Search NASA⌕ Search

SEARCH · Search NASA

Results for “hardware performance”

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 631 records · Page 35

Reproduction of the Yucca Mountain Project TSPA-LA Model Runs using TSPA Computing Systems

Sandia National Laboratories (SNL) conducted an evaluation of total system performance assessment (TSPA) related computing systems for the previously considered Yucca Mountain Project (YMP). This was done to maintain the operational readiness of the computing infrastructure (computer hardware and software) and knowledge capability for total system performance assessment (TSPA) type analysis, as directed by the National Nuclear Security Administration (NNSA), DOE 2010. This work is a continuation of the ongoing readiness evaluation reported in Lee and Hadgu (2014). The current work examined main components of the computing system identified in the previous work (Lee and Hadgu, 2014) to ensure the operational readiness of the TSPA-LA model capability on the server cluster. The TSPA computing hardware and storage system were replaced in late 2014 to maintain core capability and improve computation efficiency. One floating license of GoldSim Version 9.60.300 was installed on the upgraded cluster head node, and its distributed processing capability was mapped on the cluster processors. Other supporting software was tested and installed to support the TSPA-type analysis on the server cluster. All the TSPA-LA modeling cases were tested and verified for the model reproducibility on the upgraded 2014 server cluster (CL2014). All test runs were executed on multiple processors on the server cluster utilizing the GoldSim distributed processing capability, and all runs completed successfully. The model reproducibility verification was evaluated by two approaches: numerical value comparison and graphical comparison. The analysis demonstrated an excellent reproducibility of the TSPA-LA model runs on the upgraded server cluster. The 2014 server cluster and supporting software systems are fully operational to support TSPA- LA type analysis.

97 MATHEMATICS AND COMPUTING↗

Reproduction of the Yucca Mountain Project TSPA-LA Model Runs using TSPA Computing Systems

Sandia National Laboratories (SNL) conducted an evaluation of total system performance assessment (TSPA) related computing systems for the previously considered Yucca Mountain Project (YMP). This was done to maintain the operational readiness of the computing infrastructure (computer hardware and software) and knowledge capability for total system performance assessment (TSPA) type analysis, as directed by the National Nuclear Security Administration (NNSA), DOE 2010. This work is a continuation of the ongoing readiness evaluation reported in Lee and Hadgu (2014). The current work examined main components of the computing system identified in the previous work (Lee and Hadgu, 2014) to ensure the operational readiness of the TSPA-LA model capability on the server cluster. The TSPA computing hardware and storage system were replaced in late 2014 to maintain core capability and improve computation efficiency. One floating license of GoldSim Version 9.60.300 was installed on the upgraded cluster head node, and its distributed processing capability was mapped on the cluster processors. Other supporting software was tested and installed to support the TSPA-type analysis on the server cluster. All the TSPA-LA modeling cases were tested and verified for the model reproducibility on the upgraded 2014 server cluster (CL2014). All test runs were executed on multiple processors on the server cluster utilizing the GoldSim distributed processing capability, and all runs completed successfully. The model reproducibility verification was evaluated by two approaches: numerical value comparison and graphical comparison. The analysis demonstrated an excellent reproducibility of the TSPA-LA model runs on the upgraded server cluster. The 2014 server cluster and supporting software systems are fully operational to support TSPA- LA type analysis.

97 MATHEMATICS AND COMPUTING↗

Proxy Applications for Converged Workloads: DMC LDRD Initiative

Modern scientific applications are complicated and require coordination of several components. Proxy application driven software-hardware co-design plays a vital role in driving innovation among the developments of applications, software infrastructure and hardware architecture. Proxy applications are self-contained and simplified codes that are intended to model the performance-critical computations within applications. Applications executing on modern High Performance Computing (HPC) systems are susceptible to network congestion, insufficient memory bandwidth within and across compute nodes, and inadvertent loss of performance due to bugs and unoptimized programming models. Modern numerical simulations and machine learning models play a critical role in studying physical phenomenon under myriad uncertainties. Such applications often exhibit irregular computation and memory accesses at specific regions of the application code, which can contribute to various performance bottlenecks at scale. To mitigate such issues and prepare the next generation hardware for a variety of computation and data movement contingencies, a well-known practice is to consider "proxy" applications as representative motifs for various classes of scientific applications. While there is disagreement in the HPC community on the mechanisms of construction of the proxy applications, there is a strong consensus on their positive impact in co-design. Proxy Applications for Converged Workloads (PACER) is about facilitating software-hardware co-design through proxy applications with the goal of improving the performance of converged science workflows on heterogeneous systems.

97 MATHEMATICS AND COMPUTING↗

High Quality, Low Cost Bulk Gallium Nitride Substrates Grown by the Electrochemical Solution Growth Method

The objective of this project was to develop the Electrochemical Solution Growth (ESG) method conceived / patented at Sandia National Laboratory into a commercially viable bulk gallium nitride (GaN) growth process that can be scaled to low cost, high quality, and large area GaN wafer substrate manufacturing. The goal was to advance the ESG growth technology by demonstrating rotating seed growth at the lab scale and then transitioning process to prototype commercial system, while validating the GaN material and electronic / optical device quality. The desired outcome of the project is a prototype commercial process for US-based manufacturing of high quality, large area, and lower cost GaN substrates that can drive widespread deployment of energy efficient GaN-based power electronic and optical devices. In year 1 of the project (Sept 2012 – Dec 2013) the overall objective was to demonstrate crystalline GaN growth > 100um on a GaN seed crystal. The development plan included tasks to demonstrate and implement a method for purifying reagent grade salts, develop the reactor 1 process for rotating seed Electrochemical Solution Growth (ESG) of GaN, grow and characterize ESG GaN films, develop a fluid flow and reaction chemistry model for GaN film growth, and design / build an improved growth reactor capable of scaling to 50mm seed diameter. The first year’s project objectives were met in some task areas including salt purification, film characterization, modeling, and reactor 2 design / fabrication. However, the key project objective of the growth of a crystalline GaN film on the seed template was not achieved. Amorphous film growth on the order of a few tenths of a micron has been detected with a film composition including Ga and N, plus several other impurities originating from the process solution and hardware. The presence of these impurities, particularly the oxygen, has inhibited the demonstration of crystalline GaN film growth on the seed template. However, the presence of both Ga and N at the growth surface indicates that the reactor hardware physics is all functioning properly; achieving film growth is a matter of controlling the chemistry at the interface. The impurities originating from the hardware are expected to be straightforward to eliminate. Activities were defined for an extension of budget period 1 to eliminate the undesired impurities originating from the reactor hardware and interfering with crystalline GaN film growth. The budget period 1 extension was negotiated during the 1st half of 2014. The budget period 1 extension spanned approximately from August 2014 to August 2015. The project objective for this extension period was to demonstrate at least 0.5um crystalline GaN film on a GaN seed in the lab scale reactor. The focus of the budget 1 extension period from August 2014 to August 2015 was to eliminate oxygen contamination interference with GaN film growth. The team procured the highest purity lowest oxygen salt for testing. Low oxygen crucible materials such as silicon carbide were installed and evaluated in the laboratory reactor. Growth experiments were performed with high purity salt, high purity hardware, and optimized oxide removal from the seed surface. Experiments were characterized with methods including UV inspection, profilometry, x-ray diffraction (XRD) to determine crystalline structure, optical and scanning electron microscopy, photoluminescence, x-ray photon spectroscopy (XPS), transmission electron microscopy (TEM), and secondary ion mass spectroscopy (SIMS). Despite successfully integrating the low oxygen materials in the laboratory reactor, the goal of depositing 0.5um of crystalline GaN on the MOCVD GaN seed was not met. Very thin (ca. 10nm) cubic phase GaN deposition was observed on the hexagonal MOCVD GaN seeds. But there was a competing etching reaction which was also observed and thought to be related to the presence of metallic lithium, a byproduct of the LiCl-KCl salt used as the process medium. The etching reaction could potentially be addressed by alternate salts not containing lithium, but would necessitate starting all over on the reactor and process design. Further, controlling the reaction of Ga and N in the bulk salt to favor deposition on the seed has proved to be very difficult and unlikely to be solved within the scope of this project in a manner consistent with the original objective for wafer or crystal scale thickness for GaN deposition on a GaN seed. Upon completion of the budget 1 extension period in August 2015 the project partners and DOE agreed to stop work on the project.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

A CHIL Validation of Machine Learning-Assisted Methods for Real-Time Controls of Solar PV for Grid Services

Recent research has highlighted the potential for solar to act as a zero-marginal-cost and zero-emission flexibility resource on the bulk power system when operated with advanced control systems. To increase the performance of these systems, leading technologies, including machine learning (ML) and hierarchical inverter set point allocation, have been proposed; however, these technologies lack comprehensive validation under real-world application scenarios. This paper addresses this gap by designing and developing a controller-hardware-in-the-loop framework to evaluate the performance of different flexible solar technologies in responding to automatic generation control signals in a closed-loop fashion. Simulation results indicate the superior performance of an ML-based approach compared to the conventional reference-control grouping-based approach, showcasing its potential to support grid stability and operational efficiency.

closed-loop validation↗

A CHIL Validation of Machine Learning-Assisted Methods for Real-Time Controls of Solar PV for Grid Services: Preprint

Recent research has highlighted the potential for solar to act as a zero-marginal-cost and zero-emission flexibility resource on the bulk power system when operated with advanced control systems. To increase the performance of these systems, leading technologies, including machine learning (ML) and hierarchical inverter set point allocation, have been proposed; however, these technologies lack comprehensive validation under real-world application scenarios. This paper addresses this gap by designing and developing a controller-hardware-in-the-loop framework to evaluate the performance of different flexible solar technologies in responding to automatic generation control signals in a closed-loop fashion. Simulation results indicate the superior performance of an ML-based approach compared to the conventional reference-control grouping-based approach, showcasing its potential to support grid stability and operational efficiency.

closed-loop validation↗

The effect of laser spot shapes on polar-direct-drive implosions on the National Ignition Facility

Ongoing polar-direct-drive (PDD) implosions on the National Ignition Facility (NIF) [J. D. Lindl and E. I. Moses, Phys. Plasmas 18, 050901 (2011)] use existing NIF hardware, including indirect-drive phase plates. This limits the performance achievable in these implosions. Spot shapes are identified that significantly improve the uniformity of PDD NIF implosions; outer surface deviation is reduced by a factor of 7 at the end of the laser pulse and hot-spot distortion is reduced by a factor of 2 when the shell has converged by a factor of ∼10. As a result, the neutron yield increases by approximately a factor of 2. This set of laser spot shapes is a combination of circular and elliptical spots, along with elliptical spot shapes modulated by an additional higher-intensity ellipse offset from the center of the beam. This combination is motivated in this paper. It is also found that this improved implosion uniformity is obtained independent of the heat conduction model. This work indicates that significant improvement in performance can be obtained robustly with the proposed spot shapes.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Validating the simulation of large-scale parallel applications using statistical characteristics

Simulation is a widely adopted method to analyze and predict the performance of large-scale parallel applications. Validating the hardware model is highly important for complex simulations with a large number of parameters. Common practice involves calculating the percent error between the projected and the real execution time of a benchmark program. However, in a high-dimensional parameter space, this coarse-grained approach often suffers from parameter insensitivity, which may not be known a priori. Moreover, the traditional approach cannot be applied to the validation of software models, such as application skeletons used in online simulations. In this work, we present a methodology and a toolset for validating both hardware and software models by quantitatively comparing fine-grained statistical characteristics obtained from execution traces. Although statistical information has been used in tasks like performance optimization, this is the first attempt to apply it to simulation validation. Lastly, our experimental results show that the proposed evaluation approach offers significant improvement in fidelity when compared to evaluation using total execution time, and the proposed metrics serve as reliable criteria that progress toward automating the simulation tuning process.

97 MATHEMATICS AND COMPUTING↗

Modeling the Office of Science Ten Year FacilitiesPlan: The PERI Architecture Tiger Team

The Performance Engineering Institute (PERI) originally proposed a tiger team activity as a mechanism to target significant effort to the optimization of key Office of Science applications, a model that was successfully realized with the assistance of two JOULE metric teams. However, the Office of Science requested a new focus beginning in 2008: assistance in forming its ten year facilities plan. To meet this request, PERI formed the Architecture Tiger Team, which is modeling the performance of key science applications on future architectures, with S3D, FLASH and GTC chosen as the first application targets. In this activity, we have measured the performance of these applications on current systems in order to understand their baseline performance and to ensure that our modeling activity focuses on the right versions and inputs of the applications. We have applied a variety of modeling techniques to anticipate the performance of these applications on a range of anticipated systems. While our initial findings predict that Office of Science applications will continue to perform well on future machines from major hardware vendors, we have also encountered several areas in which we must extend our modeling techniques in order to fulfill our mission accurately and completely. In addition, we anticipate that models of a wider range of applications will reveal critical differences between expected future systems, thus providing guidance for future Office of Science procurement decisions, and will enable DOE applications to exploit machines in future facilities fully.

99 GENERAL AND MISCELLANEOUS↗

Space Nuclear Power Autonomous Control Algorithm and Control Element Test Bed

Nuclear thermal rockets are currently NASA’s preferred option for use in a manned mission to Mars in the 2040s. The communication delay between an Earth ground station and a spacecraft heading toward Mars can be up to 20 min. Therefore, controlling the nuclear rocket engine would require either a full-time reactor operator on the mission or an autonomous control system for the reactor. The latter idea of making space nuclear reactors fully autonomous has drawn more interest from stakeholders, but such an autonomous control system must be rigorously tested and validated before it is certified for human use. The cost of a full ground test for a space nuclear reactor is tremendous, so a nonnuclear mock reactor test bed was created to test and validate control elements and control algorithms for space nuclear reactors. The test bed consists of control element hardware that inputs physical measurement data into a reactor emulator to produce the reactor’s performance under steady-state, transient, and fault conditions. The control element hardware consists of six full-sized control drums equipped with servo drives and motors and is instrumented with optical encoders, resolvers, and torque sensors for drum movement characterization. In addition to the drums, a two-phase flow loop was designed and built to mimic the valves and turbomachinery associated with the propellant flow through a nuclear thermal rocket engine; components such as pressure sensors, flow meters, thermocouples, and tachometers are instrumented throughout the loop to characterize the fluid flow, valve, and turbomachinery behavior of the system. The data from the physical hardware (e.g., drum position, propellant flow rates) are input to a nuclear reactor simulator to determine the actual nuclear reactor parameters, and the data are sent back to a control algorithm to complete the control loop. The ability to conduct numerous tests of the control systems and autonomous algorithms can help validate the instrumentation and control aspects for a space nuclear reactor for every possible fault situation.

Wilson, Brandon↗

A CHIL Validation of Machine Learning-Assisted Methods for Real-Time Controls of Solar PV for Grid Services

Recent research has highlighted the potential for solar to act as a zero-marginal-cost and zero-emission flexibility resource on the bulk power system when operated with advanced control systems. To increase the performance of these systems, leading technologies, including machine learning (ML) and hierarchical inverter set point allocation, have been proposed; however, these technologies lack comprehensive validation under real-world application scenarios. This paper addresses this gap by designing and developing a controller-hardware-in-the-loop framework to evaluate the performance of different flexible solar technologies in responding to automatic generation control signals in a closed-loop fashion. Simulation results indicate the superior performance of an ML-based approach compared to the conventional reference-control grouping-based approach, showcasing its potential to support grid stability and operational efficiency.

14 SOLAR ENERGY↗

Optimizing the performance of reactive molecular dynamics simulations for many-core architectures

Reactive molecular dynamics simulations are computationally demanding. Reaching spatial and temporal scales where interesting scientific phenomena can be observed requires efficient and scalable implementations on modern hardware. In this article, we focus on optimizing the performance of the widely used LAMMPS/ReaxC package for many-core architectures. As hybrid parallelism allows better leverage of the increasing on-node parallelism, we adopt thread parallelism in the construction of bonded and nonbonded lists and in the computation of complex ReaxFF interactions. To mitigate the I/O overheads due to large volumes of trajectory data produced and to save users the burden of post-processing, we also develop a novel in situ tool for molecular species analysis. We analyze the performance of the resulting ReaxC-OMP package on two different architectures: (i) Mira, an IBM Blue Gene/Q system and (ii) Cori-II, a Cray XC-40 sytem with Knights Landing processors. For Pentaerythritol tetranitrate (PETN) systems of sizes ranging from 32 thousand to 16.6 million particles, we observe speedups in the range of 1.5-4.5x. We observe sustained performance improvements for up to 262,144 cores (1,048,576 processes) of Mira and a weak scaling efficiency of 91.5% in large simulations containing 16.6 million particles. Finally, the in situ molecular species analysis tool incurs only insignificant overheads across various system sizes and runs configurations.

97 MATHEMATICS AND COMPUTING↗

Final Report PetaScale Application Development Analysis Grant Number DE-FG02-04ER25629

The results obtained from this project will fundamentally change the way we look at computer performance analysis. These results are made possible by the precise definition of a consistent system of measurement with a set of primary units designed specifically for computer performance analysis. This system of units, along with their associated dimensions, allows us to apply the methods of dimensional analysis, based on the Pi Theorem, to define scaling and self-similarity relationships. These relationships reveal new insights into experimental results that otherwise seems only vaguely correlated. Applying the method to cache-miss data revealed scaling relationships that were not seen by those who originally collected the data. Applying dimensional analysis to the performance of parallel numerical algorithms revealed that computational force is a unifying concept for understanding the interaction between hardware and software. The efficiency of these algorithms depends, in a very intimate way, on the balance between hardware forces and software forces. Analysis of five different algorithms showed that performance analysis can be reduced to a study of the differential geometry of the efficiency surface. Each algorithm defines a set of curvilinear coordinates, specific to that algorithm, and different machines follow different paths along the surface depending on the difference in balance between hardware forces and software forces. Two machines with the same balance in forces follow the same path and are self-similar. The most important result from the project is the statement of the Principle of Computational Least Action. This principle follows from the identification of a dynamical system underlying computer performance analysis. Instructions in a computer are modeled as a classical system under the influence of computational forces. Each instruction generates kinetic energy during execution, and the sum of the kinetic energy for all instructions produces a kinetic energy spectrum as a function of time. These spectra look very much like the spectra used by chemists to analyze properties of molecules. Large spikes in the spectra reveal events during execution, like cache misses, that limit performance. The area under the kinetic energy spectrum is the computational action generated by the program. This computational action defines a normed metric space that measures the size of a program in terms of its action norm and the distance between programs in terms of the norm of the difference of their action. This same idea can be applied to a set of programmers writing code and leads to a computational action metric that measures programmer productivity. In both cases, experimental evidence suggests that highly efficient programs and highly productive programmers generate the least computational action.

97 MATHEMATICS AND COMPUTING↗

Extreme Scale Computing to Secure the Nation

Since the dawn of modern electronic computing in the mid 1940's, U.S. national security programs have been dominant users of every new generation of high-performance computer. Indeed, the first general-purpose electronic computer, ENIAC (the Electronic Numerical Integrator and Computer), was used to calculate the expected explosive yield of early thermonuclear weapons designs. Even the U. S. numerical weather prediction program, another early application for high-performance computing, was initially funded jointly by sponsors that included the U.S. Air Force and Navy, agencies interested in accurate weather predictions to support U.S. military operations. For the decades of the cold war, national security requirements continued to drive the development of high performance computing (HPC), including advancement of the computing hardware and development of sophisticated simulation codes to support weapons and military aircraft design, numerical weather prediction as well as data-intensive applications such as cryptography and cybersecurity U.S. national security concerns continue to drive the development of high-performance computers and software in the U.S. and in fact, events following the end of the cold war have driven an increase in the growth rate of computer performance at the high-end of the market. This mainly derives from our nation's observance of a moratorium on underground nuclear testing beginning in 1992, followed by our voluntary adherence to the Comprehensive Test Ban Treaty (CTBT) beginning in 1995. The CTBT prohibits further underground nuclear tests, which in the past had been a key component of the nation's science-based program for assuring the reliability, performance and safety of U.S. nuclear weapons. In response to this change, the U.S. Department of Energy (DOE) initiated the Science-Based Stockpile Stewardship (SBSS) program in response to the Fiscal Year 1994 National Defense Authorization Act, which requires, 'in the absence of nuclear testing, a progam to: (1) Support a focused, multifaceted program to increase the understanding of the enduring stockpile; (2) Predict, detect, and evaluate potential problems of the aging of the stockpile; (3) Refurbish and re-manufacture weapons and components, as required; and (4) Maintain the science and engineering institutions needed to support the nation's nuclear deterrent, now and in the future'. This program continues to fulfill its national security mission by adding significant new capabilities for producing scientific results through large-scale computational simulation coupled with careful experimentation, including sub-critical nuclear experiments permitted under the CTBT. To develop the computational science and the computational horsepower needed to support its mission, SBSS initiated the Accelerated Strategic Computing Initiative, later renamed the Advanced Simulation & Computing (ASC) program (sidebar: 'History of ASC Computing Program Computing Capability'). The modern 3D computational simulation capability of the ASC program supports the assessment and certification of the current nuclear stockpile through calibration with past underground test (UGT) data. While an impressive accomplishment, continued evolution of national security mission requirements will demand computing resources at a significantly greater scale than we have today. In particular, continued observance and potential Senate confirmation of the Comprehensive Test Ban Treaty (CTBT) together with the U.S administration's promise for a significant reduction in the size of the stockpile and the inexorable aging and consequent refurbishment of the stockpile all demand increasing refinement of our computational simulation capabilities. Assessment of the present and future stockpile with increased confidence of the safety and reliability without reliance upon calibration with past or future test data is a long-term goal of the ASC program. This will be accomplished through significant increases in the scientific bases that underlie the computational tools. Computer codes must be developed that replace phenomenology with increased levels of scientific understanding together with an accompanying quantification of uncertainty. These advanced codes will place significantly higher demands on the computing infrastructure than do the current 3D ASC codes. This article discusses not only the need for a future computing capability at the exascale for the SBSS program, but also considers high performance computing requirements for broader national security questions. For example, the increasing concern over potential nuclear terrorist threats demands a capability to assess threats and potential disablement technologies as well as a rapid forensic capability for determining a nuclear weapons design from post-detonation evidence (nuclear counterterrorism).

99 GENERAL AND MISCELLANEOUS↗

RACB: Resource Aware Cache Bypass on GPUs

Caches are universally used in computing systems to hide long off-chip memory access latencies. Unlike CPUs, massive threads running simultaneously on GPUs bring a tremendous pressure on memory hierarchy. As a result, the limitation of cache resources becomes a bottleneck for a GPU to exploit thread-level parallelism (TLP) and memory-level parallelism (MLP) and achieve high performance. In this paper, we propose a mechanism to bypass L1D and L2 cache based on the availability of cache resources. Our proposed mechanism is based on the observation that a huge number of stalls coming from limited cache resources prohibit GPUs from providing a higher throughput. So we propose Resource Aware Cache Bypass (RACB) with minor hardware changes to eliminate such stalls to improve performance. We examine the effectiveness of this approach when applied to L1D and L2 cache separately as well as together. Evaluation results with NVIDIA Computing SDK show that RACB generally improves performance the most when applied to both L1D and L2 cache, which is up to 88.05% and on an average of 16.73%, additionally, energy is saved up to 22.35% and on an average of 5.88% with minor hardware overheads.

Dai, Hongwen↗

Practical Insights on Applying Simulation-Based Control Methods in Experimental Studies

Advanced nuclear reactors are crucial to the future of energy both in the United States and around the globe. In contrast to the current operating fleet, they are characterized as being deployable in remote locations and able to operate in semi-autonomous or autonomous fashion. This leap forward necessitates a new reactor control paradigm. Because advanced nuclear reactors are still under development in the United States, the creation of new control methods to achieve autonomous operations has been based on systems modeling and simulation. However, an important factor in successfully deploying these new control methods is the ability to seamlessly transition from simulation environments to real-world settings. Control methods tested in both simulation and experimental settings need to be investigated in the context of advanced reactor applications. This work developed a series of simple controllers for Idaho National Laboratory (INL)’s Microreactor Applications Research Validation and Evaluation (MARVEL) microreactor operating in load-following scenarios. These controllers were tested in both simulation and experimental settings, and a comparative performance analysis was performed. The simulation tests leveraged the Control and Optimization Modular Modeling Application for Nuclear Deployment (COMMAND) software developed in a previous stage of the current effort, along with the MARVEL Reactor Excursion and Leak Analysis Program (RELAP5-3D) and Monte Carlo N-Particle (MCNP) models. The experimental tests leveraged the COMMAND software, MARVEL models, and the U.S. Department of Energy Microreactor Program’s Microreactor Automated Control System (MACS). MACS was developed to serve as a control method testbed. It was customized to mirror the MARVEL microreactor, and COMMAND enabled MACS to emulate the physics of MARVEL. The load-following controller was developed using the simulation platform, with efforts to emulate real systems by introducing actuator saturation and noise. These factors were incrementally accounted for in the controller design. After finalizing the controller design, it was implemented with the experimental setup. The experimental conditions tested included an initial test under conditions similar to the final simulation test, and two additional scenarios. The first scenario introduced additional actuator saturation to account for equipment aging over time, which was unknown to the controller. The second scenario introduced sensor delay, a phenomenon anticipated with the use of remote operations or wireless communication in advanced reactors. These tests revealed several notable differences. While the controller performed well in simulation, it exhibited several limitations when transitioning to hardware. The main challenges involved maintaining the steady-state target power, as evidenced by larger error values between the true reactor power and setpoint power, as well as persistent oscillations in controlled reactor power. These issues could lead to unacceptable transient conditions in real reactor testing. Introducing actuator aging and stochastic delays in the experimental setup significantly impacted controller performance, resulting in increased overshoot and undershoot, and exacerbated error and oscillations previously mentioned. These findings underscore the importance of experimental testbeds for testing and validating control methods, as controllers developed using only theory and/or simulation may perform unexpectedly when applied to actual hardware. This research emphasizes the need for an experimental testbed for achieving such validation.

46 - INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AN↗

Mixed-precision iterative refinement using tensor cores on GPUs to accelerate solution of linear systems

Double-precision floating-point arithmetic (FP64) has been the de facto standard for engineering and scientific simulations for several decades. Problem complexity and the sheer volume of data coming from various instruments and sensors motivate researchers to mix and match various approaches to optimize compute resources, including different levels of floating-point precision. In recent years, machine learning has motivated hardware support for half-precision floating-point arithmetic. A primary challenge in high-performance computing is to leverage reduced-precision and mixed-precision hardware. We show how the FP16/FP32 Tensor Cores on NVIDIA GPUs can be exploited to accelerate the solution of linear systems of equations Ax = b without sacrificing numerical stability. The techniques we employ include multiprecision LU factorization, the preconditioned generalized minimal residual algorithm (GMRES), and scaling and auto-adaptive rounding to avoid overflow. We also show how to efficiently handle systems with multiple right-hand sides. On the NVIDIA Quadro GV100 (Volta) GPU, we achieve a 4×-5× performance increase and 5× better energy efficiency versus the standard FP64 implementation while maintaining an FP64 level of numerical stability.

GMRES↗

Operational experience with the CMS data acquisition system

The data-acquisition (DAQ) system of the CMS experiment at the LHC performs the read-out and assembly of events accepted by the first level hardware trigger. Assembled events are made available to the high-level trigger (HLT), which selects interesting events for offline storage and analysis. The system is designed to handle a maximum input rate of 100 kHz and an aggregated throughput of 100 GB/s originating from approximately 500 sources and 10^8 electronic channels. An overview of the architecture and design of the hardware and software of the DAQ system is given. We report on the performance and operational experience of the DAQ and its Run Control System in the first two years of collider runs of the LHC, both in proton-proton and Pb-Pb collisions. We present an analysis of the current performance, its limitations, and the most common failure modes and discuss the ongoing evolution of the HLT capability needed to match the luminosity ramp-up of the LHC.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗