Search NASA⌕ Search

SEARCH · Search NASA

Results for “AI workloads”

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 OSTI2026

Future Generation High Performance Computing Center (FG-HPCC): RFI Technical Considerations

Lawrence Livermore National Security, LLC (LLNS) is interested in receiving information about technologies that could be available in the 2029-2030 timeframe that may serve to enable the vision for a Future Generation High Performance Computing (HPC) Center (FG-HPCC) described in this document. The future HPC Center vision has been conceived to meet the future mission needs of the Advanced Simulation and Computing (ASC) Program within the National Nuclear Security Administration (NNSA). LLNS envisions a center composed not of many independent clusters, but of heterogeneous elements accessible to users as a single system. The capabilities will be integrated to create a scalable, flexible, yet tightly coupled computing center capable of integrated HPC, AI, and cloud-like workloads.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2022

AI4IO: A suite of AI-based tools for IO-aware scheduling

Traditional workload managers do not have the capacity to consider how IO contention can increase job runtime and even cause entire resource allocations to be wasted. Whether from bursts of IO demand or parallel file systems (PFS) performance degradation, IO contention must be identified and addressed to ensure maximum performance. In this paper, we present AI4IO (AI for IO), a suite of tools using AI methods to prevent and mitigate performance losses due to IO contention. AI4IO enables existing workload managers to become IO-aware. Currently, AI4IO consists of two tools: PRIONN and CanarIO. PRIONN predicts IO contention and empowers schedulers to prevent it. CanarIO mitigates the impact of IO contention when it does occur. We measure the effectiveness of AI4IO when integrated into Flux, a next-generation scheduler, for both small- and large-scale IO-intensive job workloads. Our results show that integrating AI4IO into Flux improves the workload makespan up to 6.4%, which can account for more than 18,000 node-h of saved resources per week on a production cluster in our large-scale workload.

Wyatt, II, Michael R.↗
DOE OSTI2023

2022 Operational Assessment Report - Argonne Leadership Computing Facility

This Operational Assessment Report describes how the Argonne Leadership Computing Facility (ALCF) met or exceeded every one of its goals for the calendar year (CY) 2022. In CY 2022, ALCF operated Theta, an Intel-based Cray XC40 system (11.7 petaflops) augmented with 24 NVIDIA DGX A100-based nodes (3.9 petaflops), that supports diverse workloads, integrating data analytics with artificial intelligence (AI) training and learning in a single platform; and Polaris, a 44-petaflop AMD and NVIDIA-based HPE Apollo 6500 Gen10+ system that provides a powerful new platform to prepare applications and workloads for Aurora, Argonne National Laboratory’s (Argonne’s) upcoming Intel-Hewlett Packard Enterprise (HPE) exascale supercomputer.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2024

Studying CPU and memory utilization of applications on Fujitsu A64FX and Nvidia Grace Superchip

ARM-based manycore CPU architectures are well-positioned to provide the rising memory throughput requirements of modern data intensive scientific applications in High Performance Computing (HPC). The Fujitsu A64FX CPU platform is based on the ARM v8.2A architecture, and is the processor of the flagship Japanese supercomputer - "Fugaku", which was previously ranked as the #1 supercomputer in the world according to the Top500 list. The Nvidia Grace superchip features 144 Neoverse V2 cores based on the ARMv9 architecture with 4x128b SVE2, providing exceptional computational power. The chip supports up to 480GB of memory, making it ideal for AI, machine learning, and scientific computing workloads. In this paper, we conduct a thorough performance exploration of a variety of parallel bandwidth-sensitive benchmarks and applications compiled with the native Fujitsu compiler on a Fugaku A64FX compute node and ARM (LLVM) Compiler on an NVIDIA Grace superchip compute node, engaging all the computational cores per cluster using OpenMP multithreading (assuming the cores can drive the available bandwidth). Our ultimate goals are to study the resource utilization of scientific applications and benchmarks on A64FX and Grace superchip, considering graph application scenarios ( GAP Benchmark suite) and eleven appli- cation proxies from the Rodinia heterogeneous benchmark suite (considering domains such as Data Mining, Bioinformatics, Fluid Dynamics, Pattern Recognition, etc.). Through exhaustive performance monitoring, we quantify the resource utilization of diverse OpenMP-based HPC applications on both the Fujitsu A64FX and the Nvidia Grace Superchip platforms.

benchmarking, Performance Analysis, High performan↗
DOE OSTI2025

I/O in Machine Learning Applications on HPC Systems: A 360-degree Survey

Growing interest in Artificial Intelligence (AI) has resulted in a surge in demand for faster methods of Machine Learning (ML) model training and inference. This demand for speed has prompted the use of high performance computing (HPC) systems that excel in managing distributed workloads. Because data is the main fuel for AI applications, the performance of the storage and I/O subsystem of HPC systems is critical. In the past, HPC applications accessed large portions of data written by simulations or experiments or ingested data for visualizations or analysis tasks. ML workloads perform small reads spread across a large number of random files. This shift of I/O access patterns poses several challenges to modern parallel storage systems. In this paper, we survey I/O in ML applications on HPC systems, and target literature within a 6-year time window from 2019 to 2024. We define the scope of the survey, provide an overview of the common phases of ML, review available profilers and benchmarks, examine the I/O patterns encountered during offline data preparation, training, and inference, and explore I/O optimizations utilized in modern ML frameworks and proposed in recent literature. Lastly, we seek to expose research gaps that could spawn further R&D.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2025

In-Transit Data Transport Strategies for Coupled AI-Simulation Workflow Patterns

Coupled AI-Simulation workflows are becoming the major workloads for HPC facilities, and their increasing complexity necessitates new tools for performance analysis and prototyping of new in-situ workflows. We present SimAI-Bench, a tool designed to both prototype and evaluate these coupled workflows. In this paper, we use SimAI-Bench to benchmark the data transport performance of two common patterns on the Aurora supercomputer: a one-to-one workflow with co-located simulation and AI training instances, and a many-to-one workflow where a single AI model is trained from an ensemble of simulations. For the one-to-one pattern, our analysis shows that node-local and DragonHPC data staging strategies provide excellent performance compared Redis and Lustre file system. For the many-to-one pattern, we find that data transport becomes a dominant bottleneck as the ensemble size grows. Our evaluation reveals that file system is the optimal solution among the tested strategies for the many-to-one pattern.

Tummalapalli, Harikrishna [Argonne National Labora↗
NASA NTRSDate not supplied

Transforming Science Prioritization Processes Using Artificial Intelligence

Artificial Intelligence (AI) and Machine Learning (ML) have potential to augment significantly the current labor-intensive processes of science prioritization, specifically by the National Academies’ Decadal Survey on behalf of NASA and NSF. Here we summarize what we believe to be the first exploratory demonstration-of-concept results from an application of AI/ML to Survey science prioritization. Specifically, we applied Latent Dirichlet Allocation (LDA) and Natural Language Processing (NLP) to reveal trends in published astrophysics research that may indicate science priorities and which could be applied to strategic planning. For the purpose of the work that we summarize here, AI/ML is able to analyze – that is, to “understand,” in a manner of speaking – a vast amount of text to reveal complex relationships among research topics, including the growth or decline of science community activities in those topics over time. We trained ourselves and AI/ML algorithms by using ~400,000 abstracts in the period 1998 to 2010 to “forecast” the Academies’ Astro2010 recommendations and compare with the solicited white papers. Comparing our results with actual Astro2010 recommendations allowed us to identify candidate metrics that better predicted the actual results of the Survey. We found, for example, that Compound Annual Growth Rate (CAGR) of papers published in a topic area is a good proxy measure for importance of this topic area of research. With this training complete, we identified candidate astrophysics astrophysics science priorities for the 2021+ period using the research during 2007 - 2019 . We conclude that appropriate application of AI can potentially significantly reduce the current workload of the Decadal Survey processes and reveal otherwise unrecognized characteristics in the body of astronomical research. We emphasize throughout the exploratory nature of our work, encouraging colleagues to pursue promising results further. Our most critical governing assumption was that increased (or decreased) research activity can be used to identify scientific or technology topic areas worthy of increased (or decreased) future emphasis. We discuss advantages, limitations, and recognize the “black box” nature of our technique. We note ethics issues associated, for example, with using AI/ML to reveal “hidden” meanings and biases in published work. Furthermore, inevitable improvements in AI may soon enable widespread and welcome identification of and advocacy for science and technology priorities by disparate and diverse groups and organizations. Consequently, we continue to urge a near-term, in-depth evaluation of appropriate applications of AI, including implications and consequences, as well as support for multiple follow-on assessments, of which ours is only a beginning.

Artificial Intelligence↗
DOE OSTI2025

Empirically-calibrated H100 node power models for accurate AI training energy estimation

Accurately quantifying the energy use of artificial intelligence (AI) training is critical for infrastructure planning, carbon accounting, and sustainable data center operation, but few studies have directly measured the power consumption of production workloads on contemporary hardware. By combining empirical measurements from Brookhaven National Laboratory during AI training on 8-graphics-processing-unit H100 systems with open-source benchmarking data, we develop statistical models relating computational intensity to node-level power consumption. We measure the gap between manufacturer-rated thermal design power (TDP) and actual power demand during AI training. Our analysis reveals that even computationally intensive workloads operate at only 76% of the 10.2 kW TDP rating. Our architecture-specific model, calibrated to floating-point operations, predicts energy consumption with 11.4% mean absolute percentage error, significantly outperforming TDP-based approaches (27%–37% error). We identified distinct power signatures between transformer and convolutional neural network architectures, with transformers showing characteristic fluctuations that may impact grid stability. These results provide a measurement-grounded basis for improving AI training energy estimates, enabling more reliable infrastructure sizing, cost projections, and environmental impact assessments.

Newkirk, Alex C↗
DOE OSTI2021

Argonne Leadership Computing Facility: 2021 Operational Assessment Report

This Operational Assessment Report describes how the Argonne Leadership Computing Facility (ALCF) met or exceeded every one of its goals for calendar year (CY) 2021 as an advanced scientific computing center. In CY 2021, the ALCF operated its production resource, Theta, an Intel-based Cray XC40 system (11.7-petaflops) augmented with 24 NVIDIA DGX A100-based nodes (3.9-petaflops) that supports diverse workloads, integrating data analytics with artificial intelligence (AI) training and learning in a single platform. In 2021, we began deploying Polaris, our newest 40- petaflops system, and augmented this powerful testbed system with an additional 28 nodes to support the integration of real-time experiments and HPC resources. We also deployed our two largest storage systems yet, named Grand and Eagle, that will bring new services to our users and will power data-driven research for years to come. Last year, Theta delivered a total of 20.8 million node-hours to 16 Innovative and Novel Computational Impact on Theory and Experiment (INCITE) projects and 7.2 million node-hours to ASCR Leadership Computing Challenge (ALCC) projects (32 awarded during the 2020–2021 ALCC year and 17 awarded during the 2021–2022 ALCC year), as well as substantial support to Director’s Discretionary (DD) projects (5.5 million node-hours). As Table ES.1 shows, Theta performed exceptionally well in terms of overall availability (95.1 percent), scheduled availability (99.4 percent), and utilization (98.1 percent; Table 2.1). As of the submission date of this document, ALCF’s user community has published 249 papers in high-quality, peer-reviewed journals and technical proceedings. At the 2021 International Conference for High Performance Computing, Networking, Storage and Analysis (SC’21), Argonne researchers won two HPCwire Readers’ Choice Awards and were part of a Gordon Bell Prize finalist team recognized for developing an AI-enabled, multi-resolution simulation framework for studying complex biomolecular machines. Their framework was used to observe the SARS-CoV-2 replication-transcription machinery in action, by directly integrating experimental data. ALCF also provided a comprehensive program of high-performance computing (HPC) support services to help our community make productive use of the facility’s diverse and growing collection of resources. We are now entering the exascale era, with exascale machines being planned for national laboratories across the country, including Aurora at Argonne National Laboratory (Argonne) in 2023. ALCF researchers have been leading and guiding numerous strategic activities that will push the boundaries of what’s possible in computational science and engineering and allow us to deliver science on day one.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2026

iDDS: intelligent distributed dispatch and scheduling for workflow orchestration

The intelligent distributed dispatch and scheduling (iDDS) service is a versatile workflow orchestration system designed for large-scale, distributed scientific computing. iDDS extends traditional workload and data management by integrating data-aware execution, conditional logic, and programmable workflows, enabling automation of complex and dynamic processing pipelines. Originally developed for the ATLAS experiment at the large hadron collider, iDDS has evolved into an experiment-agnostic platform that supports both template-driven workflows and a Function-as-a-Task model for Python-based orchestration. This paper presents the architecture and core components of iDDS, highlighting its scalability, modular message-driven design, and integration with systems such as PanDA and Rucio. We demonstrate its versatility through real-world use cases: fine-grained tape resource optimization for ATLAS, orchestration of large Directed Acyclic Graph (DAG) workflows for the Rubin Observatory, distributed hyperparameter optimization for machine learning applications, active learning for physics analyses, and AI-assisted detector design at the electron–ion collider. By unifying workload scheduling, data movement, and adaptive decision-making, iDDS reduces operational overhead and enables reproducible, high-throughput workflows across heterogeneous infrastructures. We conclude with current challenges and future directions, including interactive, cloud-native, and serverless workflow support.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2026

A Methodology to Evaluate the Grid Reliability Impact of Oscillations Induced by Large Loads

The rapid growth of hyperscale AI data centers is bringing renewed attention to the reliability risk that sustained forced oscillations pose to bulk power systems, with cyclic computational workloads emerging as a new forcing source. Unlike the broadband, stochastic disturbances from traditional industrial loads such as arc furnaces, AI training and inference facilities can inject large active power swings concentrated at specific frequencies over extended durations - characteristics that existing grid planning practices do not account for. While the North American Electric Reliability Corporation (NERC) has recognized this gap and called for system-level studies of large load interconnections, no standardized methodology exists to screen, simulate, and quantify these risks at the planning stage. This report presents the Risk Assessment Tool for Large Load-induced Events (RATLLE), a Python-based, publicly available script suite developed at the Pacific Northwest National Laboratory to evaluate bulk power system reliability risks from data center-induced oscillations. RATLLE implements a three-module workflow: a screening module that identifies vulnerable interconnection locations and excitable system modes; a simulation module that models cyclic data center load behavior using a commercial positive sequence simulation platform; and an analysis module that computes risk metrics and generates interactive visualization dashboards. The risk metrics, formulated around simulation observables, map oscillation impacts to a three-stage severity scale spanning latent equipment fatigue through imminent cascading failure. The methodology is demonstrated on two Western Electricity Coordinating Council (WECC) system models: a publicly available 240-bus reduced representation and a detailed 2031 Heavy Winter planning case. Case studies illustrate that even modest 50 MW forced oscillations at resonant frequencies can produce wide-area power swings, N-1 security constraint violations, and cascading generator trips through protection actions - outcomes that would not occur under normal operating conditions without oscillations present. The results underscore the need for standardized oscillation impact assessment in large load interconnection studies and provide a reproducible, extensible framework for utilities to adopt or customize within their existing planning workflows.

Biswas, Shuchismita↗
DOE OSTI2026

ChatMPI: LLM-Driven MPI Code Generation for HPC Workloads

The Message Passing Interface (MPI) standard plays a crucial role in enabling scientific applications for parallel computing and is an essential component in high-performance computing (HPC). However, implementing MPI code manually—especially applying a proper domain decomposition and communication pattern—is a challenging and error-prone task. We present ChatMPI, an AI assistant for MPI parallelization of sequential C codes. In our analysis, we focus on testing six essential HPC workloads, which are based on Basic Linear Algebra Subprograms levels 1, 2, and 3 as well as sparse, stencil, and iterative operations. We analyze the process of creating ChatMPI by using the ChatHPC library. This lightweight large language model (LLM)–based infrastructure enables HPC experts to efficiently create and supervise trustworthy AI capabilities for critical HPC software tasks. We study the data required for training (fine-tuning) ChatMPI to generate parallel codes that not only use MPI syntax correctly but also apply HPC techniques to reduce memory communication and maximize performance by using proper work decomposition. With a relatively small training dataset composed of a few dozen prompts and fewer than 15 minutes of fine-tuning on one node equipped with two NVIDIA H100 GPUs, ChatMPI elevates trustworthiness for MPI code generation of current LLMs (e.g., Code Llama, ChatGPT-4o and ChatGPT 5). Additionally, we evaluate the performance of the MPI codes generated by ChatMPI in comparison with the ones generated by ChatGPT-4o and ChatGPT-5. The codes generated by ChatMPI provide up to a 4 × boost in performance by using better problem decomposition, communication patterns, and HPC techniques (e.g., communication avoiding).

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

Toward a Seamless Integration of Computing, Experimental, and Observational Science Facilities: A Blueprint to Accelerate Discovery

The Department of Energy, Office of Science operates world-leading facilities for experimental, observational, and computational science. DOE supercomputing facilities will reach performance at the scale of ExaFLOPs in the coming years, enabling new vistas of scale and precision for large scale simulations and data analysis. Experimental scientific facilities are undergoing similar upgrades that will lead to higher data rates and correspondingly larger computational demands, and will increase the need for near-real-time processing and resilient support for more complex workflows. A transformation of science is underway, with workloads at supercomputing facilities increasingly driven by this explosion of data from instruments and experimental facilities, as well as the accelerating use of Artificial Intelligence (AI) as a tool for scientific discovery. A seamless integration of computing, networking, instruments, and experimental facilities is required to support these emerging workloads and open up a new frontier of U.S. leadership in scientific discovery. We propose to accomplish this by providing frictionless access to the ASCR supercomputing facilities. We describe our vision of combining the power of ASCR supercomputers and networking infrastructure into an integrated scalable fabric, available to end user scientists via interfaces that aim to automate and simplify access to high performance computing systems. This will enable unprecedented computational science capabilities for experimental and observational facilities, and will create new opportunities to combine large simulations and modeling with experimental facility data analysis. This blueprint for creating an integrated network of computational and experimental facilities will provide an enriched discovery environment and open doors for new scientific communities to access the DOE’s world-leading computing and networking capabilities.

97 MATHEMATICS AND COMPUTING↗
NASA NTRSDate not supplied

Using Artificial Intelligence and Machine Learning to Enhance Mission Design and Operations of the Habitable Worlds Observatory (HWO)

One key aspect in the development of HWO is the early deployment of artificial intelligence (AI) and machine learning (ML) to enhance mission science and operations. Our subtask group is part of the HWO AI/ML working group and focuses on AI and ML for mission operations. Our task group seeks to educate other HWO working groups about AI and ML capabilities for mission operations, investigate how to bridge technology gaps, and enable new capabilities particularly in the areas of observational scheduling, instrument health monitoring, and downlink operations. We focus on mission tasking / scheduling both for mission analysis in development and operations. AI and ML for mission scheduling includes: tools to support proposal calls and review, ensuring fairness in calls for proposals, community peer reviews and ease workloads, as well as in-flight and ground software development (e.g., using natural language processing (NLP) to support process automation from requirements). AI and ML for the mission’s development and operations include 1) anomaly detection and prediction (from onboard and ground based tools) to monitor the spacecraft’s health, 2) ground-based automated scheduling for mission operations including long-term and short-term planning and maintenance, and 3) flight system flexible execution (as flight proven for Spitzer and JWST) to enable robust execution despite execution variations, and 4) data analysis for prioritization (e.g., real-time data evaluation leading to autonomous actions and adjustments, high-priority identification, onboard data compression, etc.). Incorporation of ML and AI will enable HWO to address the major science questions related to exoplanet characterization, general astrophysics, and solar system exploration and also extend the boundaries of space mission technologies.

Mark Moussa↗
DOE OSTI2025

CGSim: A Simulation Framework for Large Scale Distributed Computing Environment

Large-scale distributed computing infrastructures such as the Worldwide LHC Computing Grid (WLCG) require comprehensive simulation tools for evaluating performance, testing new algorithms, and optimizing resource allocation strategies. However, existing simulators suffer from limited scalability, hardwired algorithms, lack of real-time monitoring, and inability to generate datasets suitable for modern machine learning approaches. We present CGSim, a simulation framework for large-scale distributed computing environments that addresses these limitations. Built upon the validated SimGrid simulation framework, CGSim provides high-level abstractions for modeling heterogeneous grid environments while maintaining accuracy and scalability. Key features include a modular plugin mechanism for testing custom workflow scheduling and data movement policies, interactive real-time visualization dashboards, and automatic generation of event-level datasets suitable for AI-assisted performance modeling. We demonstrate CGSim’s capabilities through a comprehensive evaluation using production ATLAS PanDA workloads, showing significant calibration accuracy improvements across WLCG computing sites. Scalability experiments show near-linear scaling for multi-site simulations, with distributed workloads achieving 6 × better performance compared to single-site execution. The framework enables researchers to simulate WLCG-scale infrastructures with hundreds of sites and thousands of concurrent jobs within practical time budget constraints on commodity hardware.

Vatsavai, Sairam Sri [Brookhaven National Laborato↗
DOE OSTI2024

Real-time High-resolution X-Ray Computed Tomography

Computed Tomography (CT) serves as a key imaging technology that relies on computationally intensive filtering and back-projection algorithms for 3D image reconstruction. While conventional high-resolution image reconstruction (> 2K3) solutions provide quick results, they typically treat reconstruction as an offline workload to be performed remotely on large-scale HPC systems. The growing demand for post-construction AI-driven analytics and the need for real-time adjustments call for high-resolution reconstruction solutions that are feasible on local computing resources, i.e. a multi-GPU server at most. In this paper, we propose a novel approach that utilizes Tensor Cores to optimize image reconstruction without sacrificing precision. We also introduce a framework designed to enable real-time execution of end-to-end distributed image reconstruction in a multi-GPU environment. Evaluations conducted on a single Nvidia A100 and H100 GPU show performance improvements of 1.91 × and 2.15 × compared to highly optimized production libraries. Furthermore, our framework, when deployed on 8-card Nvidia A100 GPU system, demonstrates the ability to reconstruct real-world datasets into 20483 volumes (32 GB) in slightly more than one minute and 40963 volumes (256 GB) in 7 minutes.

Wu, Du↗
DOE OSTI2026

Integrating AI Data Centers with the Power Grid

The rapid expansion of artificial intelligence (AI) has triggered an unprecedented surge in electricity demand, with US data center energy use projected to double or triple 2023 levels by 2028. This exponential growth places strain on grid infrastructure, which can hinder timely construction of desired computing capacity. To bridge this supply-demand gap, utilities and AI developers are increasingly turning to demand flexibility, a strategy that incentivizes shifting or reducing power use during peak periods of grid stress. Data centers are uniquely equipped for flexible operations due to their digital workloads, built-in redundancy, and onsite energy assets. This article outlines four primary mechanisms to enable data center flexibility: computational load flexibility (shifting tasks temporally or geographically), flexible use of core facility infrastructure adjustments, energy storage utilization, and onsite electricity generation. To encourage adoption, utilities are deploying new tariff designs, including voluntary interruptible service riders, mandated flexibility requirements, and streamlined interconnection processes for flexible loads. For the highly capitalized and rapidly growing AI industry, the primary motivators for embracing these strategies are expediting facility interconnection, satisfying emerging regulatory mandates, and mitigating community resistance. While demand flexibility cannot substitute the long-term need for new bulk power generation, it serves as an essential, immediate solution for enabling near-term deployment. By transforming data centers from grid stressors into stabilizing assets, flexible operations can ensure reliable grid integration, ease market pressures, and support a resilient power system.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2022

Strategies for Integrating Deep Learning Surrogate Models with HPC Simulation Applications

The emerging trend of the convergence of high performance computing (HPC), machine learning/deep learning (ML/DL), and big data analytics presents a host of challenges for large-scale computing campaigns that seek best practices to interleave traditional scientific simulation-based workloads with ML/DL models. A portfolio of systematic approaches to incorporate deep learning into modeling and simulation serves a vital need when we support AI for science at a computing facility. In this paper, we evaluate several strategies for deploying deep learning surrogate models in a representative physics application on supercomputers at the Oak Ridge Leadership Computing Facility (OLCF). We discuss a set of recommended deployment architectures and implementation approaches. We analyze and evaluate these alternatives and show their performance and scalability up to 1000 GPUs on two mainstream platforms equipped with different deep learning hardware and software stacks.

Yin, Junqi↗