Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “inference accelerators”

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 55 records · Page 3
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

A GPU-Accelerated Population Generation, Sorting, and Mutation Kernel for an Optimization-Based Causal Inference Model

We develop a GPU-accelerated machine learning generative adversarial network model that can be used with observational data for the purpose of constructing causal inferences. The theoretical basis of our machine learning model is novel and is conceptualized to be operable and scalable for high performance computing platforms. Our GPU-accelerated code enables large-scale parallelization of the computation within a common and accessible computing environment. This will expand the reach of our model and empower research in new substantive domains while maintaining the underlying theoretical properties.

Cho, Wendy K. Tam↗
DOE OSTI2025

LLM-Inference-Bench: Inference Benchmarking of Large Language Models on AI Accelerators

Large Language Models (LLMs) have propelled groundbreaking advancements across several domains and are commonly used for text generation applications. However, the computational demands of these complex models pose significant challenges, requiring efficient hardware acceleration. Benchmarking the performance of LLMs across diverse hardware platforms is crucial to understanding their scalability and throughput characteristics. We introduce LLM-Inference-Bench, a comprehensive benchmarking suite to evaluate the hardware inference performance of LLMs. We thoroughly analyze diverse hardware platforms, including GPUs from Nvidia and AMD and specialized AI accelerators, Intel Habana and SambaNova. Our evaluation includes several LLM inference frameworks and models from LLaMA, Mistral, and Qwen families with 7B and 70B parameters. Our benchmarking results reveal the strengths and limitations of various models, hardware platforms, and inference frameworks. We provide an interactive dashboard to help identify configurations for optimal performance for a given hardware platform.

Chitty-Venkata, Krishna Teja↗
DOE OSTI2021

I-GCN: A Graph Convolutional Network Accelerator with Runtime Locality Enhancement through Islandization

In this paper, we propose a novel hardware accelerator for GCN inference called I-GCN that significantly improves data locality and reduces unnecessary computation through a new online graph restructuring algorithm we refer to as islandization. The proposed algorithm finds clusters of nodes with strong internal but weak external connections. The islandization process yields two major benefits. First, by processing islands rather than individual nodes, there is better on-chip data reuse and fewer off-chip memory accesses. Second, there is less redundant computation as aggregation for common/shared neighbors in an island can be reused. The parallel search, identification, and leverage of graph islands are all handled purely in hardware at runtime working in an incremental pipelined manner. This is done without any preprocessing of the graph data or adjustment of the GCN model structure.

Geng, Tong↗
DOE OSTI2025

hls4ml: A Flexible, Open-Source Platform for Deep Learning Acceleration on Reconfigurable Hardware

We present hls4ml, a free and open-source platform that translates machine learning (ML) models from modern deep learning frameworks into high-level synthesis (HLS) code that can be integrated into full designs for field-programmable gate arrays (FPGAs) or application-specific integrated circuits (ASICs). With its flexible and modular design, hls4ml supports a large number of deep learning frameworks and can target HLS compilers from several vendors, including Vitis HLS, Intel oneAPI and Catapult HLS. Together with a wider eco-system for software-hardware co-design, hls4ml has enabled the acceleration of ML inference in a wide range of commercial and scientific applications where low latency, resource usage, and power consumption are critical. In this paper, we describe the structure and functionality of the hls4ml platform. The overarching design considerations for the generated HLS code are discussed, together with selected performance results.

FOS: Computer and information sciences↗
DOE OSTI2025

hls4ml: A Flexible, Open-Source Platform for Deep Learning Acceleration on Reconfigurable Hardware

We present hls4ml, a free and open-source platform that translates machine learning (ML) models from modern deep learning frameworks into high-level synthesis (HLS) code that can be integrated into full designs for field-programmable gate arrays (FPGAs) or application-specific integrated circuits (ASICs). With its flexible and modular design, hls4ml supports a large number of deep learning frameworks and can target HLS compilers from several vendors, including Vitis HLS, Intel oneAPI and Catapult HLS. Together with a wider eco-system for software-hardware co-design, hls4ml has enabled the acceleration of ML inference in a wide range of commercial and scientific applications where low latency, resource usage, and power consumption are critical. In this paper, we describe the structure and functionality of the hls4ml platform. The overarching design considerations for the generated HLS code are discussed, together with selected performance results.

Schulte, Jan-Frederik [Purdue U.] (ORCID:000000034↗
DOE OSTI2021

GPU coprocessors as a service for deep learning inference in high energy physics

In the next decade, the demands for computing in large scientific experiments are expected to grow tremendously. During the same time period, CPU performance increases will be limited. At the CERN Large Hadron Collider (LHC), these two issues will confront one another as the collider is upgraded for high luminosity running. Alternative processors such as graphics processing units (GPUs) can resolve this confrontation provided that algorithms can be sufficiently accelerated. In many cases, algorithmic speedups are found to be largest through the adoption of deep learning algorithms. We present a comprehensive exploration of the use of GPU-based hardware acceleration for deep learning inference within the data reconstruction workflow of high energy physics. We present several realistic examples and discuss a strategy for the seamless integration of coprocessors so that the LHC can maintain, if not exceed, its current performance throughout its running.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗
DOE OSTI2024

Simulating the CMS High Granularity Calorimeter with ML

Detector simulation is a key component of physics analysis and related activities in CMS. In the upcoming High Luminosity LHC era, simulation will be required to use a smaller fraction of computing in order to satisfy resource constraints. At the same time, CMS will be upgraded with the new High Granularity Calorimeter (HGCal), which requires significantly more resources to simulate than the existing CMS calorimeters. This computing challenge motivates the use of generative machine learning models as surrogates to replace full physics-based simulation. We study the application of state-of-the-art diffusion models to simulate particle showers in the CMS HGCal. We will discuss methods to overcome the challenges posed by the high-dimensional, irregular geometry of the HGCal. The quality of the showers produced by the diffusion model will be assessed by comparison to the full GEANT4-based simulation. The increase in simulation throughput will be quantified and methods to accelerate the diffusion model inference will also be discussed.

Amram, Oz↗
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 OSTI2023

Towards On-Chip Learning for Low Latency Reasoning with End-to-End Synthesis

The Software Defined Architectures (SODA) Synthesizer is an open-source compiler-based tool able to automatically generate domain-specialized systems targeting Application-Specific Integrated Circuits (ASICs) or Field Programmable Gate Arrays (FPGAs) starting from high-level programming. SODA is composed of a frontend, SODA-OPT, which leverages the multilevel intermediate representation (MLIR) framework to interface with productive programming tools (e.g., machine learning frame-works), identify kernels suitable for acceleration, and perform high-level optimizations, and of a state-of-the-art high-level synthesis backend, Bambu from the PandA framework, to generate custom accelerators. One specific application of the SODA Synthesizer is the generation of accelerators to enable ultra-low latency inference and control on autonomous systems for scientific discovery (e.g., electron microscopes, sensors in particle accelerators, etc.). This paper provides an overview of the flow in the context of the generation of accelerators for edge processing to be integrated in transmission electron microscopy (TEM) devices, focusing on use cases from precision material synthesis. We show the tool in action with an example of design space exploration for inference on reconfigurable devices with a conventional deep neural network model (LeNet). Finally, we discuss the research directions and opportunities enabled by SODA in the area of autonomous control for scientific experimental workflows.

Castellana, Vito G.↗
DOE OSTI2022

Detecting Thermally-Induced Spinodal Decomposition with Picosecond Ultrasonics in Cast Austenitic Stainless Steels

Destructive techniques to monitor nuclear reactor component health may not always be available during service, as they are time-consuming and often require pre-installed inspection coupons. Non-destructive evaluation (NDE) techniques can bridge this gap by rapidly identifying the state of mission-critical reactor components, via inference between NDE-measurable material properties and those of ultimate interest, such as ductility and toughness. Here, we demonstrate one such inference about the health of thermally aged cast austenitic stainless steels. Observations of surface acoustic wave peak (SAW) splitting correlate with spinodal decomposition-induced embrittlement as destructively measured by Charpy impact energy. Elastodynamic calculations and molecular dynamics simulations of the effects of spinodal decomposition on elastic moduli support that the new acoustic modes present are due to stiffening in the δ-ferrite domains. Finally, this discovery enables one to probe structure-property relationships in materials in a greatly accelerated manner, suggesting that similar inference methods can be used to determine material fitness-for-service, or to quickly uncover new structure-property relationships.

304-type stainless steel↗
DOE OSTI2024

Towards Automated Generation of Chiplet-Based Systems

The Software Defined Architectures (SODA) Synthesizer is an open-source compiler-based tool able to automatically generate domain-specialized systems targeting Application- Specific Integrated Circuits (ASICs) or Field Programmable Gate Arrays (FPGAs) starting from high-level programming. SODA is composed of a high-level frontend, SODA-OPT, which leverages the multilevel intermediate representation (MLIR) framework to interface with productive programming tools (e.g., machine learning frameworks), identify kernels suitable for acceleration, and perform high-level optimizations, and of a state-of-the-art high-level synthesis backend, Bambu from the PandA framework, to generate custom accelerators. One specific application of the SODA Synthesizer is the generation of accelerators to enable ultra-low latency inference and control on autonomous systems for scientific discovery (e.g., electron microscopes, sensors in particle accelerators, etc.). This talk will discuss ongoing work on the SODA synthesizer to enable no-human-in-the-loop generation and design space exploration of the chiplets for highly specialized artificial intelligence accelerators. Connecting these highly specialized chiplets to general-purpose cores or programmable accelerators will allow to quickly deploy autonomous systems for scientific discovery.

Limaye, Ankur M.↗
DOE OSTI2023

FPGA Acceleration of GCN in Light of the Symmetry of Graph Adjacency Matrix

Graph Convolutional Neural Networks (GCNs) are widely used to process large-scale graph data. Different from deep neural networks (DNNs), GCNs are sparse, irregular, and unstructured, posing unique challenges to hardware acceleration with regular processing elements (PEs). In particular, the adjacency matrix of a GCN is extremely sparse, leading to frequent but irregular memory access, low spatial/temporal data locality and poor data reuse. Furthermore, a realistic graph usually consists of unstructured data (e.g., unbalanced distributions), creating significantly different processing times and imbalanced workload for each node in GCN acceleration. To overcome these challenges, we propose an end-to-end hardware-software co-design to accelerate GCNs on resource-constrained FPGAs with the features including: (1) A custom dataflow that leverages symmetry along the diagonal of the adjacency matrix to accelerate feature aggregation for undirected graphs. We utilize either the upper or the lower triangular matrix of the adjacency matrix to perform aggregation in GCN to improve data reuse. (2) Unified compute cores for both aggregation and transform phases, with full support to the symmetry-based dataflow. These cores can be dynamically reconfigured to the systolic mode for transformation or as individual accumulators for aggregation in GCN processing. (3) Preprocessing of the graph in software to rearrange the edges and features to match the custom dataflow. This step improves the regularity in memory access and data reuse in the aggregation phase. Moreover, we quantize the GCN precision from FP32 to INT8 to reduce the memory footprint without losing the inference accuracy. We implement our accelerator design in Intel Stratix10 MX FPGA board with HBM2, and demonstrate 1.3x-110.5x improvement in end-to-end GCN latency as compared to the state-of the-art FPGA implementations, on the graph datasets of Cora, Pubmed, Citeseer and Reddit.

Nair, Gopikrishnan R.↗
DOE OSTI2024

A Magnetized Strongly Turbulent Corona as the Source of Neutrinos from NGC 1068

Abstract The cores of active galactic nuclei are potential accelerators of 10–100 TeV cosmic rays, in turn producing high-energy neutrinos. This picture was confirmed by the compelling evidence of a TeV neutrino signal from the nearby active galaxy NGC 1068, leaving open the question of what is the site and mechanism of cosmic-ray acceleration. One candidate is the magnetized turbulence surrounding the central supermassive black hole. Recent particle-in-cell simulations of magnetized turbulence indicate that stochastic cosmic-ray acceleration is nonresonant, in contrast to the assumptions of previous studies. We show that this has important consequences on a self-consistent theory of neutrino production in the corona, leading to a more rapid cosmic-ray acceleration than previously considered. The turbulent magnetic-field fluctuations needed to explain the neutrino signal are consistent with a magnetically powered corona. We find that strong turbulence, with turbulent magnetic energy density higher than 1% of the rest-mass energy density, naturally explains the normalization of the IceCube neutrino flux, in addition to the neutrino spectral shape. Only a fraction of the protons in the corona, which can be directly inferred from the neutrino signal, are accelerated to high energies. Thus, in this framework, the neutrino signal from NGC 1068 provides a testbed for particle acceleration in magnetized turbulence.

Fiorillo, Damiano F. G. (ORCID:0000000349279850)↗
DOE OSTI2023

End-to-end AI framework for interpretable prediction of molecular and crystal properties

We introduce an end-to-end computational framework that allows for hyperparameter optimization using the DeepHyper library, accelerated model training, and interpretable AI inference. The framework is based on state-of-the-art AI models including CGCNN, PhysNet, SchNet, MPNN, MPNN-transformer, and TorchMD-NET. We employ these AI models along with the benchmark QM9, hMOF, and MD17 datasets to showcase how the models can predict user-specified material properties within modern computing environments. We demonstrate transferable applications in the modeling of small molecules, inorganic crystals and nanoporous metal organic frameworks with a unified, standalone framework. We have deployed and tested this framework in the ThetaGPU supercomputer at the Argonne Leadership Computing Facility, and in the Delta supercomputer at the National Center for Supercomputing Applications to provide researchers with modern tools to conduct accelerated AI-driven discovery in leadership-class computing environments. We release these digital assets as open source scientific software in GitLab, and ready-to-use Jupyter notebooks in Google Colab.

36 MATERIALS SCIENCE↗
DOE OSTI2024

A differentiable simulation package for performing inference of synchrotron-radiation-based diagnostics

The direction of particle accelerator development is ever-increasing beam quality, currents and repetition rates. This poses a challenge to traditional diagnostics that directly intercept the beam due to the mutual destruction of both the beam and the diagnostic. An alternative approach is to infer beam parameters non-invasively from the synchrotron radiation emitted in bending magnets. However, inferring the beam distribution from a measured radiation pattern is a complex and computationally expensive task. To address this challenge we present SYRIPY ( SYnchrotron Radiation In PYthon ), a software package intended as a tool for performing inference of synchrotron-radiation-based diagnostics. SYRIPY has been developed using PyTorch , which makes it both differentiable and able to leverage the high performance of GPUs, two vital characteristics for performing statistical inference. The package consists of three modules: a particle tracker, Lienard–Wiechert solver and Fourier optics propagator, allowing start-to-end simulation of synchrotron radiation detection to be carried out. SYRIPY has been benchmarked against SRW , the prevalent numerical package in the field, showing good agreement and up to a 50× speed improvement. Finally, we have demonstrated how SYRIPY can be used to perform Bayesian inference of beam parameters using stochastic variational inference.

43 PARTICLE ACCELERATORS↗
DOE OSTI2025

Neural network-based model of galaxy power spectrum: fast full-shape galaxy power spectrum analysis

ABSTRACT We present a neural network-based emulator for the galaxy redshift-space power spectrum that enables several orders of magnitude acceleration in the galaxy clustering parameter inference, while preserving 3$\sigma$ accuracy better than 0.5 per cent up to $k_{\mathrm{max}}$ = 0.25 $\, h\text{Mpc}^{-1}$ within Lambda-cold dark matter ($\Lambda$CDM) and around 0.5 per cent $w_0$–$w_a$CDM. Our surrogate model only emulates the galaxy bias-invariant terms of one-loop perturbation theory predictions, these terms are then combined analytically with galaxy bias terms, counter-terms, and stochastic terms in order to obtain the non-linear redshift-space galaxy power spectrum. This allows us to avoid any galaxy bias prescription in the training of the emulator, which makes it more flexible. Moreover, we include the redshift $z \in [0,1.4]$ in the training which further avoids the need for re-training the emulator. We showcase the performance of the emulator in recovering the cosmological parameters of $\Lambda$CDM by analysing the suite of 25 AbacusSummit simulations that mimic the Dark Energy Spectroscopic Instrument luminous red galaxies at $z=0.5$ and 0.8, together as the emission line galaxies at $z=0.8$. We obtain similar performance in all cases, demonstrating the reliability of the emulator for any galaxy sample at any redshift in $0 \lt z \lt 1.4$. We will make our emulator public at github repository.

Trusov, Svyatoslav (ORCID:0000000224146720)↗
DOE OSTI2022

Proteome-scale Deployment of Protein Structure Prediction Workflows on the Summit Supercomputer

Deep learning has contributed to major advances in the prediction of protein structure from sequence, a fundamental problem in structural bioinformatics. With predictions now approaching the accuracy of crystallographic experiments, and with accelerators like GPUs and TPUs making inference using large models rapid, genome-level structure prediction becomes an obvious aim. Leadership-class computing resources can be used to perform genome-scale protein structure prediction using state-of-the-art deep learning models, providing a wealth of new data for systems biology applications. Here we describe our efforts to efficiently deploy the AlphaFold v.2 program, for full-proteome structure prediction, at scale on the Oak Ridge Leadership Computing Facility's resources, including the Summit supercomputer. We performed inference to produce the predicted structures for 40,526 protein sequences, corresponding to four prokaryotic proteomes and one plant proteome, using under 4,400 total Summit node hours, equivalent to using the majority of the supercomputer for a little over one hour. We also designed an optimized structure refinement that reduced the time for the relaxation stage of the AlphaFold pipeline by over 10X for longer sequences. We demonstrate the types of analyses that can be performed on proteome-scale collections of sequences, including a search for novel quaternary structures and implications for functional annotation.

Gao, Mu↗