SEARCH · Engineering Papers

Results for “hardware efficiency”

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.

SourceResearch connectionFrom yearThrough year

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

At least 757 records · Page 42

All sources NASA NTRS DOE OSTI

DOE OSTI2020

Reliability Analysis for Unreliable FSM Computations

Finite State Machines (FSMs) are fundamental in both hardware design and software development. However, the reliability of FSM computations remains poorly understood. This becomes increasingly critical as the technology scaling becomes more aggressive. Existing reliability analyses are mainly designed for generic computations and are unaware of the special error tolerance characteristics in FSM computations. In this work, we introduce RelyFSM – a state-level reliability analysis framework for FSM computations. By qualitatively reasoning about the transition structures in FSM computations, RelyFSM can precisely capture their inherent error tolerance. Basically, RelyFSM takes an FSM specification as input and analyzes its reliability for a specific execution. At the core of the analysis is a set of mathematical models that accurately define the behaviors of unreliable FSM executions. Our evaluation with a spectrum of FSM benchmarks confirms the accuracy and efficiency of RelyFSM in analyzing the reliability of real-world FSM computations.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2023

Binary operations on neuromorphic hardware with application to linear algebraic operations and stochastic equations

Abstract Non-von Neumann computational hardware, based on neuron-inspired, non-linear elements connected via linear, weighted synapses—so-called neuromorphic systems—is a viable computational substrate. Since neuromorphic systems have been shown to use less power than CPUs for many applications, they are of potential use in autonomous systems such as robots, drones, and satellites, for which power resources are at a premium. The power used by neuromorphic systems is approximately proportional to the number of spiking events produced by neurons on-chip. However, typical information encoding on these chips is in the form of firing rates that unarily encode information. That is, the number of spikes generated by a neuron is meant to be proportional to an encoded value used in a computation or algorithm. Unary encoding is less efficient (produces more spikes) than binary encoding. For this reason, here we present neuromorphic computational mechanisms for implementing binary two’s complement operations. We use the mechanisms to construct a neuromorphic, binary matrix multiplication algorithm that may be used as a primitive for linear differential equation integration, deep networks, and other standard calculations. We also construct a random walk circuit and apply it in Brownian motion simulations. We study how both algorithms scale in circuit size and iteration time.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2021

Computational Offload with BlueField Smart NICs

The recent introduction of a new generation of "smart NICs" have provided new accelerator platforms that include CPU cores or reconfigurable fabric in addition to traditional networking hardware and packet offloading capabilities. While there are currently several proposals for using these smartNICs for low-latency, in-line packet processing operations, there remains a gap in knowledge as to how they might be used as computational accelerators for traditional high-performance applications. This work aims to look at benchmarks and mini-applications to evaluate possible benefits of using a smartNIC as a compute accelerator for HPC applications. We investigate NVIDIA's current-generation BlueField-2 card, which includes eight Arm CPUs along with a small amount of storage, and we test the networking and data movement performance of these cards compared to a standard Intel server host. We then detail how two different applications, YASK and miniMD can be modified to make more efficient use of the BlueField-2 device with a focus on overlapping computation and communication for operations like neighbor building and halo exchanges. Our results show that while the overall compute performance of these devices is limited, using them with a modified miniMD algorithm allows for potential speedups of 5 to 20% over the host CPU baseline with no loss in simulation accuracy.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2023

Efficiency and Output Power Sensitivity to Control Parameter Variations in an LCC-Series Wireless Power Transfer System

In wireless power transfer (WPT) systems, output power (Pout) can be controlled through input parameters such as, input voltage (Vin) , duty cycle (d) , and switching frequency (fsw) . The impact of these control parameters on the overall system efficiency has not been explored in previous studies. This paper focuses on the efficiency (η) and Pout sensitivity to these control parameter variations in an LCC-Series tuned WPT system. There are multiple combinations of Vin, d , and fsw that can achieve a specified target Pout . However, it is important to determine the optimal combination of these control variables that will yield the maximum η for a target Pout . A MATLAB/Simulink model is built to calculate and measure η and Pout . In order to achieve this result, several simulations are run where Vin,d , and fsw are swept across a wide range to cover all possible combinations. MATLAB's Sensitivity Analysis toolbox is used to evaluate how the parameters influence the model's output. Finally, an experimental prototype is implemented. The optimal values acquired using the sensitivity analysis is then plugged into the hardware prototype and compared with the simulation. The results obtained confirm the validity of the proposed method.

Harave, Sudarshan↗

DOE OSTI2026

Artificial Intelligence Thermostat to Detect Faults

Residential air conditioners and heat pumps often experience faults due to inadequate maintenance, which can severely reduce efficiency or even cause system failure. Common issues include dirty or clogged air filters and refrigerant leaks. These problems degrade performance and increase energy use and operating costs. This study presents a smart thermostat with embedded artificial intelligence to detect such faults and alert homeowners when maintenance is needed. The thermostat uses low-cost measurements—including return-air temperature, relative humidity, supply-air temperature, outdoor-air temperature, and condenser subcooling—to identify abnormal operations. Because different faults produce distinct response patterns, tailored algorithms are developed to recognize characteristic fault signatures. The investigation is built on a detailed co-simulation platform that couples EnergyPlus with the DOE/ORNL Heat Pump Design Model (HPDM). EnergyPlus represents the building’s dynamic environment, while HPDM is a high-fidelity, hardware-based model that can simulate fault-free performance as well as a wide range of faults, including gradual degradation such as minor refrigerant leakage. This platform provides a virtual training and testing environment that helps distinguish fault-induced behavior from normal operation and supports development of robust diagnostic algorithms. Using this framework, a Dynamic Bayesian Network was developed to identify two common faults—gradual refrigerant charge loss and indoor airflow blockage—and the AI-embedded thermostat was verified through annual building simulations.

Shen, Bo [ORNL] (ORCID:0000000336600393)↗

DOE OSTI2020

Adapting In Situ Accelerators for Sparsity With Granular Matrix Reordering

Neural network (NN) inference is an essential part of modern systems and is found at the heart of numerous applications ranging from image recognition to natural language processing. In situ NN accelerators can efficiently perform NN inference using resistive crossbars, which makes them a promising solution to the data movement challenges faced by conventional architectures. Although such accelerators demonstrate significant potential for dense NNs, they often do not benefit from sparse NNs, which contain relatively few non-zero weights. Processing sparse NNs on in situ accelerators results in wasted energy to charge the entire crossbar where most elements are zeros. To address this limitation, this paper proposes Granular Matrix Reordering (GMR): a preprocessing technique that enables an energy-efficient computation of sparse NNs on in situ accelerators. GMR reorders the rows and columns of sparse weight matrices to maximize the crossbars' utilization and minimize the total number of crossbars needed to be charged. The reordering process does not rely on sparsity patterns and incurs no accuracy loss. Finally, GMR achieves an average of 28% and up to 34% reduction in energy consumption over seven pruned NNs across four different pruning methods and network architectures.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2020

Collaborative Research: Improved Efficiency and Coupling of the Radiation Code in the ACME Earth System Model. Final Report

This final report details all work performed on the project by both project partners. This project provided support to properly couple RTE+RRTMGP, a high-performance broadband radiation code, within DOE’s Energy Exascale Earth System Model (E3SM). RTE+RRTMGP is a successor to the RRTMG radiation code, which has been widely accepted for its speed and accuracy by the global modeling community, and has been in use in the NCAR CESM for many years and was implemented in the initial version of E3SM. However, the computational cost of RRTMG remains high relative to other components in part due to its complexity and to its inefficient use of modern optimization strategies, issues that were rectified by the development of RTE+RRTMGP. Many of the accomplishment in this project necessitated significant collaboration with the E3SM development team. One focus of the project was to enhance the code’s optimization on the limited number of emerging computing systems on which the model is expected be used, including Many Integrated Core (MIC) architectures and Graphics Processing Unit (GPU) hardware. We also developed additional capabilities for RTE+RRTMGP that E3SM scientists identified as important for the planned applications of the model. The result of our project was optimization of a key physical component (radiative transfer calculations) of E3SM, directly supporting E3SM’s overarching global modeling objectives. More broadly, this project provided overall advancements in the use of radiative transfer calculations in atmospheric modeling and simulation, particularly for climate.

54 ENVIRONMENTAL SCIENCES↗

DOE OSTI2022

Rapidly Deployable Acoustic Monitoring and Localization System Based on a Low-Cost Wave Buoy Platform

The primary objective of this project is to develop a cost-effective, fit-for-purpose environmental monitoring system, “NoiseSpotter®,” that characterizes, classifies, and provides accurate location information for anthropogenic and natural sounds in near real-time. NoiseSpotter was developed to support the evaluation of potential acoustic effects of marine energy (ME) projects. By utilizing a compact array of three acoustic particle motion sensors, NoiseSpotter triangulates individual bearings to provide sound source localization to within 5% accuracy, allowing the ability to discern ME device sounds relative to other confounding sounds in the environment, while providing location estimates to nearby marine mammals for environmental mitigation purposes. The ME industry needs proven solutions to meet environmental impact assessment needs. The NoiseSpotter® includes off-the-shelf, modular components that are easy to assemble and disassemble. Its acoustic particle motion sensors are commercially available and the data logger and real-time telemetry system is designed to be plug-and-play. The entire system is relatively compact and can be deployed from small vessels. NoiseSpotter’s near real-time capability enables operational monitoring of ME sounds, particularly during early stages of technology adoption to facilitate mitigation of potential noise effects. Widespread adoption of the technology for acoustic monitoring of ME devices requires that it be cost effective; hence the anticipated commercial cost of system hardware is $35,000. This project contributes to reducing barriers to ME testing through support of scientific research focused on reducing or mitigating environmental risks and lowering costs and complexity of environmental monitoring. This project has developed an acoustic monitoring system, NoiseSpotter® (U.S. Patent No. 11,156,734 and U.S. Registered Trademark No. 6,442,313), to detect and characterize baseline noise and sounds from ME operations and support geolocation of detected sounds. The intended outcome of the project is to mitigate concerns about the potential for ME device noise to alter marine mammal or fish behavior. NoiseSpotter® enables cost-effective, near real-time acoustic monitoring of an operational ME device relative to ambient environmental noise and provides a technical basis for ME developers seeking to navigate the permitting process in an efficient manner.

16 TIDAL AND WAVE POWER↗

DOE OSTI2025

Q1-2024 Solar Cost Benchmarks

Each year, the U.S. Department of Energy’s (DOE) Solar Energy Technologies Office (SETO) and its national laboratory partners develop cost benchmarks for U.S. solar photovoltaic (PV) systems. These benchmarks track progress toward reducing solar costs and guide R&D priorities. Unlike typical studies that report only $/W, SETO uses intrinsic units (e.g., $/m² for mounting structures) to better capture how technology improvements such as module efficiency would impact system costs. This allows flexible modeling where inputs can vary significantly to assess cost sensitivity. Costs are reported in two ways: Minimum Sustainable Price (MSP): Long term, financially viable price under stable market conditions. Modeled Market Price (MMP): Actual market price, influenced by short term distortions such as tariffs or subsidies. Three national labs collect cost data from industry stakeholders, ensuring no duplication in outreach to stakeholders. Data reflects real transactions (primarily from Q1) and is weighted based on the number of sources per cost element. The PV System Cost Model (PVSCM) divides total installed system cost into eight categories: 1. Module (PV) 2. Inverter 3. Energy Storage System (ESS) 4. Structural BOS (SBOS) 5. Electrical BOS (EBOS) 6. Fieldwork 7. Office work 8. Other (developer/EPC costs) The first five are hardware costs, while the last three are soft costs. Each category includes fixed and variable cost components, where “size” depends on context (e.g., manufacturing capacity for modules vs. system capacity for installation costs). Variable costs are expressed using appropriate intrinsic units. The model reflects the owner’s upfront overnight capital cost, excluding tax credits. Tariffs and subsidies are treated as temporary market distortions affecting MMP but not MSP. PVSCM is implemented in Excel, where cost elements are aggregated into total system cost. Additional sheets handle unit conversions and operation & maintenance (O&M), with O&M costs levelized over the system’s lifetime.

14 SOLAR ENERGY↗

DOE OSTI2024

Improved particle-flow event reconstruction with scalable neural networks for current and future particle detectors

Abstract Efficient and accurate algorithms are necessary to reconstruct particles in the highly granular detectors anticipated at the High-Luminosity Large Hadron Collider and the Future Circular Collider. We study scalable machine learning models for event reconstruction in electron-positron collisions based on a full detector simulation. Particle-flow reconstruction can be formulated as a supervised learning task using tracks and calorimeter clusters. We compare a graph neural network and kernel-based transformer and demonstrate that we can avoid quadratic operations while achieving realistic reconstruction. We show that hyperparameter tuning significantly improves the performance of the models. The best graph neural network model shows improvement in the jet transverse momentum resolution by up to 50% compared to the rule-based algorithm. The resulting model is portable across Nvidia, AMD and Habana hardware. Accurate and fast machine-learning based reconstruction can significantly improve future measurements at colliders.

Physics↗

DOE OSTI2021

CAN-D: A Modular Four-Step Pipeline for Comprehensively Decoding Controller Area Network Data

Controller area networks (CANs) are a broadcast protocol for real-time communication of critical vehicle subsystems. Original equipment manufacturers of passenger vehicles hold secret their mappings of CAN data to vehicle signals, and these definitions vary according to make, model, and year. Without these mappings, the wealth of real-time vehicle information hidden in the CAN packets is uninterpretable, severely impeding vehicle-related research, including CAN cybersecurity and privacy studies, aftermarket tuning, efficiency and performance monitoring, and fault diagnosis to name a few. Guided by the four-part CAN signal definition, we present CAN-D (CAN-Decoder), a modular, four-step pipeline for identifying each signal's boundaries (start bit and length), endianness (byte ordering), signedness (bit-to-integer encoding), and by leveraging diagnostic standards, augmenting a subset of the extracted signals with meaningful, physical interpretation. En route to CAN-D, we provide a comprehensive review of the CAN signal reverse engineering research. All previous methods ignore endianness and signedness, rendering them incapable of decoding many standard CAN signal definitions. Incorporating endianness grows the search space from 128 to 4.72E21 signal tokenizations and introduces a web of changing dependencies. In response, we formulate, formally analyze, and provide an efficient solution to an optimization problem, allowing identification of the optimal set of signal boundaries and byte orderings. In addition, we provide two novel, state-of-the-art signal boundary classifiers—both of which are superior to previous approaches in precision and recall in three different test scenarios—and the first signedness classification algorithm, which exhibits a $>$ 97% F-score. Altogether, CAN-D is the only solution with the potential to extract any CAN signal that is also the state of the art. In evaluation on 10 vehicles of different makes, CAN-D's average $\ell ^1$ error is five times better (81% less) than all previous methods and exhibits lower average error, even when considering only signals that meet prior methods’ assumptions. Finally, CAN-D is implemented in lightweight hardware, allowing for an on-board diagnostic (OBD-II) plugin for real-time in-vehicle CAN decoding.

42 ENGINEERING↗

DOE OSTI2022

Upgrade of Gamma Spectrometry Systems for ORNL TRISO Fuel PIE

Gamma spectrometry is a key element in much of the post-irradiation examination (PIE) work performed under the Advanced Gas Reactor Fuel Development and Qualification (AGR) Program (Demkowicz et al. 2015; Stempien et al. 2021). Gamma spectrometers are integrated into three major capabilities used at the Oak Ridge National Laboratory (ORNL) Irradiated Fuels Examination Laboratory (IFEL) for PIE of tristructural-isotropic (TRISO) coated particles and fuel compacts: the Core Conduction Cooldown Test Facility (CCCTF), the Vertical Counting System (VCS), and the Irradiated Microsphere Gamma Analyzer (IMGA). The CCCTF includes liquid-nitrogen-cooled traps to extract 85 Kr out of the He sweep gas that passes through the furnace in which the fuel compacts are heated during safety testing. Analysis of the 85 Kr activity in the traps is the primary indicator for TRISO failure during safety testing. The VCS is a system used to accurately measure gamma emission from components placed in a lead-shielded chamber. It is used to count the CCCTF deposition cups after removal from furnace. Each cup resides in the CCCTF furnace for typically 12–24 h and is periodically replaced with a fresh cup throughout the safety test. Metallic fission products collect on the water-cooled cups and several gamma-emitting isotopes ( 110 mAg, 134 Cs, 137 Cs, 154 Eu, and 155 Eu) are often measured and provide indication of the retention performance of the TRISO coatings. The VCS is also used to measure the presence of these isotopes on the CCCTF tantalum liner and sweep gas inlet tube for the determination of cup collection efficiency, as well as support other gamma spectrometry needs related to calibration of the 85 Kr fission gas traps and various other special PIE tasks. The IMGA uses gamma spectrometry to measure the inventory of gamma-emitting isotopes in individual TRISO particles. An automated particle handling system within the IMGA hot cell removes each particle from a source vial and positions it in front of a gamma detector, and output from the gamma spectrometer is used by the IMGA software to determine a destination vial such that particles are sorted according to their inventory and retention characteristics. At the conclusion of the AGR-1 and AGR-2 PIE campaigns, the gamma spectrometer systems used at ORNL to support that PIE had reached the end of its life cycle due to gradual obsolescence of the hardware and software. Upgrade of the Canberra Genie 2000 software used by these systems to a Windows 10 version was not a viable option, because the newest Windows 10 version offered by Mirion (the new owner of the Canberra technology) did not include the dynamic-link libraries (DLLs) needed for integration with the custom PIE software used with the CCCTF and IMGA, and Mirion had no current plans for development and release of Windows 10 versions of these DLLs with the Model S560 Genie 2000 Programming Library. Ultimately a switch was made to ORTEC gamma spectrometry systems, which appeared to be a more sustainable solution due to more proactive vendor support. The ORTEC conversion involved replacing the aging detector preamplifier and multichannel analyzer (MCA) hardware, upgrading the obsolete Windows 7 computers to Windows 10 compatible models, adopting ORTEC GammaVision software, and extensive modification of the ORNL-developed Visual Basic .NET (VB.NET) programs that provide the CCCTF and IMGA user interfaces.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

DOE OSTI2025

Development and validation of a software for simulating γ-γ coincidence emission and detection probabilities

Gamma-gamma coincidence spectrometers have the potential to significantly enhance detection sensitivity for ultra-trace radionuclide measurements. The implementation of these spectrometers, however, is limited by the complexity of acquisition hardware, data processing and quantification. This work reports development of a novel radionuclide quantification software for γ-γ coincidence measurements. For any radionuclide, the software parses the Evaluated Nuclear Structure Data File (ENSDF) database, recursively simulating all possible γ-γ coincidence signatures and their respective emission and detection probabilities. Implemented using Python programming language, the software employs several strategies to boost overall computational performance. Since coincidence-based spectrometers are of notable interest in monitoring compliance for the Comprehensive Nuclear-Test-Ban Treaty (CTBT), the software’s execution was tested for 84 CTBT-relevant radionuclides. To date, the software has been experimentally validated for 15 radionuclides using the Advanced Radionuclide Gamma spectrOmeter (ARGO) at Pacific Northwest National Laboratory, USA (PNNL). Notably, the software can be operated in convergence mode, whereby coincidence detection efficiency’s convergence behavior can help avoid unreliable radionuclide activity estimates. With growing number of coincidence spectrometers worldwide, this paper aims to assist the radiation metrology community in developing similar software for their system.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

DOE OSTI2021

Primary Frequency Control Using Motor Drives for Short Term Grid Disturbances

Industrial motor systems make up a quarter of all electric sales in the United States. Variable speed drives (VSDs) can provide energy efficiency savings to the customer by regulating motor speed based on specific and varying needs. In addition to the benefits provided to the customer, VSDs can provide support to the grid through ancillary services. The Center for Ultra-Wide-Area Resilient Electric Energy Transmission Networks (CURENT) developed a power electronics converter-based grid emulator to allow testing of various power system architectures and demonstration of key technologies in monitoring, control, actuation, and visualization. This paper proposes using an active front-end VSD's connected motor load to provide frequency regulation to a large scale power grid. Each part of the emulator is described including motor and power electronics model and control. The proposed frequency regulation is implemented in VSDs and modeled in both a transmission system in EMTDC/PSCAD and verified on CURENT's hardware testbed.

24 POWER TRANSMISSION AND DISTRIBUTION↗

DOE OSTI2024

Foldy: An open-source web application for interactive protein structure analysis

Foldy is a cloud-based application that allows non-computational biologists to easily utilize advanced AI-based structural biology tools, including AlphaFold and DiffDock. With many deployment options, it can be employed by individuals, labs, universities, and companies in the cloud without requiring hardware resources, but it can also be configured to utilize locally available computers. Foldy enables scientists to predict the structure of proteins and complexes up to 6000 amino acids with AlphaFold, visualize Pfam annotations, and dock ligands with AutoDock Vina and DiffDock. In our manuscript, we detail Foldy’s interface design, deployment strategies, and optimization for various user scenarios. We demonstrate its application through case studies including rational enzyme design and analyzing proteins with domains of unknown function. Furthermore, we compare Foldy’s interface and management capabilities with other open and closed source tools in the field, illustrating its practicality in managing complex data and computation tasks. Our manuscript underlines the benefits of Foldy as a day-to-day tool for life science researchers, and shows how Foldy can make modern tools more accessible and efficient.

59 BASIC BIOLOGICAL SCIENCES↗

DOE OSTI2024

Concept Study of Robotic Camera-Based Foreign Object Detection for EV Wireless Charging

Wireless charging of an electric vehicle (EV) is an emerging charging technology promising convenient, autonomous, and highly efficient EV charging without requiring heavy gauge cables. However, due to the strong electromagnetic field created by this process that surrounds the wireless charger, the presence of foreign objects can detrimentally interact with it, thus affecting wireless power transfer (WPT) performance or leading to harmful and unwanted safety risks. This paper presents the results for a concept study on a robotic camera-based foreign object detection (FOD) system, as a supplement to the industry-existing overlapped FOD coil array method, for EV wireless charging. A Raspberry PI 4 control board and compatible Raspberry PI Camera Module 2 are used to implement camera-based object detection. The FOD program was developed using a state-of-the-art deep learning object detection model with the OpenCV and Pytorch library and is compatible with camera module hardware. A dry-run test with Raspberry PI and a camera module was conducted and the preliminary FOD function was verified. The feasibility assessment is also validated by comparing the performance of five existing state-of-the-art deep learning object detection models for vehicles, animals, persons, and metals subsets, respectively. Satisfactory performance on the benchmark datasets is observed by the tests, but further improvements are needed in future work when detecting small-sized metallic objects. A programable robotic car is also under development as ongoing work for carrying the Raspberry PI and camera module while moving for the maintenance process.

24 POWER TRANSMISSION AND DISTRIBUTION↗

DOE OSTI2020

Report on local data recovery approaches suitable for weather and climate prediction (Deliverable 1.3) (V.1.0)

Numerical weather and climate prediction rates as one of the scientific applications whose accuracy improvements greatly depend on the growth of the available computing power. As the number of cores in top computing facilities pushes into the millions, increasing average frequency of hardware and software failures forces users to review their algorithms and systems in order to protect simulations from breakdown. This report surveys approaches for fault-tolerance in numerical algorithms and system resilience in parallel simulations from the perspective of numerical weather and climate prediction systems. A selection of existing strategies is analyzed, featuring interpolation-restart and compressed checkpointing for the numerics, in-memory checkpointing, user-level failure mitigation-based and backup-based methods for the systems. Numerical examples showcase the performance of the techniques in addressing faults, with particular emphasis on iterative solvers for linear systems, a staple of atmospheric fluid flow solvers. The potential impact of these strategies is discussed in relation to current development of numerical weather prediction algorithms and systems towards the exascale. Trade-offs between performance, efficiency and effectiveness of resiliency strategies are analyzed and some recommendations outlined for future developments.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2024

Analyzing inference workloads for spatiotemporal modeling

Ensuring power grid resiliency, forecasting climate conditions, and optimization of transportation infrastructure are some of the many application areas where data is collected in both space and time. Spatiotemporal modeling is about modeling those patterns for forecasting future trends and carrying out critical decision-making by leveraging machine learning/deep learning. Once trained offline, field deployment of trained models for near real-time inference could be challenging because performance can vary significantly depending on the environment, available compute resources and tolerance to ambiguity in results. Users deploying spatiotemporal models for solving complex problems can benefit from analytical studies considering a plethora of system adaptations to understand the associated performance-quality trade-offs. To facilitate the co-design of next-generation hardware architectures for field deployment of trained models, it is critical to characterize the workloads of these deep learning (DL) applications during inference and assess their computational patterns at different levels of the execution stack. In this paper, we develop several variants of deep learning applications that use spatiotemporal data from dynamical systems. We study the associated computational patterns for inference workloads at different levels, considering relevant models (Long short-term Memory, Convolutional Neural Network and Spatio-Temporal Graph Convolution Network), DL frameworks (Tensorflow and PyTorch), precision (FP16, FP32, AMP, INT16 and INT8), inference runtime (ONNX and AI Template), post-training quantization (TensorRT) and platforms (Nvidia DGX A100 and Sambanova SN10 RDU). Overall, our findings indicate that although there is potential in mixed-precision models and post-training quantization for spatiotemporal modeling, extracting efficiency from contemporary GPU systems might be challenging. Instead, co-designing custom accelerators by leveraging optimized High Level Synthesis frameworks (such as SODA High-Level Synthesizer for customized FPGA/ASIC targets) can make workload-specific adjustments to enhance the efficiency.

97 MATHEMATICS AND COMPUTING↗