Search NASA⌕ Search

SEARCH · Search NASA

Results for “High gradient acceleration”

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 973 records · Page 54
All sourcesNASA NTRSDOE OSTI
DOE OSTI2012

Accelerating Satellite Image Based Large-Scale Settlement Detection with GPU

Computer vision algorithms for image analysis are often computationally demanding. Application of such algorithms on large image databases\---- such as the high-resolution satellite imagery covering the entire land surface, can easily saturate the computational capabilities of conventional CPUs. There is a great demand for vision algorithms running on high performance computing (HPC) architecture capable of processing petascale image data. We exploit the parallel processing capability of GPUs to present a GPU-friendly algorithm for robust and efficient detection of settlements from large-scale high-resolution satellite imagery. Feature descriptor generation is an expensive, but a key step in automated scene analysis. To address this challenge, we present GPU implementations for three different feature descriptors\-- multiscale Historgram of Oriented Gradients (HOG), Gray Level Co-Occurrence Matrix (GLCM) Contrast and local pixel intensity statistics. We perform extensive experimental evaluations of our implementation using diverse and large image datasets. Our GPU implementation of the feature descriptor algorithms results in speedups of 220 times compared to the CPU version. We present an highly efficient settlement detection system running on a multiGPU architecture capable of extracting human settlement regions from a city-scale sub-meter spatial resolution aerial imagery spanning roughly 1200 sq. kilometers in just 56 seconds with detection accuracy close to 90\%. This remarkable speedup gained by our vision algorithm maintaining high detection accuracy clearly demonstrates that such computational advancements clearly hold the solution for petascale image analysis challenges.

Patlolla, Dilip Reddy↗
DOE OSTI2013

Comparison of methods for the measurement of radiation dose distributions in high dose rate (HDR) brachytherapy: Ge-doped optical fiber, EBT3 Gafchromic film, and PRESAGE{sup Registered-Sign} radiochromic plastic

Purpose: Dose distribution measurement in clinical high dose rate (HDR) brachytherapy is challenging, because of the high dose gradients, large dose variations, and small scale, but it is essential to verify accurate treatment planning and treatment equipment performance. The authors compare and evaluate three dosimetry systems for potential use in brachytherapy dose distribution measurement: Ge-doped optical fibers, EBT3 Gafchromic film with multichannel analysis, and the radiochromic material PRESAGE{sup Registered-Sign} with optical-CT readout. Methods: Ge-doped SiO{sub 2} fibers with 6 {mu}m active core and 5.0 mm length were sensitivity-batched and their thermoluminescent properties used via conventional heating and annealing cycles. EBT3 Gafchromic film of 30 {mu}m active thickness was calibrated in three color channels using a nominal 6 MV linear accelerator. A 48-bit transmission scanner and advanced multichannel analysis method were utilized to derive dose measurements. Samples of the solid radiochromic polymer PRESAGE{sup Registered-Sign }, 60 mm diameter and 100 mm height, were analyzed with a parallel beam optical CT scanner. Each dosimetry system was used to measure the dose as a function of radial distance from a Co-60 HDR source, with results compared to Monte Carlo TG-43 model data. Each system was then used to measure the dose distribution along one or more lines through typical clinical dose distributions for cervix brachytherapy, with results compared to treatment planning system (TPS) calculations. Purpose-designed test objects constructed of Solid Water and held within a full-scatter water tank were utilized. Results: All three dosimetry systems reproduced the general shape of the isolated source radial dose function and the TPS dose distribution. However, the dynamic range of EBT3 exceeded those of doped optical fibers and PRESAGE{sup Registered-Sign }, and the latter two suffered from unacceptable noise and artifact. For the experimental conditions used in this study, the useful range from an isolated HDR source was 5-40 mm for fibers, 3-50 mm for EBT3, and 4-21 mm for PRESAGE{sup Registered-Sign }. Fibers demonstrated some over-response at very low dose levels, suffered from volume averaging effects in the dose distribution measurement, and exhibited up to 9% repeatability variation over three repeated measurements. EBT3 demonstrated excellent agreement with Monte Carlo and TPS dose distributions, with up to 3% repeatability over three measurements. PRESAGE{sup Registered-Sign} gave promising results, being the only true 3D dosimeter, but artifacts and noise were apparent. Conclusions: The comparative response of three emerging dosimetry systems for clinical brachytherapy dose distribution measurement has been investigated. Ge-doped optical fibers have excellent spatial resolution for single-direction measurement but are currently too large for complex dose distribution assessment. The use of PRESAGE{sup Registered-Sign} with optical-CT readout gave promising results in the measurement of true 3D dose distributions but further development work is required to reduce noise and improve dynamic range for brachytherapy dose distribution measurements. EBT3 Gafchromic film with multichannel analysis demonstrated accurate and reproducible measurement of dose distributions in HDR brachytherapy. Calibrated dose measurements were possible with agreement within 1.5% of TPS dose calculations. The suitability of EBT3 as a dosimeter for 2D quality control or commissioning work has been demonstrated.

62 RADIOLOGY AND NUCLEAR MEDICINE↗
NASA NTRS1986

The clumpy circumstellar medium around young supernova remnants

Each of the youngest supernova remnants known in the Milky Way, Cas A, Kepler's SNR, and Tycho's SNR, shows a different morphological structure caused by different conditions in the progenitor stars and their surroundings. In all three cases, however, the observed shells have a thickness of about 1/4 the radius, a sharp outer edge, and significant brightness irregularities. These features require that the circumstellar medium be highly clumped. To investigate the phenomenon, models of the expansion have been constructed using a one-dimensional spherical hydrodynamic code. As a supernova shock moves down the external density gradient of the star, material behind the shock begins to go into free expansion. Then as surrounding material is encountered a reverse shock moving back into the ejectum will be formed. Until the expansion has swept up about eight times the ejected mass when the situation can be considered as a point explosion in its surroundings, the dynamics are controlled by conditions between the shocks. The region is also where the synchrotron radio emission from relativistic electrons trapped in magnetic fields arises. Initial particles and fields are accelerated and amplified by eddy motion at the interface between the ejected and swept-up material and at the boundaries of clumps. Polarimetry shows that these SNR have a net radial orientation of their magnetic fields apparently from stretching by Rayleigh-Taylor instabilities at the contact surfaces. Without clumps the observed shell is much too narrow and steep on the inside.

Dickel, J. R.↗
DOE OSTI2012

High-Field Combined-Function Magnets for a 1.5 x 1.5 TeV Muon Collider Storage Ring

Designs and parameters of dipoles and combined-function quadrupoles for bending arc lattice of a 1.5x1.5 TeV muon collider with an average luminosity of 4x1034 cm{sup -2}s{sup -1} are presented. The magnets use the Nb{sub 3}Sn superconductor and provide the required gradients and fields with the appropriate operating margins and field quality. The magnet apertures accommodate tungsten liners to minimize the dynamic heat load in the superconducting coils.

43 PARTICLE ACCELERATORS↗
DOE OSTI2017

Cavity Processing and Preparation of 650 MHz Elliptical Cell Cavities for PIP-II

The PIP-II project at Fermilab requires fifteen 650 MHz SRF cryomodules as part of the 800 MeV LINAC that will provide a high intensity proton beam to the Fermilab neutrino program. A total of fifty-seven high-performance SRF cavities will populate the cryomodules and will operate in both pulsed and continuous wave modes. These cavities will be processed and prepared for performance testing utilizing adapted cavity processing infrastructure already in place at Fermilab and Argonne. The processing recipes implemented for these structures will incorporate state-of-the art processing and cleaning techniques developed for 1.3 GHz SRF cavities for the ILC, XFEL, and LCLS-II projects. This paper describes the details of the processing recipes and associated chemistry, heat treatment, and cleanroom processes at the Fermilab and Argonne cavity processing facilities. This paper also presents single and multi-cell cavity test results with quality factors above 5·10¹⁰ and accelerating gradients above 30 MV/m.

43 PARTICLE ACCELERATORS↗
NASA NTRS2021

High-Fidelity Simulation of Turbulent Flow Past a Gaussian Bump

A spanwise-periodic computation of a turbulent flow past a Gaussian bump is performed in the form of a hybrid direct numerical simulation and wall-resolved large-eddy simulation. A fourth-order spatially-accurate flow solver is employed to perform the simulation, using 10.2 billion grid points for a Reynolds number of 170000 based on the bump height. The key findings from the simulation are reported in the acceleration and deceleration flow regions associated with the bump shape. Significant anisotropy in the normal Reynolds stresses, along both the wall-normal and streamwise directions, is observed within the acceleration region. The ratio between the Reynolds shear stress and turbulent kinetic energy in that region also experiences significant deviations from the norms of a zero pressure gradient turbulent boundary layer. The chosen Reynolds number generates strong flow separation in the adverse pressure gradient region, which is in contrast with a previous simulation at half the Reynolds number that only indicated incipient separation. An internal layer generated in the acceleration region evolves into a free shear layer that develops in the deceleration region and separates. Proper modeling of this inner layer appears crucial to predict the flow separation. Surface curvature effects on the attached flow development are also discussed.

Turbulent Boundary Layer↗
NASA NTRS1995

A review of solar wind ion and electron plasma distribution functions: Present understanding and Ulysses results

Our present understanding of the distribution functions of the solar wind ion and electron thermal/suprathermal populations will be reviewed covering selected theoretical and observational topics. Roughly half the review will be devoted to recent discoveries (since Solar Wind 7). Among recent results are those of the Ulysses mission. The SWICS experiment observed pick-up protons and alpha particles, and acceleration of these particles at interplanetary shocks. Positive ion phenomena observed by SWOOPS include enhanced proton-alpha particle differential streaming both: (1) at high latitudes; and (2) in the ecliptic beyond 1 AU in the vicinity of shocks. The SWOOPS positive ion observations also demonstrate relative constancy of the entropy per proton at high latitudes beyond about 2.5 AU. Double beaming in both protons and alphas is observed by SWOOPS on either side of the current sheet, but not within; a possible explanation is reconnection at the edge of the current sheet. SWOOPS observed bi-directional streaming of electrons beyond 2 AU not only within Coronal Mass Ejections, but also upstream of corotating interplanetary shocks. Latitudinal and radial gradients of Ulysses electron core and halo temperatures will be discussed, as will work in progress on the three dimensional (in velocity space) properties of ion distributions.

Goldstein, B. E.↗
DOE OSTI2012

Update on the R&D of vertical buffered electropolishing on Nb samples and SRF single cell cavities

Electropolishing (EP) has become a popular choice as the final step of the surface removal process during the fabrication of Nb superconducting radio frequency (SRF) cavities. One of the major reasons for the choice is that Nb SRF cavities treated by EP tend to have a better chance to reach an accelerating gradient of 30 MV/m or higher than those treated by buffered chemical polishing (BCP). This advantage of EP over BCP can at least be partially attributed to the smoother Nb surfaces that EP can produce. Recently a Nb surface removal technique called buffered electropolishing (BEP) was developed at JLab. The root mean square of the Nb surfaces treated by BEP can be as smooth as 20nm over a surface area of 200X200 ?m2. In this contribution, R&D efforts of vertical BEP on Nb small flat samples and SRF single cell cavities since the last SRF conference in 2009 will be updated. It is shown that under a suitable condition, BEP can have a Nb removal rate as high as 10 ?m/mim that is more than 25 and 5 times quicker than those of EP and BCP(112) respectively. Possible mechanisms responsible for the high Nb removal rate are proposed. Clues on the optimization of vertical BEP and EP treatments on Nb SRF cavities from recent experimental results obtained on a Nb single cell demountable cavity will be discussed.

Wu, Andy↗
DOE OSTI2016

Modification and validation of an analytical source model for external beam radiotherapy Monte Carlo dose calculations

Purpose: A dose calculation tool, which combines the accuracy of the dose planning method (DPM) Monte Carlo code and the versatility of a practical analytical multisource model, which was previously reported has been improved and validated for the Varian 6 and 10 MV linear accelerators (linacs). The calculation tool can be used to calculate doses in advanced clinical application studies. One shortcoming of current clinical trials that report dose from patient plans is the lack of a standardized dose calculation methodology. Because commercial treatment planning systems (TPSs) have their own dose calculation algorithms and the clinical trial participant who uses these systems is responsible for commissioning the beam model, variation exists in the reported calculated dose distributions. Today’s modern linac is manufactured to tight specifications so that variability within a linac model is quite low. The expectation is that a single dose calculation tool for a specific linac model can be used to accurately recalculate dose from patient plans that have been submitted to the clinical trial community from any institution. The calculation tool would provide for a more meaningful outcome analysis. Methods: The analytical source model was described by a primary point source, a secondary extra-focal source, and a contaminant electron source. Off-axis energy softening and fluence effects were also included. The additions of hyperbolic functions have been incorporated into the model to correct for the changes in output and in electron contamination with field size. A multileaf collimator (MLC) model is included to facilitate phantom and patient dose calculations. An offset to the MLC leaf positions was used to correct for the rudimentary assumed primary point source. Results: Dose calculations of the depth dose and profiles for field sizes 4 × 4 to 40 × 40 cm agree with measurement within 2% of the maximum dose or 2 mm distance to agreement (DTA) for 95% of the data points tested. The model was capable of predicting the depth of the maximum dose within 1 mm. Anthropomorphic phantom benchmark testing of modulated and patterned MLCs treatment plans showed agreement to measurement within 3% in target regions using thermoluminescent dosimeters (TLD). Using radiochromic film normalized to TLD, a gamma criteria of 3% of maximum dose and 2 mm DTA was applied with a pass rate of least 85% in the high dose, high gradient, and low dose regions. Finally, recalculations of patient plans using DPM showed good agreement relative to a commercial TPS when comparing dose volume histograms and 2D dose distributions. Conclusions: A unique analytical source model coupled to the dose planning method Monte Carlo dose calculation code has been modified and validated using basic beam data and anthropomorphic phantom measurement. While this tool can be applied in general use for a particular linac model, specifically it was developed to provide a singular methodology to independently assess treatment plan dose distributions from those clinical institutions participating in National Cancer Institute trials.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2022

Spin-piston problem for a ferromagnetic thin film: Shock waves and solitons

In this report the unsteady, nonlinear magnetization dynamics induced by spin injection in an easy-plane ferromagnetic channel subject to an external magnetic field are studied analytically. Leveraging a dispersive hydrodynamic description, the Landau-Lifshitz equation is recast in terms of hydrodynamic-like variables for the magnetization's perpendicular component (spin density) and azimuthal phase gradient (fluid velocity). Spin injection acts as a moving piston that generates nonlinear, dynamical spin textures in the ferromagnetic channel with downstream quiescent spin density set by the external field. In contrast to the classical problem of a piston accelerating a compressible gas, here, variable spin injection and field lead to a rich variety of nonlinear wave phenomena from oscillatory spin shocks to solitons and rarefaction (expansion) waves. A full classification of solutions is provided using nonlinear wave modulation theory by identifying two key aspects of the fluid-like dynamics: subsonic/supersonic conditions and convex/nonconvex hydrodynamic flux. Familiar waveforms from the classical piston problem such as rarefaction waves and shocks manifest in their spin-based counterparts as smooth and highly oscillatory transitions, respectively, between two distinct magnetic states. The spin shock is an example of a dispersive shock wave, which arises in many physical systems. New features without a gas dynamics counterpart include composite wave complexes with “contact” spin shocks and rarefactions. Magnetic supersonic conditions lead to two pronounced piston edge behaviors including a stationary soliton and an oscillatory wave train. These coherent wave structures have physical implications for the generation of high frequency spin waves from pulsed injection and persistent, stable stationary and/or propagating solitons in the presence of magnetic damping. The analytical results are favorably compared with numerical simulations.

36 MATERIALS SCIENCE↗
NASA NTRS1981

Linear theory of tearing in a high-beta plasma

The linear dispersion relation for a collisionless plasma in a sheared one-dimensional current sheet is calculated with reference to conditions in the daytime magnetopause. Calculations are extended to include plasmas with beta approximately equal to 1. It is found that the tearing mode eigenstructure and temporal growth rate are a sensitive function of the ratios l sub s/l sub G, l sub s/l sub T, and l sub s/l sub n, where l sub s is the shearing length of the magnetic field, l sub G is the gradient scale length, and l sub T is the temperature gradient scale length. In particular, if beta is approximately equal to 1, and l sub s is less than l sub G, l sub n, and l sub T, then the thickness of the layer over which particles are resonantly accelerated by the induction magnetic field is approximately rho, a thermal gyroradius. If the above conditions are not satisfied, plasma gradients may electrostatically stabilize the mode.

Quest, K. B.↗
NASA NTRS1985

High-thermal-gradient Superalloy Crystal Growth

Single, (001)-oriented crystals of PWA 1480 were processed in alumina/silica shell molds in a laboratory high gradient furnace. The furnace employs a graphite resistance heated element, a radiation baffle, and a water cooled radiation trap below the baffle. All crystals were grown in vacuum (10 torr) and all heat transfer was radiative. The element is constructed with a variable cross section that is tapered just above the baffle to maximize heat input and therefore thermal gradient. A maximum alloy temperature of 1600 C was used. A thermal gradient of 130 deg C/cm was recorded at 1370 C just above the solidus of the PWA 1480 alloys. Crystal bars with 14.4 and 17.5 mm diameters were grown in alumina/silica shell molds. Each crystal was started from a 1.6 mm pencil seed at a rate of 76 mm/hr and slowly accelerated to a rate of 200 mm/hr under computer control. Volume percent porosity and average pore size were measured as functions of distance in representative bars. Low cycle fatigue behavior and stress rupture properties were determined.

Pearson, D. D.↗
DOE OSTI2010

Dense Heterogeneous Continuum Model of Two-Phase Explosion Fields

A heterogeneous continuum model is proposed to describe the dispersion of a dense Aluminum particle cloud in an explosion. Let {alpha}{sub 1} denote the volume fraction occupied by the gas and {alpha}{sub 2} the fraction occupied by the solid, satisfying the volume conservation relation: {alpha}{sub 1} + {alpha}{sub 2} = 1. When the particle phase occupies a non-negligible volume fraction (i.e., {alpha}{sub 2} > 0), additional terms, proportional to {alpha}{sub 2}, appear in the conservation laws for two-phase flows. These include: (i) a particle pressure (due to particle collisions), (ii) a corresponding sound speed (which produces real eigenvalues for the particle phase system), (iii) an Archimedes force induced on the particle phase (by the gas pressure gradient), and (iv) multi-particle drag effects (which enhance the momentum coupling between phases). These effects modify the accelerations and energy distributions in the phases; we call this the Dense Heterogeneous Continuum Model. A characteristics analysis of the Model equations indicates that the system is hyperbolic with real eigenvalues for the gas phase: {l_brace}v{sub 1}, v{sub 1} {+-} {alpha}{sub 1}{r_brace} and for the 'particle gas' phase: {l_brace}v{sub 2}, v{sub 2} {+-}{alpha}{sub 2}{r_brace} and the particles: {l_brace}v{sub 2}{r_brace}, where v{sub i} and {alpha}{sub i} denote the velocity vector and sound speed of phase i. These can be used to construct a high-order Godunov scheme to integrate the conservation laws of a dense heterogeneous continuum.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2010

Phase relations and hardness trends of ZrO[subscript 2] phases at high pressure

We use high-resolution synchrotron x-ray powder diffraction and density-functional theory (DFT) to investigate the phase stability, equations of state (EOSs), and mechanical hardness of zirconia (ZrO{sub 2}) up to {approx}54 and 160 GPa, respectively. For the equilibrium phase at ambient conditions (MI), we provide an experimental EOS that is comparable to results obtained from room-pressure Brillouin scattering experiments and bulk modulus-volume systematics but different from previous high-pressure experiments. The experimental second-order Birch-Murnaghan EOS parameters of MI-ZrO{sub 2} are: ambient-pressure volume (V{sub 0}) of 35.15({+-}0.03) {angstrom}{sup 3}/f.u. with an ambient-pressure bulk modulus K{sub 0} of 210({+-}28) GPa. For the high-pressure OI phase, we find that the K{sub 0} = 290({+-}11) GPa, which is 19%-32% higher than previously determined, and V{sub 0} = 33.65({+-}0.07) {angstrom}{sup 3}/f.u. The small volume decrease of 3.4% across the MI {yields} OI transition at {approx}10 GPa is associated with a 38% increase in the bulk modulus consistent with our DFT calculations that predict a {approx}36% and 39% increase in K{sub 0} for the generalized gradient and local density approximations, respectively. In contrast to the EOS of MI and OI, we find that our experimental EOS for the high-pressure OII phase is in good agreement with previous measurements. The large volume decrease across the OI {yields} OII phase transition as obtained from both our experiments and calculations is {approx}10%. Our estimates, using scaling relations, indicate that this phase, while dense and quenchable, may have a comparatively low mechanical hardness of {approx} 10 GPa.

43 PARTICLE ACCELERATORS↗
DOE OSTI2010

Momentum transfer in the boundary layer when there is acceleration and combustion of ethanol as it evaporates behind a barrier

Experimental data have been gathered on the local parameters of a boundary layer gas flow with ethanol combustion behind a 3 mm-high rib. These parameters include averaged velocities, temperatures, concentrations of stable substances and OH radicals and mass fluxes on the wall. The temperature and composition of the gases were studied with probe methods (thermocouple and chromatography). To measure the velocity and concentration of radicals, we applied the laser optical measurement methods of LDA and LIF. We propose a way of processing the obtained data utilizing balances in the continuity and motion equations. The influence of incident flow acceleration on the viscous and turbulent shear stresses at the wall and in the volume of the boundary layer has been analyzed to determine the acceleration parameters to be K = (0, 0.7, 1.3, and 4.1) x 10{sup -6}. It is shown that without a longitudinal pressure gradient (K = 0) on the combustion that is behind the rib, OH radicals accumulate with their highest concentrations existing in areas that do not coincide with the flame front and are shifted toward the oxidizer. The main mechanism of momentum transfer is connected to the boundary layer separation. In the presence of acceleration caused by a negative longitudinal pressure gradient, the detachment area does not show up (in experiments with K > 0.7 x 10{sup -6}); shear stresses increase substantially and reach one percent of the dynamic pressure. The general level of turbulent stresses in the reacting boundary layer becomes higher than in the case without acceleration. The incident airflow is also accelerated by the reacting boundary layer in which the maximum velocity is formed. (author)

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2017

Thermal conductivity of self-ion irradiated nanocrystalline zirconium thin films

Thermomechanical stability and high thermal conductivity are important for nuclear cladding material performance and reliability, which degrade over time under irradiation. The literature suggests nanocrystalline materials as radiation tolerant, but little or no evidence is present from thermal transport perspective. In this study, we irradiated 10 nm grain size zirconium thin films with 800 keV Zr + beam from a 6 MV HVE Tandem accelerator to achieve various doses of 3 × 10 10 to 3.26 × 10 14 ions/cm 2 , corresponding to displacement per atom (dpa) of 2.1 × 10 –4 to 2.28. Transmission electron microscopy showed significant grain growth, texture evolution and oxidation in addition to the creation of displacement defects due to the irradiation. The specimens were co-fabricated with micro-heaters to establish thermal gradients that were mapped using infrared thermometry. An energy balance approach was used to estimate the thermal conductivity of the specimens, as function of irradiation dosage. As a result, up to 32% reduction of thermal conductivity was measured for the sample exposed to a dose of 2.1 dpa (3 × 10 14 ions/cm 2 ).

36 MATERIALS SCIENCE↗
DOE OSTI2025

Enabling Scientific Applications with Performance-Portability and High-Productivity for Multi-GPU Programming with JACC.Multi

This work bridges the gap between multi-GPU computing and high-productivity, performance-portable programming solutions. Our goal is to enhance scientific applications with a productive and portable solution—program once, deploy everywhere—for multi-GPU programming with no cost to programmability. To accomplish this, we implemented JACC.Multi, which is part of the Julia for ACCelerators (JACC) performance-portable framework. JACC. Multi is the only high-level, portable metaprogramming solution that targets multi-GPU environments and is integrated in a readily accessible programming language (e.g., Julia language). With transparent GPU-to-GPU communication, JACC. Multi is optimized for scientific application workloads and is portable for NVIDIA and AMD accelerators. For the evaluation, we use two modern multi-GPU systems: Hudson, which features two NVIDIA H100 Hopper GPUs per node, and Frontier, which features four AMD MI250X GPUs per node, each with two Graphics Compute Dies (GCDs) for a total of eight GCDs per node. Additionally, as part of the evaluation, we use JACC (one GPU), MPI+JACC, and JACC. Multi codes that implement well-known and widely used scientific algorithms/kernels such as the conjugate gradient algorithm and an explicit forward Euler solver that requires GPU-to-GPU communication. Overall, JACC. Multi codes achieve better performance than MPI+JACC codes and significant speedups over JACC (one GPU), with up to 1.9× on Hudson and 6× on Frontier.

Valero Lara, Pedro [ORNL] (ORCID:0000000214794310)↗
DOE OSTI2021

Replicated Computational Results (RCR) Report for “Adaptive Precision Block-Jacobi for High Performance Preconditioning in the Ginkgo Linear Algebra Software”

The article by Flegar et al. titled “Adaptive Precision Block-Jacobi for High Performance Preconditioning in the Ginkgo Linear Algebra Software” presents a novel, practical implementation of an adaptive precision block-Jacobi preconditioner. Performance results using state-of-the-art GPU architectures for the block-Jacobi preconditioner generation and application demonstrate the practical usability of the method, compared to a traditional full-precision block-Jacobi preconditioner. A production-ready implementation is provided in the Ginkgo numerical linear algebra library. In this report, the Ginkgo library is reinstalled and performance results are generated to perform a comparison to the original results when using Ginkgo’s Conjugate Gradient solver with either the full or the adaptive precision block-Jacobi preconditioner for a suite of test problems on an NVIDIA GPU accelerator. After completing this process, the published results are deemed reproducible.

97 MATHEMATICS AND COMPUTING↗