Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “heterogeneous benchmark”

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 73 records · Page 4

AI Benchmark Democratization and Carpentry

Benchmarks are a cornerstone of modern machine learning, enabling reproducibility, comparison, and scientific progress. However, AI benchmarks are increasingly complex, requiring dynamic, AI-focused workflows. Rapid evolution in model architectures, scale, datasets, and deployment contexts makes evaluation a moving target. Large language models often memorize static benchmarks, causing a gap between benchmark results and real-world performance. Beyond traditional static benchmarks, continuous adaptive benchmarking frameworks are needed to align scientific assessment with deployment risks. This calls for skills and education in AI Benchmark Carpentry. From our experience with MLCommons, educational initiatives, and programs like the DOE's Trillion Parameter Consortium, key barriers include high resource demands, limited access to specialized hardware, lack of benchmark design expertise, and uncertainty in relating results to application domains. Current benchmarks often emphasize peak performance on top-tier hardware, offering limited guidance for diverse, real-world scenarios. Benchmarking must become dynamic, incorporating evolving models, updated data, and heterogeneous platforms while maintaining transparency, reproducibility, and interpretability. Democratization requires both technical innovation and systematic education across levels, building sustained expertise in benchmark design and use. Benchmarks should support application-relevant comparisons, enabling informed, context-sensitive decisions. Dynamic, inclusive benchmarking will ensure evaluation keeps pace with AI evolution and supports responsible, reproducible, and accessible AI deployment. Community efforts can provide a foundation for AI Benchmark Carpentry.

von Laszewski, Gregor [Virginia U.]↗

Calculated and Measured Dose Distributions in an Anthropomorphic RANDO Phantom - Paper 105

Patient-specific quality assurance (QA) is a major component of the Intensity Modulated Radiation Therapy (IMRT) clinical process. The objective of this often time-intensive measurement process is to verify that the prescribed dose is accurately delivered to the target volume while sparing surrounding healthy tissue. The IMRT dose distributions can be measured using a wide variety of techniques and measurement devices including film (radiographic and radiochromic), ion chamber arrays, diode arrays, and Electronic Portal Imaging Devices (EPIDs). Of all the available modalities, film is one of the most versatile and most frequently utilized measurement systems. In rotation delivery techniques (such as volume modulated arc therapy and helical tomotherapy), composite dose distributions are obtained by irradiating a test phantom containing dosimeter(s). When film is utilized as the dosimeter, it can be placed in coronal, sagittal, or transverse planes inside the phantom. Because the film is irradiated using the rotational delivery sequence used to treat the patient, the dose given to specific anatomical structures (such as the parotids, spinal cord, bladder, rectum etc...) can be evaluated. Thomas et al (Med Phys 32: 3793-3800) recently reported the results of IMRT verifications for ten patients planned and treated with helical tomotherapy. This study reported that nearly 92% of the film pixels within the analyzed region of interest were within a gamma criterion of 2 mm and 2%. Like most tomotherapy users, Thomas et al used a test phantom made of homogeneous water equivalent material. Heterogeneous phantoms that simulate the different tissue densities in the human body do exist, but are infrequently used for patient-specific IMRT QA. The potential advantage of using heterogeneous phantoms is that the measured dose distributions will more adequately reflect the clinical dose distributions in the patients. The use of homogenous phantoms during IMRT quality assurance can potentially mask errors that arise from limitations in the treatment planning system's modeling of density corrections. The purpose of this study was to evaluate the dosimetric accuracy of a helical tomotherapy system using an anthropomorphic RANDO phantom (The Phantom Laboratory, Salem, NY) for patient-specific QA. The results of this study will allow a benchmark to be established for dosimetric accuracy in heterogeneous materials. Performing patient-specific QA is an integral part of the IMRT patient treatment process. Test phantoms are typically homogeneous, and do test the treatment planning systems ability to account for heterogeneity corrections. In this study, an anthropomorphic RANDO phantom was used to evaluate the doses calculated by a helical tomotherapy system. The dosimetric agreement between calculated and measured was worse than published results in homogenous phantoms, especially for target volumes located in the lung. Further investigation will be needed to determine if these errors are due to film response at heterogeneous tissue interfaces or from actual dose calculation errors. (authors)

61 RADIATION PROTECTION AND DOSIMETRY↗

Development Status of PROTEUS-MOC

The PROTEUS-MOC solver is intended to resolve the outstanding computational challenges that the PROTEUS-SN is limited by and bridge the gap between PROTEUS-SN and existing homogenized structured grid solvers. The primary push behind all of this work is to carry out coupled thermal-mechanical, fluid dynamics, and neutronics physics problems. The PROTEUS-MOC development has taken advantage of the combined experience with PROTEUS-SN and DeCART with the goal of producing a tool which combines the best features of both tools. Taking the DeCART tool as the model, it merges a subgroup cross section interpolation technique with a 2D method of characteristics approach and a 1D SN formulation. DeCART has been specifically demonstrated to be accurate on PWR and VHTR problems, but it has displayed several issues with regard to accuracy on thin axial mesh sizes which are attributable to the 2D1D scheme it is built upon. There are also notable problems with performance and scalability all of which we hope to resolve in PROTEUS-MOC. PROTEUS-MOC is at its core, based upon the success achieved with MOCFE which demonstrated excellent parallelization with respect to angle and good parallelization with respect to space. More recent efforts have indicated parallelization with respect to energy is also feasible. The work completed yielded a novel methodology where the axial basis functions are treated separately from the radial ones, but the geometry is extruded. This eliminates the need to carry out and store 3D ray tracing data and simultaneously allows the coarse axial node sizes typical with DeCART. To rapidly build a prototype, a standard power iteration algorithm was constructed using an unpreconditioned GMRES solution algorithm for solving the flux vector in each fission source iteration. The new PROTEUS-MOC solver was deployed on two test problems: Takeda and a heterogeneous cylindrical heavy water system. The Takeda benchmark problem is a standard homogeneous test problem which was previously found to be too difficult for the 3D MOCFE code to solve. Using the PROTEUS-MOC we were able to show space-angle convergence although full convergence requires further improvements in performance. The second benchmark problem was constructed more for future work than this year, but the results generated are promising indicating that it will be very useful in highlighting the weaknesses and strengths of the PROTEUS-MOC tool. While we are far from a production tool, the results included in this report give us confidence that we are on the right path to constructing an interceding code between the plethora of structured grid homogenized assembly solvers and the supercomputer consuming PROTEUS-SN code. With time, the creation of a suitable preconditioner along with typical optimization should reduce the performance shortcomings that are observable today.

Marin-Lafleche, A.↗

Prediction on Underestimation of Statistical Uncertainty in Monte Carlo Eigenvalue Calculation for Two-dimensional Heterogeneous Color Set Geometry

In a Monte Carlo calculation, statistical uncertainty is inevitably included in the result. The statistical uncertainty is commonly estimated by the central limit theorem without considering correlation among generations. The statistical uncertainty is usually expected to decrease at the inverse square root of the number of neutron histories. However, in a Monte Carlo eigenvalue calculation in a large geometry whose dominance ratio is close to unity, the central limit theorem would not be valid and thus the statistical uncertainty of fission source distribution can be significantly underestimated when inter-cycle correlation is not appropriately taken into account. The true statistical uncertainty can be calculated by statistical processing of the results obtained by multiple Monte Carlo runs using different initial random seeds, but this approach has a drawback of large calculation cost. Therefore, prediction of true statistical uncertainty by alternate approach is desirable. One of the solutions for this issue is an evaluation method of statistical uncertainty based on the eigenfunction expansion and the Autoregressive model (AR model). In this method, the true statistical uncertainty is theoretically derived by applying the AR model to behavior of expansion coefficients of eigenfunction in fission source distribution during cycles. Underestimation ratio of statistical uncertainty of fission source distribution can be predicted by this method. Note that the underestimation ratio of statistical uncertainty is defined by the ratio of estimated error by a Monte Carlo calculation to the true statistical error. In the previous study, the proposed method was verified in a simple condition such as one energy group, one dimensional homogeneous slab geometry. Contributions of this summary are as follows. Firstly, expression of the noise term that is appeared in the AR model is theoretically derived. Secondly, prediction on underestimation ratio of statistical uncertainty is carried out in a more general case, i.e., flux of each fuel cell in 2 x 2 heterogeneous assembly geometry taken from the C5G7 benchmark problem. Through the present study, applicability of the theoretical prediction for underestimation of statistical uncertainty in practical problems will be verified. (authors)

21 SPECIFIC NUCLEAR REACTORS AND ASSOCIATED PLANTS↗

Quantifying Reaction and Rate Heterogeneity in Battery Electrodes in 3D through Operando X-ray Diffraction Computed Tomography

In composite battery electrode architectures, local limitations in ionic and electronic transport can result in nonuniform energy storage reactions. Understanding such reaction heterogeneity is important to optimizing battery performance, including rate capability and mitigating degradation and failure. Here, we use spatially resolved X-ray diffraction computed tomography to map the reaction in a composite electrode based on the LiFePO 4 active material as it undergoes charge and discharge. Accelerated reactions at the electrode faces in contact with either the separator or the current collector demonstrate that both ionic and electronic transport limit the reaction progress. The data quantify how nonuniformity of the electrode reaction leads to variability in the charge/discharge rate, both as a function of time and position within the electrode architecture. Importantly, this local variation in the reaction rate means that the maximum rate that individual cathode particles experience can be substantially higher than the average, control charge/discharge rate, by a factor of at least 2-5 times. As a result, this rate heterogeneity may accelerate rate-dependent degradation pathways in regions of the composite electrode experiencing faster-than-average reaction and has important implications for understanding and optimizing rate-dependent battery performance. Benchmarking multiscale continuum model parameters against the observed reaction heterogeneity permits extension of these models to other electrode geometries.

25 ENERGY STORAGE↗

Benchmarking Operators in Deep Neural Networks for Improving Performance Portability of SYCL

SYCL is a portable programming model for heterogeneous computing, so it is important to obtain reasonable performance portability of SYCL. Towards the goal of better understanding and improving performance portability of SYCL for machine learning workloads, we have been developing benchmarks for basic operators in deep neural networks (DNNs). These operators could be offloaded to heterogeneous computing devices such as graphics processing units (GPUs) to speed up computation. In this paper, we introduce the benchmarks, evaluate the performance of the operators on GPU-based systems, and describe the causes of the performance gap between the SYCL and Compute Unified Device Architecture (CUDA) kernels. We find that the causes are related to the utilization of the texture cache for read-only data, optimization of the memory accesses with strength reduction, use of local memory, and register usage per thread. We hope that the efforts of developing benchmarks for studying performance portability will stimulate discussion and interactions within the community.

Jin, Zheming [ORNL] (ORCID:000000027197780X)↗

ML-Based Pebble Power Reconstruction for Pebble Bed Reactor Analysis

Pebble power reconstruction has been explored to complement the conventional homogenized modeling approach in pebble bed reactor (PBR) analysis, as detailed heterogeneous geometry calculations are computationally expensive. The random distribution of pebble fuels within the core challenges the application of conventional pin power reconstruction methods. To address this, we introduce a machine learning approach based on the transformer model, composed of encoder and decoder layers, to estimate the flux and power form functions for reconstructing individual pebble neutron fluxes and powers. The homogeneous neutron flux distribution within each spectral zone (SZ) is obtained from finite element solutions of global diffusion or transport calculations. Verification tests demonstrate that the trained transformer model accurately predicts power form functions over a range of conditions, including variations in pebble enrichment, location, type, SZ size, and burnup. In particular, verification using a three-dimensional PBR benchmark with burned pebbles shows good agreement in heterogeneous pebble power distributions between Griffin and Serpent. These results highlight the potential of applying conventional pin power reconstruction approaches to PBR cores with randomly distributed pebbles.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Derivation of heterogeneous material distributions and their sensitivity to HM-coupled two-phase flow models exemplified with the LASGIT experiment

Abstract Advective gas transport in bentonite, a possible buffer material in repositories for radioactive materials, is difficult to simulate in numerical continuum models, partly due to the complicated microstructure of bentonite. To generate reliable models of repositories nevertheless, spatially distributed heterogeneous material properties can be used to allow localization of gas flow. In this study, a pore-size-dependent stochastic approach of the gas entry pressure is derived from Mercury Intrusion Porosimetry, which is used to replicate measurements from the LASGIT experiment. In addition, three benchmark tests are simulated to investigate the dependence of heterogeneous distributions of material properties on the mesh discretization, the temporal dependence, and the coupling between the processes influenced by the heterogeneous parameters. The numerical modeling results of the LASGIT experiment show that the onset of gas flow into the system and the subsequent increase in pressure and stress can be well reproduced using heterogeneous distributions. Compared to a model with homogeneous material properties, heterogeneous distributions may allow the generation of dilatancy-controlled microfractures—an important feature with regard to the advective gas flow in bentonites. However, it can be observed that the heterogeneous distributions in LASGIT are less significant, as technical gaps or differences in material types could have a greater impact.

Environmental Sciences & Ecology↗

Application of the Hybrid Diffusion-Transport Spatial Homogenization Method to a Pressurized Water Reactor Benchmark Problem - Paper 70

The recently developed Hybrid Diffusion-Transport Spatial Homogenization (DTH) method was previously tested on a benchmark problem typical of a boiling water reactor. In this summary, the DTH method is tested in a 1-D benchmark problem typical of a pressurized water reactor. The method recreates the heterogeneous flux to within 1% average flux error and 34 pcm eigenvalue error in half the time as a heterogeneous transport calculation. (authors)

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Direct Observations of Solute Dispersion in Rocks With Distinct Degree of Sub‐Micron Porosity

Abstract The transport of chemical species in rocks is affected by their structural heterogeneity to yield a wide spectrum of local solute concentrations. To quantify such imperfect mixing, advanced methodologies are needed that augment the traditional breakthrough curve analysis by probing solute concentration within the fluids locally. Here, we demonstrate the application of asynchronous, multimodality imaging by X‐ray computed tomography (XCT) and positron emission tomography (PET) to the study of passive tracer experiments in laboratory rock cores. The four‐dimensional concentration maps measured by PET reveal specific signatures of the transport process, which we have quantified using fundamental measures of mixing and spreading. We observe that the extent of solute spreading correlate strongly with the strength of subcore‐scale porosity heterogeneity measured by XCT, while dilution is enhanced in rocks containing substantial sub‐micron porosity. We observe that the analysis of different metrics is necessary, as they can differ in their sensitivity to the strength and forms of heterogeneity. The multimodality imaging approach is uniquely suited to probe the fundamental difference between spreading and mixing in heterogeneous media. We propose that when multi‐dimensional data is available, mixing and spreading can be independently quantified using the same metric. We also demonstrate that one‐dimensional transport models have limited predictive ability toward the internal evolution of the solute concentration, when the model is solely calibrated against the effluent breakthrough curves. The data set generated in this study can be used to build realistic digital rock models and to benchmark transport simulations that account deterministically for rock property heterogeneity.

Kurotori, Takeshi [Department of Chemical Engineer↗

Analytical and Computational Studies of Correlations of Hydrodynamic Fluctuations in Shear Flow

Correlations of hydrodynamic fluctuations in shear flow analytically and also by dissipative particle dynamics (DPD) simulations are studied. The hydrodynamic equations are linearized around the macroscopic velocity field and then solved by a perturbation method in Fourier-transformed space. The autocorrelation functions (ACFs) from the analytical method are compared with results obtained from DPD simulations under the same shear-flow conditions. Up to a moderate shear rate, various ACFs from the two approaches agree with each other well. At large shear rates, discrepancies between the two methods are observed, hence revealing strong additional coupling between different fluctuating variables, which is not considered in the analytical approach. In addition, the results at low and moderate shear rates can serve as benchmarks for developing multiscale algorithms for coupling of heterogeneous solvers, such as a hybrid simulation of molecular dynamics and fluctuating hydrodynamics solver, where thermal fluctuations are indispensable.

Physics↗

Comparing High Performance Computing Accelerator Programming Models

Accelerator devices are becoming a norm in High Performance Computing (HPC). With more systems opting for heterogeneous architectures, portable programming models like OpenMP and OpenACC are becoming increasingly important. The SPEC ACCEL 1.2 benchmark suite consists of comparable benchmarks in OpenCL, OpenMP 4.5, and OpenACC 2.5 that can be used to evaluate the performance and support for programming models and frameworks on heterogeneous platforms. In this paper we go beneath the normative metric of performance times and look at the individual kernels to study the usage, strengths, and weaknesses of the two prevalent portable heterogeneous programming models, OpenMP and OpenACC. From our analysis we identify that benchmarks like MRI-Q, SP and BT have better performance using OpenACC, while benchmarks like MiniGhost, LBM and LBDC do consistently better with the OpenMP programming model across supercomputers like Titan, and Summit.We deep dive into the kernels of select four benchmarks to answer questions like: Where does the benchmark spend most of its cycles? What is the parallelization strategy used? Why is one programming model more performant than the other? By identifying the similarities and dierences we want to contrast between the benchmark implementation strategies in the SPEC ACCEL 1.2 benchmarks and provide more insights into the OpenMP and OpenACC programming models.

Pophale, Swaroop S.↗

WE-A-17A-08: Evaluation of the OncentraBrachy Collapsed Cone Convolution Algorithm for Ir-192 Source Using Phantom and Real-Patient Heterogeneous Geometries

Purpose: To evaluate the commercially released Collapsed Cone convolution-based(CCC) dose calculation module of the Elekta OncentraBrachy(OcB) treatment planning system(TPS). Methods: An allwater phantom was used to perform TG43 benchmarks with single source and seventeen sources, separately. Furthermore, four real-patient heterogeneous geometries (chestwall, lung, breast and prostate) were used. They were selected based on their clinical representativity of a class of clinical anatomies that pose clear challenges. The plans were used as is(no modification). For each case, TG43 and CCC calculations were performed in the OcB TPS, with TG186-recommended materials properly assigned to ROIs. For comparison, Monte Carlo simulation was run for each case with the same material scheme and grid mesh as TPS calculations. Both modes of CCC (standard and high quality) were tested. Results: For the benchmark case, the CCC dose, when divided by that of TG43, yields hot-n-cold spots in a radial pattern. The pattern of the high mode is denser than that of the standard mode and is representative of angular dicretization. The total deviation ((hot-cold)/TG43) is 18% for standard mode and 11% for high mode. Seventeen dwell positions help to reduce “ray-effect”, with the total deviation to 6% (standard) and 5% (high), respectively. For the four patient cases, CCC produces, as expected, more realistic dose distributions than TG43. A close agreement was observed between CCC and MC for all isodose lines, from 20% and up; the 10% isodose line of CCC appears shifted compared to that of MC. The DVH plots show dose deviations of CCC from MC in small volume, high dose regions (>100% isodose). For patient cases, the difference between standard and high modes is almost undiscernable. Conclusion: OncentraBrachy CCC algorithm marks a significant dosimetry improvement relative to TG43 in real-patient cases. Further researches are recommended regarding the clinical implications of the above observations. Support provided by a CIHR grant and CCC system provided by Elekta-Nucletron.

60 APPLIED LIFE SCIENCES↗

A scalable variational inequality approach for flow through porous media models with pressure-dependent viscosity

Mathematical models for flow through porous media typically enjoy the so-called maximum principles, which place bounds on the pressure field. It is highly desirable to preserve these bounds on the pressure field in predictive numerical simulations, that is, one needs to satisfy discrete maximum principles (DMP). Unfortunately, many of the existing formulations for flow through porous media models do not satisfy DMP. This paper presents a robust, scalable numerical formulation based on variational inequalities (VI), to model non-linear flows through heterogeneous, anisotropic porous media without violating DMP. VI is an optimization technique that places bounds on the numerical solutions of partial differential equations. To crystallize the ideas, a modification to Darcy equations by taking into account pressure-dependent viscosity will be discretized using the lowest-order Raviart–Thomas (RT0) and Variational Multi-scale (VMS) finite element formulations. It will be shown that these formulations violate DMP, and, in fact, these violations increase with an increase in anisotropy. It will be shown that the proposed VI-based formulation provides a viable route to enforce DMP. Moreover, it will be shown that the proposed formulation is scalable, and can work with any numerical discretization and weak form. Aseriesof numerical benchmark problems are solved to demonstrate the effects of heterogeneity, anisotropy and non-linearity on DMP violations under the two chosen formulations (RT0 and VMS), and that of non-linearity on solver convergence for the proposed VI-based formulation. Parallel scalability on modern computational platforms will be illustrated through strong-scaling studies, which will prove the efficiency of the proposed formulation in a parallel setting. Algorithmic scalability as the problem size is scaled up will be demonstrated through novel static-scaling studies. The performed static-scaling studies can serve as a guide for users to be able to select an appropriate discretization for a given problem size.

42 ENGINEERING↗

The GPU-enabled divide-expand-consolidate RI-MP2 method (DEC-RI-MP2)

We report porting of the Divide-Expand-Consolidate Resolution of the Identity second-order Møller–Plesset perturbation (DEC-RI-MP2) method to the graphic processing units (GPUs) using OpenACC compiler directives. It is shown that the OpenACC compiler directives implementation efficiently accelerates the rate-determining step of the DEC-RI-MP2 method with minor implementation effort. Moreover, the GPU acceleration results in a better load balance and thus in an overall scaling improvement of the DEC algorithm. The resulting cross-platform hybrid MPI/OpenMP/OpenACC implementation has scalable and portable performance on heterogeneous HPC architectures. The GPU-enabled code was benchmarked using a reduced version of the S12L test set of Stefan Grimme (Grimme, Chem. Eur. J. 2012, 18, 9955) consisting of supramolecular complexes up to 158 atoms and 4292 contracted basis functions (cc-pVTZ). The test set results demonstrate the general applicability of the DEC-RI-MP2 method showing results consistent with the DEC-RI-MP2 introductory paper (Baudin et al., J. Chem. Phys. 2016, 144, 054102) on molecules of complicated electronic structures. © 2016 Wiley Periodicals, Inc.

Chemistry↗

Characterization of NiFe oxyhydroxide electrocatalysts by integrated electronic structure calculations and spectroelectrochemistry

NiFe oxyhydroxide materials are highly active electrocatalysts for the oxygen evolution reaction (OER), an important process for carbon-neutral energy storage. Recent spectroscopic and computational studies increasingly support iron as the site of catalytic activity but differ with respect to the relevant iron redox state. A combination of hybrid periodic density functional theory calculations and spectroelectrochemical experiments elucidate the electronic structure and redox thermodynamics of Ni-only and mixed NiFe oxyhydroxide thin-film electrocatalysts. The UV/visible light absorbance of the Ni-only catalyst depends on the applied potential as metal ions in the film are oxidized before the onset of OER activity. In contrast, absorbance changes are negligible in a 25% Fe-doped catalyst up to the onset of OER activity. First-principles calculations of proton-coupled redox potentials and magnetizations reveal that the Ni-only system features oxidation of Ni2+ to Ni3+, followed by oxidation to a mixed Ni3+/4+ state at a potential coincident with the onset of OER activity. Calculations on the 25% Fedoped system show the catalyst is redox inert before the onset of catalysis, which coincides with the formation of Fe4+ and mixed Ni oxidation states. The calculations indicate that introduction of Fe dopants changes the character of the conduction band minimum from Ni-oxide in the Ni-only to predominantly Fe-oxide in the NiFe electrocatalyst. These findings provide a unified experimental and theoretical description of the electrochemical and optical properties of Ni and NiFe oxyhydroxide electrocatalysts and serve as an important benchmark for computational characterization of mixedmetal oxidation states in heterogeneous catalysts.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

An Innovative Three-Dimensional Heterogeneous Coarse-Mesh Transport Method for Advanced and Generation IV Reactor Core Analysis and Design

This project has resulted in a highly efficient method that has been shown to provide accurate solutions to a variety of 2D and 3D reactor problems. The goal of this project was to develop (1) an accurate and efficient three-dimensional whole-core neutronics method with the following features: based sollely on transport theory, does not require the use of cross-section homogenization, contains a highly accurate and self-consistent global flux reconstruction procedure, and is applicable to large, heterogeneous reactor models, and to (2) create new numerical benchmark problems for code cross-comparison.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Evaluating Operators in Deep Neural Networks for Improving Performance Portability of SYCL

SYCL is a portable programming model for heterogeneous computing, so it is important to obtain reasonable performance portability of SYCL. Towards the goal of better understanding and improving performance portability of SYCL for machine learning workloads, we have been developing benchmarks for basic operators in deep neural networks (DNNs). These operators could be offloaded to heterogeneous computing devices such as graphics processing units (GPUs) to speed up computation. In this work, we introduce the benchmarks, evaluate the performance of the operators on GPU-based systems, and describe the causes of the performance gap between the SYCL and Compute Unified Device Architecture (CUDA) kernels. We find that the causes are related to the utilization of the texture cache for read-only data, optimization of the memory accesses with strength reduction, shared local memory accesses, and register usage per thread. We hope that the efforts of developing benchmarks for studying performance portability will stimulate discussion and interactions within the community.

97 MATHEMATICS AND COMPUTING↗