Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “accelerated computation”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 1,117 records · Page 62
All sourcesNASA NTRSDOE OSTI
DOE OSTI2026

Electromagnetic production of kaons on the nucleon

Studies of the electromagnetic production of strange quarks started in the 1950s as something of a curiosity that puzzled experimentalists and theorists alike. Eventually, a nascent understanding of these processes began to take shape through the first pioneering experiments dedicated to explore photo- and electroproduction that were carried out in the period from the 1950s to the 1980s. As the datasets increased, concomitant advances in theoretical models were realized. However, these initial studies also made clear that more precise data was essential to continue to move forward. A paradigm shift occurred in the 1990s with the development of second-generation facilities at ELSA, MAMI, SPring-8, and JLab. High-intensity, high duty-factor accelerators, coupled with novel detector systems and advances in computing and readout electronics, brought nuclear physics experiments forward by orders of magnitude in counting statistics compared to the first-generation efforts. This was an utter boon to strangeness physics investigations, and to date, more than 50 dedicated experiments in kaon photo- and electroproduction have been completed at facilities around the world, leading to a host of experimental observables that have enabled significant advances in the exploration of strongly interacting systems that decay via $s\bar{s}$ quark pair creation. These data have proven to be an essential complementary pathway to study the spectrum and structure of the excited states of the nucleon, and the search for missing and exotic baryon configurations. As well, investigations in these channels are requisite for exploring hypernuclear production as a probe of the $YN$ interaction and for studies of the electromagnetic form factors of strange mesons. This review was designed to provide the first-ever in-depth overview of both the experimental and theoretical progress in the field of the electromagnetic production of strangeness. This work looks back over 70 years of past developments, discusses ongoing work and near-term plans, and details future possibilities being considered for third-generation facilities. Extensive lists of the available datasets and theoretical models are provided, together with a comprehensive supporting bibliography of the field. Throughout this work, the primary impacts of these explorations are highlighted, along with connections to a wide range of related phenomenological applications. An important goal of this review is to provide a complete, (reasonably) self-contained guide into this field prepared at a level that is relevant for both new and seasoned scientists, whether experimentalists, phenomenologists, or theorists, to better understand what has been accomplished by so many dedicated folks-each building on what has come before-and to appreciate the exciting future potential for continued studies in this area.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2024

Optimizing High-Throughput Inference on Graph Neural Networks at Shared Computing Facilities with the NVIDIA Triton Inference Server

Abstract With machine learning applications now spanning a variety of computational tasks, multi-user shared computing facilities are devoting a rapidly increasing proportion of their resources to such algorithms. Graph neural networks (GNNs), for example, have provided astounding improvements in extracting complex signatures from data and are now widely used in a variety of applications, such as particle jet classification in high energy physics (HEP). However, GNNs also come with an enormous computational penalty that requires the use of GPUs to maintain reasonable throughput. At shared computing facilities, such as those used by physicists at Fermi National Accelerator Laboratory (Fermilab), methodical resource allocation and high throughput at the many-user scale are key to ensuring that resources are being used as efficiently as possible. These facilities, however, primarily provide CPU-only nodes, which proves detrimental to time-to-insight and computational throughput for workflows that include machine learning inference. In this work, we describe how a shared computing facility can use the NVIDIA Triton Inference Server to optimize its resource allocation and computing structure, recovering high throughput while scaling out to multiple users by massively parallelizing their machine learning inference. To demonstrate the effectiveness of this system in a realistic multi-user environment, we use the Fermilab Elastic Analysis Facility augmented with the Triton Inference Server to provide scalable and high-throughput access to a HEP-specific GNN and report on the outcome.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
DOE OSTI2024

High-Fidelity Accelerated Design of High-performance Electrochemical Systems

Large-scale electrification is vital to addressing the climate crisis, but several scientific and technological challenges remain to fully electrify both the chemical industry and transportation. In both of these areas, new electrochemical materials will be critical, but their development currently relies heavily on human-time-intensive experimental trial and error and computationally expensive first-principles, meso-scale and continuum simulations. To accelerate this process, our team has developed the AutoMat platform. AutoMat can accelerate development of new electrochemical materials along two avenues: first, automated input generation and management of simulations at multiple lengthscales as well as “handoff” of outputs from one lengthscale as inputs to the next; and second, replacement of the most computationally intensive simulation processes with machine-learned surrogate models. The crux of our team’s effort was not “reinventing the wheel” by developing entirely new techniques, but rather building a “superhighway” that allows existing state-of-the-art techniques to run faster and more smoothly than before. AutoMat can utilize tools spanning from first-principles quantum chemistry computations to automated robotic experimentation, and is driven by design space search techniques to reduce the number of iterations through the full simulation loop by rapidly targeting promising regions of design spaces such as single-atom alloy catalysts or blends of liquid electrolytes.

25 ENERGY STORAGE↗
NASA NTRS1968

Predicted vibration responses of Apollo structure and effects of pressure correlation lengths on response

Progressive wove and reverberant acoustic fields ore often used in the laboratory for qualification testing of aerospace structures that are subjected to unsteady aerodynamic environments which occur during flight through the atmosphere. Acoustic simulation of these in-flight environments i s achieved i f the structural vibration response to acoustic excitation is equivalent, over the frequency range of interest, to the response caused by aerodynamic turbulence. To determine the degree of response simulation that could be expected for such qualification testing of the Apollo Spacecraft, responses of a segment of the Apollo structure were estimated theoretically for both acoustic and flight environments. The vibration analysis presented in this report was performed for the Spacecraft Lunar Module Adaptor; and the three environments treated are turbulence at Mach 1 and Mach 2, a reverberant acoustic field, and a modified acoustic progressive wave field. The progressive wave field is assumed to be generated within a specially designed shroud of sixteen axially oriented ducts which is constructed around and contoured to the external skin of the vehicle. Each duct is open on the side adjacent to the skin and is driven acoustically at one end by an independent noise source, thus allowing for different pressure correlation patterns around the circumference of the vehicle. The purpose of the analysis was to determine the effects of surface pressure correlation lengths on the response characteristics of representative Apollo structure, and to select on this basis, the optimum acoustic environment for structural vibration qualification tests of the Apollo in the Spacecraft Acoustic Laboratory at MSC. The method of analysis employed consists of determining, independently, the space average, mean-square acceleration spectral density for each of the significant modes of vibration of an equivalent cylindrical shell, and then expressing the total structural response as a linear summation of these modal responses. Approximately 570 classical modes of a pinned-end cylinder were used in order to find the spectral response characteristics of the structure throughout the frequency range of 10 - 1,000 Hz. The effective forcing functions or joint acceptances of the various modes of vibration were computed and are presented graphically for each of the pressure excitation environments. A brief development of the response equations used is presented in the report. Space average acceleration response spectra were computed for twelve different cases including aerodynamic turbulence at Mach 1 and Mach 2, reverberant acoustic field, five duct correlation patterns, two cases of an axially damped progressive wave field, and two cases in which the structural damping constant was varied from the expected value. The response spectra were computed and plotted by a high speed digital computer far a flat excitation pressure spectrum; and the response spectra were numerically integrated to give one-third octave bond average responses. A complete set of digital computer programs has been developed for both flat plate and cylindrical shell structures for the above environments, and each requires run times of three minutes or less. Thus, the work presented herein, along with the results, constitutes a practical advancement to the existing state-of-the-art of vibration prediction.

R W White↗
DOE OSTI2020

Analysis of the performance of a hybrid CPU/GPU 1D2D coupled model for real flood cases

Coupled 1D2D models emerged as an efficient solution for a two-dimensional (2D) representation of the floodplain combined with a fast one-dimensional (1D) schematization of the main channel. At the same time, high-performance computing (HPC) has appeared as an efficient tool for model acceleration. In this work, a previously validated 1D2D Central Processing Unit (CPU) model is combined with an HPC technique for fast and accurate flood simulation. Due to the speed of 1D schemes, a hybrid CPU/GPU model that runs the 1D main channel on CPU and accelerates the 2D floodplain with a Graphics Processing Unit (GPU) is presented. Since the data transfer between sub-domains and devices (CPU/GPU) may be the main potential drawback of this architecture, the test cases are selected to carry out a careful time analysis. Here, the results reveal the speed-up dependency on the 2D mesh, the event to be solved and the 1D discretization of the main channel. Additionally, special attention must be paid to the time step size computation shared between sub-models. In spite of the use of a hybrid CPU/GPU implementation, high speed-ups are accomplished in some cases.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2020

Accelerating Quantum Technologies Through Fermilab SRF Expertise

Industry forecast of exponential growth in quantum computing power still predicts that transformative applications are decades away. System performance is limited by decoherence and requires 5-10x the number of physical qubits per effective qubit.

43 PARTICLE ACCELERATORS↗
DOE OSTI2022

SPEL: Software tool for Porting E3SM Land Model with OpenACC in a Function Unit Test Framework

Most high-end computers adopt hybrid architecture, porting a large-scale scientific code onto accelerators is necessary. The paper presents a generic method for porting large-scale scientific code onto accelerators using compiler directives within a modularized function unit test platform. We have implemented the method and designed a software tool (SPEL) to port the E3SM Land Model (ELM) onto the GPUs in the Summit computer. SPEL automatically generates GPU-ready test modules for all ELM functions, such as CanopyFlux, SoilTemperature, and EcosystemDynamics. SPEL breaks the ELM into a collection of standalone unit test programs for easy code verification and further performance improvement. We further optimize several ELM test modules with advanced techniques, including memory reduction, reconstructed parallel loops, and asynchronous GPU kernel launch. We hope our study will inspire new toolkit developments that expedite large-scale scientific code porting with compiler directives.

Schwartz, Peter↗
NASA NTRS1990

The effects of magnetic nozzle configurations on plasma thrusters

Plasma thrusters have been operated at power levels from 10 kw to 0.1 MW. When these devices have had magnetic fields applied to them which form a nozzle configuration for the expanding plasma, they have shown marked increases in exhaust velocity which is in direct proportion to the magnitude of the applied field. Further, recent results have shown that electrode erosion may be influenced by applied magnetic fields. This research effort is directed to the experimental and computational study of the effects of applied magnetic field nozzles in the acceleration of plasma flows. Plasma source devices which eliminate the plasma interaction in normal thrusters are studied as most basic. Normal thruster configurations were studied without applied fields and with applied magnetic nozzle fields. Unique computational studies utilize existing codes which accurately include transport processes. Unique diagnostic studies supported the experimental studies to generate new data. Both computation and diagnostics were combined to indicate the physical mechanisms and transport properties that are operative in order to allow scaling and accurate prediction of thruster performance.

York, Thomas M.↗
DOE OSTI2022

A Common Tracking Software Project

Abstract The reconstruction of the trajectories of charged particles, or track reconstruction, is a key computational challenge for particle and nuclear physics experiments. While the tuning of track reconstruction algorithms can depend strongly on details of the detector geometry, the algorithms currently in use by experiments share many common features. At the same time, the intense environment of the High-Luminosity LHC accelerator and other future experiments is expected to put even greater computational stress on track reconstruction software, motivating the development of more performant algorithms. We present here A Common Tracking Software (ACTS) toolkit, which draws on the experience with track reconstruction algorithms in the ATLAS experiment and presents them in an experiment-independent and framework-independent toolkit. It provides a set of high-level track reconstruction tools which are agnostic to the details of the detection technologies and magnetic field configuration and tested for strict thread-safety to support multi-threaded event processing. We discuss the conceptual design and technical implementation of ACTS, selected applications and performance of ACTS, and the lessons learned.

97 MATHEMATICS AND COMPUTING↗
NASA NTRS2019

Cloud Computing Methods for Near Rectilinear Halo Orbit Trajectory Design

Complicated mission design problems require innovative computational solutions. As spacecraft depart from a proposed Gateway in a Near Rectilinear Halo Orbit (NRHO), recontact analysis is required to avoid risk of collision and ensure safe operations. Escape dynamics from NRHOs are governed by multiple gravitational bodies, yielding a trajectory design space that is exhaustively large. This paper summarizes the recontact analysis for departure from the NRHO and describes how the Deep Space Trajectory Explorer (DSTE) trajectory design software incorporates high performance cloud computing to compute and visualize the orbit design space. Recent focus on exploration missions to cislunar space has kindled accelerated interest in multibody orbit solutions. Trajectory analysis in the presence of multiple gravity fields is complex, and innovative computational tools are needed to simplify complicated design spaces, to generate large quantities of data quickly, and to visualize the output for user accessibility. The Gateway mission is a prime example. The Gateway1 is proposed as a human outpost in deep space. The current baseline orbit for the Gateway is a Near Rectilinear Halo Orbit (NRHO) near the Moon.2 The NRHO exists in a regime that experiences the gravitational effects of the Earth and the Moon simultaneously, complicating orbit analysis. The mission design process benefits greatly from updated computational tools for multibody missions like the Gateway. As an example, consider the problem of assessing the risk of collision in an NRHO. As a staging location to missions to the lunar surface and beyond the Earth-Moon system, the Gateway will experience spacecraft and other objects regularly arriving and departing. Departing objects potentially include spent logistics modules, visiting crew vehicles, debris objects, wastewater particles, and cubesats. Each departure is governed by the dynamics of the Gateway orbit and the surrounding dynamical environment. Over time, any unmaintained object in such an orbit eventually departs due to the small instabilities associated with the NRHOs. A separation maneuver speeds the departure from the NRHO, but the effects of the maneuver on the spacecraft behavior depend on the location, magnitude, and direction of the burn. Escape dynamics from the NRHO with regard to these maneuver options open up an enormous potential trajectory design space where subtle changes in input can produce dramatically large changes in the results. Any departing object must avoid recontacting the Gateway as it leaves the lunar vicinity, and a recontact analysis thus involves a significant number of computations and extensive output data. To explore the dynamics of this extensive design space, the Deep Space Trajectory Explorer3 (DSTE) trajectory design software incorporates new High Performance Computing (HPC) services and novel interactive visualizations. This paper details the HPC and cloud infrastructure techniques that are implemented in the DSTE, applying the new capabilities to analysis of recontact risk with the Gateway in NRHO. NEAR RECTILINEAR HALO ORBITS The Gateway is planned to fly in a lunar NRHO as its baseline orbit. The NRHO families of orbits are subsets of the larger halo families, which originate from planar orbits near the L1 and L2 libration points; the Earth-Moon L2 halo family appears in Figure 1. Each halo orbit is perfectly periodic in the Circular Restricted 3-Body Problem (CR3BP) and becomes a quasi-periodic orbit in a higher fidelity ephemeris force model. The NRHOs are defined as those members of the halo family with bounded stability properties;2 they pass near the Moon at perilune and are nearly polar. Families exist with apolunes located both above the lunar north pole and above the lunar south pole; the Gateway is planned to reside in a southern L2 NRHO in a 9:2 resonance with the lunar synodic period. The 9:2 NRHO is characterized by a period of about 6.5 days, a perilune radius of about 3,500 km, and an apolune radius of about 71,000 km; it is strongly affected by the gravity of both the Earth and the Moon simultaneously. This NRHO offers extended communications with assets on the south pole of the Moon,4 as well as low-cost orbit maintenance and attitude control,5 favorable eclipse avoidance properties,6 and inexpensive transfers from Earth and to other destinations.5,7 The NRHO portion of the southern L2 halo family is highlighted in black in Figure 1, and the 9:2 NRHO appears in blue.

Phillips, Sean M.↗
DOE OSTI2021

Transforming Energy through Computational Science: Computing for Clean Energy

This fact sheet discusses the opportunity space for NREL's computational science capabilities to address national clean energy objectives. Achieving a carbon-free power sector by 2035 as a step towards a decarbonized U.S. energy economy in 2050 will require major advances in power generation, autonomous energy systems, transportation, and buildings/communities. Development of integrated modeling approaches for complex energy systems will be essential for deployment. Success requires developments in optimization and control theory, complemented by machine learning (ML) and artificial intelligence (AI), all of which in turn need targeted investments in breadth and scale of computing. This document defines an opportunity space where: embracing computing can link established research and development (R&D) to a decarbonization agenda; pursuing emerging approaches can accelerate the pace of technology advancement across the portfolio; and leading by example could reduce the carbon footprint of computing worldwide.

advanced computing↗
DOE OSTI2021

A Taxonomy for Classification and Comparison of Dataflows for GNN Accelerators

Recently, Graph Neural Networks (GNNs) have received a lot of interest because of their success in learning representations from graph structured data. However, GNNs exhibit different compute and memory characteristics compared to traditional Deep Neural Networks (DNNs). Graph convolutions require feature aggregations from neighboring nodes (known as the aggregation phase), which leads to highly irregular data accesses. GNNs also have a very regular compute phase that can be broken down to matrix multiplications (known as the combination phase). All recently proposed GNN accelerators utilize different dataflows and microarchitecture optimizations for these two phases. Different communication strategies between the two phases have been also used. However, as more custom GNN accelerators are proposed, the harder it is to qualitatively classify them and quantitatively contrast them. In this work, we present a taxonomy to describe several diverse dataflows for running GNN inference on accelerators. This provides a structured way to describe and compare the design-space of GNN accelerators.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

Bulky Phosphine Ligands Promote Palladium-Catalyzed Protodeboronation

The Suzuki-Miyaura cross-coupling reaction is plagued by protodeboronation, an undesirable side reaction with water that consumes the boronic acid derivatives required for the cross-coupling reaction. Meticulous mechanistic studies have previously established protodeboronation to be highly sensitive to the nature of the boronic reagent and reaction conditions. Particularly, the presence of bases, which are essential for the Suzuki-Miyaura coupling, is known to catalyze protodeboronation. However, protodeboronation catalyzed by palladium-phosphine complexes, the benchmark catalyst system for Suzuki-Miyaura cross-coupling, has been understudied compared to its base-catalyzed counterpart. Here, we demonstrate, using automated high-throughput experimentation, comprehensive computational mechanistic analyses and kinetic modeling, that protodeboronation is accelerated by palladium(II) complexes bound to bulky phosphine ligands. While sterically hindered ligands are typically used to facilitate difficult cross-couplings, these ligands can instead paradoxically impede cross-coupling product formation, requiring careful and judicious consideration when choosing ligands for Suzuki-Miyaura cross-couplings.

Ser, Cher Tian [University of Toronto, ON (Canada)↗
DOE OSTI2025

High-Fidelity CFD Simulation of Mixed Convection and Forced Convection in a Pebble Bed Test Reactor Core

The Hermes low-power [35-MW(thermal)] reactor will be built and operated by Kairos Power LLC (KP) to demonstrate its fluoride salt-cooled high-temperature reactor (FHR) technology. In the KP FHR, the reactor core is composed of randomly packed pebbles with TRISO fuel particles inside with FLiBe flow upward through the core acting as a coolant. Previous numerical and experimental studies have been limited to either a small-size bed or to a lack of detailed measurements for heat transfer. Here, to address the lack of high-fidelity heat transfer data in a real-size FHR core, in this study, we simulated a pebble bed core with 34 374 pebbles randomly packed, similar to the Hermes reactor's size. The core radius was 14 times that of the pebble diameter, while the core height was 45 times. In this work, we were particularly interested in a mixed convection regime, where buoyancy is important. Therefore, we performed several large-eddy simulations at different Reynolds numbers (160 to 1000) with gravitational force included. The spectral element computational fluid dynamics code NekRS with graphics processing unit acceleration was used for this study. The low-Mach number approximation was applied to address property changes in the FLiBe and to account for buoyancy. A pure hexahedral mesh with 60 million elements was generated by the Voronoi cell method. At the polynomial order of 5, the total degrees of freedom was 7.5 billion. The developed case in this work is the first of its kind in terms of size and complexity. The local numerical data across the domain were obtained and compared with empirical correlations. After examining the data, we found the following conclusions. For pressure drop, the Reger correlation predicted less than a 5% error. On the other hand, for heat transfer, the Wakao correlation outperformed the others. Based on our findings, we recommend the use of the Wakao correlation for the Nusselt number calculation, and for pressure drop, the KTA (Kerntechnischer Ausschuss) correclation, among the available experimental correlations. In conclusion, the Reger direct numerical simulation-driven correlation for pressure drops should also be considered, given its best agreement with our calculations.

Mixed Convection↗
DOE OSTI2025

Synthesis of single-qutrit circuits from Clifford+𝑅 gates

Here, we present two deterministic compilation algorithms for single-qutrit unitaries with O ( log 1 / ɛ ) gate depth. Each algorithm selects a nearby approximation to the target unitary and then exactly synthesizes the approximation over the Clifford + R basis. The first algorithm exhaustively searches over the group; while the second algorithm searches only for Householder reflections. The exhaustive search algorithm yields an average R count of 2.193 ( 11 ) + 8.621 ( 7 ) log 10 ( 1 / ɛ ) , albeit with a time complexity of O ( ɛ − 4.4 ) . The Householder search algorithm results in a larger average R count of 3.20 ( 13 ) + 10.77 ( 3 ) log 10 ( 1 / ɛ ) at a reduced time complexity of O ( ɛ − 0.42 ) , greatly extending the reach in ɛ . These costs correspond asymptotically to 35% and 69% more non-Clifford gates compared with synthesizing the same unitary with two qubits. Such initial results are encouraging for using the R gate as the nontransversal gate for qutrit-based computation.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2021

Enabling discovery data science through cross-facility workflows

Experimental and observational instruments for scientific research (such as light sources, genome sequencers, accelerators, telescopes and electron microscopes) increasingly require High Performance Computing (HPC) scale capabilities for data analysis and workflow processing. Next-generation instruments are being deployed with higher resolutions and faster data capture rates, creating a big data crunch that cannot be handled by modest institutional computing resources. Often these big data analysis pipelines also require near real-time computing and have higher resilience requirements than the simulation and modeling workloads more traditionally seen at HPC centers. While some facilities have enabled workflows to run at a single HPC facility, there is a growing need to integrate capabilities across HPC facilities to enable cross-facility workflows, either to provide resilience to an experiment, increase analysis throughput capabilities, or to better match a workflow to a particular architecture. In this paper we describe the barriers to executing complex data analysis workflows across HPC facilities and propose an architectural design pattern for enabling scientific discovery using cross-facility workflows that includes orchestration services, application programming interfaces (APIs), data access and co-scheduling.

Antypas, Katerina B.↗
DOE OSTI2023

ESnet/JLab FPGA Accelerated Transport

To increase the science rate for high data rates/volumes, Thomas Jefferson National Accelerator Facility (JLab) has partnered with Energy Sciences Network (ESnet) to define an edge to data center traffic shaping / steering transport capability featuring data event aware network shaping and forwarding. The keystone of this ESnet+JLab FPGA Accelerated Transport (EJFAT) is the joint development of an AI/ML directed dynamic compute work Load Balancer (LB) of UDP streamed data. The LB is a suite consisting of a Field Programmable Gate Array (FPGA) executing the dynamically configurable, low fixed latency LB data plane featuring real-time packet redirection and high throughput, and a control plane running on the FPGA host computer that monitors network and compute farm telemetry in order to make dynamic AI/ML guided decisions for destination compute host redirection/load balancing and destination resource provisioning. The LB provides for three-tier horizontal scaling across LB suites, core compute hosts, and CPUs within a host. The LB effectively provides seamless integration of edge/core computing to support direct experimental data processing for immediate use by JLab science programs and others such as the EIC as well as data centers of the future requiring high throughput and low latency for both hot and cooled data for both running experiment data acquisition systems and data center use cases.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Evolution of the SLATE linear algebra library

SLATE (Software for Linear Algebra Targeting Exascale) is a distributed, dense linear algebra library targeting both CPU-only and GPU-accelerated systems, developed over the course of the Exascale Computing Project (ECP). While it began with several documents setting out its initial design, significant design changes occurred throughout its development. In some cases, these were anticipated: an early version used a simple consistency flag that was later replaced with a full-featured consistency protocol. In other cases, performance limitations and software and hardware changes prompted a redesign. Sequential communication tasks were parallelized; host-to-host MPI calls were replaced with GPU device-to-device MPI calls; more advanced algorithms such as Communication Avoiding LU and the Random Butterfly Transform (RBT) were introduced. Early choices that turned out to be cumbersome, error prone, or inflexible have been replaced with simpler, more intuitive, or more flexible designs. Applications have been a driving force, prompting a lighter weight queue class, nonuniform tile sizes, and more flexible MPI process grids. Of paramount importance has been building a portable library that works across several different GPU architectures – AMD, Intel, and NVIDIA – while keeping a clean and maintainable codebase. Here we explore the evolving design choices and their effects, both in terms of performance and software sustainability.

Gates, Mark↗