success. Four components: ASC applications, Sustained System Performance (SSP), Extra-Large MiniApplications problems, and Micro-benchmarks.

97 MATHEMATICS AND COMPUTING↗

Computational singular perturbation analysis of stochastic chemical systems with stiffness

Computational singular perturbation (CSP) is a useful method for analysis, reduction, and time integration of stiff ordinary differential equation systems. It has found dominant utility, in particular, in chemical reaction systems with a large range of time scales at continuum and deterministic level. On the other hand, CSP is not directly applicable to chemical reaction systems at micro or meso-scale, where stochasticity plays an non-negligible role and thus has to be taken into account. In this work we develop a novel stochastic computational singular perturbation (SCSP) analysis and time integration framework, and associated algorithm, that can be used to not only construct accurately and efficiently the numerical solutions to stiff stochastic chemical reaction systems, but also analyze the dynamics of the reduced stochastic reaction systems. Furthermore, the algorithm is illustrated by an application to a benchmark stochastic differential equation model, and numerical experiments are carried out to demonstrate the effectiveness of the construction.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Construction of energy-stable projection-based reduced order models

Our paper aims to unify and extend several approaches for building stable projection-based reduced order models (ROMs) using the energy method and the concept of “energy-stability”. Attention is focused on linear time-invariant (LTI) systems. First, an approach for building energy stable Galerkin ROMs for linear hyperbolic or incompletely parabolic systems of partial differential equations (PDEs) using continuous projection is proposed. The key idea is to apply to the system a transformation induced by the Lyapunov function for the system, and to build the ROM in the transformed variables. The result of this procedure will be a ROM that is energy-stable for any choice of reduced basis. It is shown that, for many PDE systems, the desired transformation is induced by a special inner product, termed the “symmetry inner product”. Next, attention is turned to building energy-stable ROMs via discrete projection. A discrete counterpart of the continuous symmetry inner product, termed the “Lyapunov inner product”, is derived. Moreover, it is shown that the Lyapunov inner product can be computed in a black-box fashion for a stable LTI system ari sing from the discretization of a system of PDEs in space. Projection in this inner product guarantees a ROM that is energy-stable, again for any choice of reduced basis. Connections between the Lyapunov inner product and the inner product induced by the balanced truncation algorithm are made. We also made comparisons between the symmetry inner product and the Lyapunov inner product. Performance of ROMs constructed using these inner products is evaluated on several benchmark test cases.

97 MATHEMATICS AND COMPUTING↗

Deploy Nalu/Kokkos algorithmic infrastructure with performance benchmarking.

The former Nalu interior heterogeneous algorithm design, which was originally designed to manage matrix assembly operations over all elemental topology types, has been modified to operate over homogeneous collections of mesh entities. This newly templated kernel design allows for removal of workset variable resize operations that were formerly required at each loop over a Sierra ToolKit (STK) bucket (nominally, 512 entities in size). Extensive usage of the Standard Template Library (STL) std::vector has been removed in favor of intrinsic Kokkos memory views. In this milestone effort, the transition to Kokkos as the underlying infrastructure to support performance and portability on many-core architectures has been deployed for key matrix algorithmic kernels. A unit-test driven design effort has developed a homogeneous entity algorithm that employs a team-based thread parallelism construct. The STK Single Instruction Multiple Data (SIMD) infrastructure is used to interleave data for improved vectorization. The collective algorithm design, which allows for concurrent threading and SIMD management, has been deployed for the core low-Mach element- based algorithm. Several tests to ascertain SIMD performance on Intel KNL and Haswell architectures have been carried out. The performance test matrix includes evaluation of both low- and higher-order methods. The higher-order low-Mach methodology builds on polynomial promotion of the core low-order control volume nite element method (CVFEM). Performance testing of the Kokkos-view/SIMD design indicates low-order matrix assembly kernel speed-up ranging between two and four times depending on mesh loading and node count. Better speedups are observed for higher-order meshes (currently only P=2 has been tested) especially on KNL. The increased workload per element on higher-order meshes bene ts from the wide SIMD width on KNL machines. Combining multiple threads with SIMD on KNL achieves a 4.6x speedup over the baseline, with assembly timings faster than that observed on Haswell architecture. The computational workload of higher-order meshes, therefore, seems ideally suited for the many-core architecture and justi es further exploration of higher-order on NGP platforms. A Trilinos/Tpetra-based multi-threaded GMRES preconditioned by symmetric Gauss Seidel (SGS) represents the core solver infrastructure for the low-Mach advection/diffusion implicit solves. The threaded solver stack has been tested on small problems on NREL's Peregrine system using the newly developed and deployed Kokkos-view/SIMD kernels. fforts are underway to deploy the Tpetra-based solver stack on NERSC Cori system to benchmark its performance at scale on KNL machines.

97 MATHEMATICS AND COMPUTING↗

Multi-GPU Implementation of a 3D Finite Difference Time Domain Earthquake Code on Heterogeneous Supercomputers

We have developed a highly scalable 3D Finite Difference GPU code for use in earthquake engineering and disaster management through regional petascale earthquake simulations. This MPI-CUDA code is based on a widely-used wave propagation code called AWP-ODC and restructured for high throughput and efficiency on a heterogeneous computing architecture. We present an effective communication reduction technique for leveraging GPUs with minimal PCI-e overhead, and a novel overlapping method to fully hide data communication latency between GPUs. The optimization concept used in this work can be extended to general stencil computing on a structured grid. The benchmarks demonstrated sustained 100 TFlops in single precision for 49 billion mesh points using 952 GPUs on the NCCS Titan Phase 5 system, which is a 77-fold speedup compared to the CPU version of the code. This multi-GPU implementation has been validated and used for a large-scale verification wave propagation simulation of Mw5.4 Chino Hills earthquake using 128 GPUs.

58 GEOSCIENCES↗