Parallel Stochastic Gradient Descent: The case for native GPU-side GPI

Size: px
Start display at page:

Download "Parallel Stochastic Gradient Descent: The case for native GPU-side GPI"

Transcription

1 Parallel Stochastic Gradient Descent: The case for native GPU-side GPI J. Keuper Competence Center High Performance Computing Fraunhofer ITWM, Kaiserslautern, Germany Mark Silberstein Accelerated Computer Systems Lab Technion Israel Institute of Technology

2 Accelerated Systems Lab Operating system support for accelerators GPU file system layer GPU networking API GPU virtual memory and huge data sets OS support for optimized GPU-SSD transfers GPU RDMA FPGA-CPU SoCs, near-data I/O accelerators SGX, accelerator security October 2016 Mark Technion 2

3 Outline Stochastic Gradual Descent in a nutshell Data-parallel distributed algorithm GPI implementation Communication bottlenecks and sparsity GPU-native GPI-2 communications October 2016 Mark Technion 3

4 SGD in a nutshell Common technique for training ML models Optimizes a loss function by modifying model's parameters Computations October 2016 Mark Technion 4

5 Parallelization Intrinsically sequential Computations Data parallel October 2016 Mark Technion 5

6 Parallelization via Master Worker Parameter server (w, s 0 ) (w, s 1 ) w w Worker Worker Sub-batch 0 Sub-batch 1 October 2016 Mark Technion 6

7 SGD with Deep Neural Networks Network output Layer n w n Forward pass Layer 2 w 2 Backward pass Layer 1 w 1 October 2016 Mark Technion 7

8 GPUs are used for computations Network output Layer n w n Forward pass Layer 2 w 2 Backward pass Layer 1 w 1 October 2016 Mark Technion 8

9 Data parallel SGD with DNNs Parameter server Layer n w n Layer n w n Layer n w n Layer 2 w 2 Layer 2 w 2 Layer 2 w 2 Layer 1 w 1 Layer 1 w 1 Layer 1 w 1 October 2016 Mark Technion 9

10 Problem: scalability limit Time until convergence Machines October 2016 Mark Technion 10

11 Communication bottleneck Communications grow linearly with the number of nodes! Communication time per node Keuper, Preundt Distributed training of deep neural networks: Theoretical and practical Limits of Parallel Scalability. To appear at MLHCP16 October 2016 Mark Technion 11

12 Steps toward improved scalability Asynchronous, zero-copy I/O Direct transfer from/to GPU memory GPI-2 Sparsity-aware compressed communications GPU-side networking Ongoing research October 2016 Mark Technion 12

13 Background: GPI October 2016 Mark Technion 13

14 Special Requirement 1: 1-sided communications Previous computations Layer 3 w 3 (w 3, s 0 ) Current computations Layer 2 w 2 Parameter server Layer 1 w 1 October 2016 Mark Technion 14

15 Special Requirement 1: 1-sided communications Previous computations Layer 3 w 3 (w 3, s 0 ) Current computations Layer 2 w 2 Parameter server Layer 1 w 1 GPI-2 PGAS model makes it easy to implement Cons: high memory requirements due to zero-copy October 2016 Mark Technion 15

16 Special requirement 2: Direct data transfer from GPU CPU NIC NIC CPU GPU GPU Extra-hop in CPU memory: extra latency lower bandwidth more complex pipelining October 2016 Mark Technion 16

17 Special requirement 2: Direct data transfer from GPU GPI-2 leverages GPUDirectRDMA, allows CPU to move data from GPU to NIC CPU NIC NIC CPU GPU GPU October 2016 Mark Technion 17

18 Reducing network traffic via smart compression Observation 1: updates many zeros or close to zero values are sent 40% values Fully connected layer: AlexNet iteration #10 October 2016 Mark Technion 18

19 Reducing network traffic via smart compression Observation 2: updates become more sparse toward convergence 95% values Fully connected layer: AlexNet Iteration #100 October 2016 Mark Technion 19

20 Special requirement 3: Predicated send send(vector,len, F()): foreach i<len if (F(vector[i])) send(vector[i]) E October 2016 Mark Technion 20

21 Problem with predicated send on GPUs send(vector,len, F()): foreach i<len if (F(vector[i])) send(vector[i]) Must be done on the GPU close to the data Must be done on CPU because no GPU send October 2016 Mark Technion 21

22 Problem with predicated send on GPUs send(vector,len, F()): foreach i<len if (F(vector[i])) send(vector[i]) Must be done on the GPU close to the data Goal: enable GPU networking October 2016 Mark Technion 22

23 GPUrdma and GPU-side networking Daoud, Wated, Silberstein GPUrdma: GPU-side library for high performance networking from GPU kernels, ROSS16 October 2016 Mark Technion 23

24 GPUrdma enables networking from GPU without CPU October 2016 Mark Technion 24

25 GPUrdma is faster for small messages GPU x3 CPU October 2016 Mark Technion 25

26 GPU-to-GPU GPU-side GPI-2 Preliminary results 52Gbit/s max throughput vs. 38Gbit/s GPI usec one-way latency 4x performance on toy applications Ongoing collaboration with NVIDIA and Mellanox October 2016 Mark Technion 26

27 Conclusions Improved scalability of distributed SGD requires Careful communication-computation overlap via one-sided communications Optimized GPU-NIC data path Smart sparsity-aware data compression GPU-side networking Thank you! October 2016 Mark Technion 27

CafeGPI. Single-Sided Communication for Scalable Deep Learning

CafeGPI. Single-Sided Communication for Scalable Deep Learning CafeGPI Single-Sided Communication for Scalable Deep Learning Janis Keuper itwm.fraunhofer.de/ml Competence Center High Performance Computing Fraunhofer ITWM, Kaiserslautern, Germany Deep Neural Networks

More information

Distributed Training of Deep Neural Networks: Theoretical and Practical Limits of Parallel Scalability

Distributed Training of Deep Neural Networks: Theoretical and Practical Limits of Parallel Scalability Distributed Training of Deep Neural Networks: Theoretical and Practical Limits of Parallel Scalability Janis Keuper Itwm.fraunhofer.de/ml Competence Center High Performance Computing Fraunhofer ITWM, Kaiserslautern,

More information

Asynchronous Parallel Stochastic Gradient Descent. A Numeric Core for Scalable Distributed Machine Learning Algorithms

Asynchronous Parallel Stochastic Gradient Descent. A Numeric Core for Scalable Distributed Machine Learning Algorithms Asynchronous Parallel Stochastic Gradient Descent A Numeric Core for Scalable Distributed Machine Learning Algorithms J. Keuper and F.-J. Pfreundt Competence Center High Performance Computing Fraunhofer

More information

Towards Scalable Machine Learning

Towards Scalable Machine Learning Towards Scalable Machine Learning Janis Keuper itwm.fraunhofer.de/ml Competence Center High Performance Computing Fraunhofer ITWM, Kaiserslautern, Germany Fraunhofer Center Machnine Larning Outline I Introduction

More information

GPUnet: networking abstractions for GPU programs

GPUnet: networking abstractions for GPU programs net: networking abstractions for programs Mark Silberstein Technion Israel Institute of Technology Sangman Kim, Seonggu Huh, Xinya Zhang Yige Hu, Emmett Witchel University of Texas at Austin Amir Wated

More information

Poseidon: An Efficient Communication Architecture for Distributed Deep Learning on GPU Clusters

Poseidon: An Efficient Communication Architecture for Distributed Deep Learning on GPU Clusters Poseidon: An Efficient Communication Architecture for Distributed Deep Learning on GPU Clusters Hao Zhang Zeyu Zheng, Shizhen Xu, Wei Dai, Qirong Ho, Xiaodan Liang, Zhiting Hu, Jianliang Wei, Pengtao Xie,

More information

Accelerator-centric operating systems

Accelerator-centric operating systems Accelerator-centric operating systems Rethinking the role of s in modern computers Mark Silberstein EE, Technion System design challenge: Programmability and Performance 2 System design challenge: Programmability

More information

Technologies for High Performance Data Analytics

Technologies for High Performance Data Analytics Technologies for High Performance Data Analytics Dr. Jens Krüger Fraunhofer ITWM 1 Fraunhofer ITWM n Institute for Industrial Mathematics n Located in Kaiserslautern, Germany n Staff: ~ 240 employees +

More information

CS 179 Lecture 16. Logistic Regression & Parallel SGD

CS 179 Lecture 16. Logistic Regression & Parallel SGD CS 179 Lecture 16 Logistic Regression & Parallel SGD 1 Outline logistic regression (stochastic) gradient descent parallelizing SGD for neural nets (with emphasis on Google s distributed neural net implementation)

More information

Memory Bandwidth and Low Precision Computation. CS6787 Lecture 10 Fall 2018

Memory Bandwidth and Low Precision Computation. CS6787 Lecture 10 Fall 2018 Memory Bandwidth and Low Precision Computation CS6787 Lecture 10 Fall 2018 Memory as a Bottleneck So far, we ve just been talking about compute e.g. techniques to decrease the amount of compute by decreasing

More information

High Performance Computing

High Performance Computing High Performance Computing 9th Lecture 2016/10/28 YUKI ITO 1 Selected Paper: vdnn: Virtualized Deep Neural Networks for Scalable, MemoryEfficient Neural Network Design Minsoo Rhu, Natalia Gimelshein, Jason

More information

Memory Bandwidth and Low Precision Computation. CS6787 Lecture 9 Fall 2017

Memory Bandwidth and Low Precision Computation. CS6787 Lecture 9 Fall 2017 Memory Bandwidth and Low Precision Computation CS6787 Lecture 9 Fall 2017 Memory as a Bottleneck So far, we ve just been talking about compute e.g. techniques to decrease the amount of compute by decreasing

More information

Building the Most Efficient Machine Learning System

Building the Most Efficient Machine Learning System Building the Most Efficient Machine Learning System Mellanox The Artificial Intelligence Interconnect Company June 2017 Mellanox Overview Company Headquarters Yokneam, Israel Sunnyvale, California Worldwide

More information

Intelligent Hybrid Flash Management

Intelligent Hybrid Flash Management Intelligent Hybrid Flash Management Jérôme Gaysse Senior Technology&Market Analyst jerome.gaysse@silinnov-consulting.com Flash Memory Summit 2018 Santa Clara, CA 1 Research context Analysis of system &

More information

Parallelism. CS6787 Lecture 8 Fall 2017

Parallelism. CS6787 Lecture 8 Fall 2017 Parallelism CS6787 Lecture 8 Fall 2017 So far We ve been talking about algorithms We ve been talking about ways to optimize their parameters But we haven t talked about the underlying hardware How does

More information

High-Performance Data Loading and Augmentation for Deep Neural Network Training

High-Performance Data Loading and Augmentation for Deep Neural Network Training High-Performance Data Loading and Augmentation for Deep Neural Network Training Trevor Gale tgale@ece.neu.edu Steven Eliuk steven.eliuk@gmail.com Cameron Upright c.upright@samsung.com Roadmap 1. The General-Purpose

More information

PacketShader: A GPU-Accelerated Software Router

PacketShader: A GPU-Accelerated Software Router PacketShader: A GPU-Accelerated Software Router Sangjin Han In collaboration with: Keon Jang, KyoungSoo Park, Sue Moon Advanced Networking Lab, CS, KAIST Networked and Distributed Computing Systems Lab,

More information

Democratizing Machine Learning on Kubernetes

Democratizing Machine Learning on Kubernetes Democratizing Machine Learning on Kubernetes Joy Qiao, Senior Solution Architect - AI and Research Group, Microsoft Lachlan Evenson - Principal Program Manager AKS/ACS, Microsoft Who are we? The Data Scientist

More information

Exploiting InfiniBand and GPUDirect Technology for High Performance Collectives on GPU Clusters

Exploiting InfiniBand and GPUDirect Technology for High Performance Collectives on GPU Clusters Exploiting InfiniBand and Direct Technology for High Performance Collectives on Clusters Ching-Hsiang Chu chu.368@osu.edu Department of Computer Science and Engineering The Ohio State University OSU Booth

More information

N V M e o v e r F a b r i c s -

N V M e o v e r F a b r i c s - N V M e o v e r F a b r i c s - H i g h p e r f o r m a n c e S S D s n e t w o r k e d f o r c o m p o s a b l e i n f r a s t r u c t u r e Rob Davis, VP Storage Technology, Mellanox OCP Evolution Server

More information

Pouya Kousha Fall 2018 CSE 5194 Prof. DK Panda

Pouya Kousha Fall 2018 CSE 5194 Prof. DK Panda Pouya Kousha Fall 2018 CSE 5194 Prof. DK Panda 1 Motivation And Intro Programming Model Spark Data Transformation Model Construction Model Training Model Inference Execution Model Data Parallel Training

More information

MELLANOX EDR UPDATE & GPUDIRECT MELLANOX SR. SE 정연구

MELLANOX EDR UPDATE & GPUDIRECT MELLANOX SR. SE 정연구 MELLANOX EDR UPDATE & GPUDIRECT MELLANOX SR. SE 정연구 Leading Supplier of End-to-End Interconnect Solutions Analyze Enabling the Use of Data Store ICs Comprehensive End-to-End InfiniBand and Ethernet Portfolio

More information

Accelerating Data Centers Using NVMe and CUDA

Accelerating Data Centers Using NVMe and CUDA Accelerating Data Centers Using NVMe and CUDA Stephen Bates, PhD Technical Director, CSTO, PMC-Sierra Santa Clara, CA 1 Project Donard @ PMC-Sierra Donard is a PMC CTO project that leverages NVM Express

More information

Building the Most Efficient Machine Learning System

Building the Most Efficient Machine Learning System Building the Most Efficient Machine Learning System Mellanox The Artificial Intelligence Interconnect Company June 2017 Mellanox Overview Company Headquarters Yokneam, Israel Sunnyvale, California Worldwide

More information

The Future of High Performance Interconnects

The Future of High Performance Interconnects The Future of High Performance Interconnects Ashrut Ambastha HPC Advisory Council Perth, Australia :: August 2017 When Algorithms Go Rogue 2017 Mellanox Technologies 2 When Algorithms Go Rogue 2017 Mellanox

More information

Adaptable Computing The Future of FPGA Acceleration. Dan Gibbons, VP Software Development June 6, 2018

Adaptable Computing The Future of FPGA Acceleration. Dan Gibbons, VP Software Development June 6, 2018 Adaptable Computing The Future of FPGA Acceleration Dan Gibbons, VP Software Development June 6, 2018 Adaptable Accelerated Computing Page 2 Three Big Trends The Evolution of Computing Trend to Heterogeneous

More information

Near-Data Processing for Differentiable Machine Learning Models

Near-Data Processing for Differentiable Machine Learning Models Near-Data Processing for Differentiable Machine Learning Models Hyeokjun Choe 1, Seil Lee 1, Hyunha Nam 1, Seongsik Park 1, Seijoon Kim 1, Eui-Young Chung 2 and Sungroh Yoon 1,3 1 Electrical and Computer

More information

Scalable Distributed Training with Parameter Hub: a whirlwind tour

Scalable Distributed Training with Parameter Hub: a whirlwind tour Scalable Distributed Training with Parameter Hub: a whirlwind tour TVM Stack Optimization High-Level Differentiable IR Tensor Expression IR AutoTVM LLVM, CUDA, Metal VTA AutoVTA Edge FPGA Cloud FPGA ASIC

More information

CS 6453: Parameter Server. Soumya Basu March 7, 2017

CS 6453: Parameter Server. Soumya Basu March 7, 2017 CS 6453: Parameter Server Soumya Basu March 7, 2017 What is a Parameter Server? Server for large scale machine learning problems Machine learning tasks in a nutshell: Feature Extraction (1, 1, 1) (2, -1,

More information

2017 Storage Developer Conference. Mellanox Technologies. All Rights Reserved.

2017 Storage Developer Conference. Mellanox Technologies. All Rights Reserved. Ethernet Storage Fabrics Using RDMA with Fast NVMe-oF Storage to Reduce Latency and Improve Efficiency Kevin Deierling & Idan Burstein Mellanox Technologies 1 Storage Media Technology Storage Media Access

More information

Onto Petaflops with Kubernetes

Onto Petaflops with Kubernetes Onto Petaflops with Kubernetes Vishnu Kannan Google Inc. vishh@google.com Key Takeaways Kubernetes can manage hardware accelerators at Scale Kubernetes provides a playground for ML ML journey with Kubernetes

More information

SDA: Software-Defined Accelerator for Large- Scale DNN Systems

SDA: Software-Defined Accelerator for Large- Scale DNN Systems SDA: Software-Defined Accelerator for Large- Scale DNN Systems Jian Ouyang, 1 Shiding Lin, 1 Wei Qi, Yong Wang, Bo Yu, Song Jiang, 2 1 Baidu, Inc. 2 Wayne State University Introduction of Baidu A dominant

More information

Profiling DNN Workloads on a Volta-based DGX-1 System

Profiling DNN Workloads on a Volta-based DGX-1 System Profiling DNN Workloads on a Volta-based DGX-1 System Saiful A. Mojumder 1, Marcia S Louis 1, Yifan Sun 2, Amir Kavyan Ziabari 3, José L. Abellán 4, John Kim 5, David Kaeli 2, Ajay Joshi 1 1 ECE Department,

More information

Decentralized and Distributed Machine Learning Model Training with Actors

Decentralized and Distributed Machine Learning Model Training with Actors Decentralized and Distributed Machine Learning Model Training with Actors Travis Addair Stanford University taddair@stanford.edu Abstract Training a machine learning model with terabytes to petabytes of

More information

Asynchronous Parallel Stochastic Gradient Descent

Asynchronous Parallel Stochastic Gradient Descent Asynchronous Parallel Stochastic Gradient Descent ABSTRACT A Numeric Core for Scalable Distributed Machine Learning Algorithms The implementation of a vast majority of machine learning (ML) algorithms

More information

Linear Regression Optimization

Linear Regression Optimization Gradient Descent Linear Regression Optimization Goal: Find w that minimizes f(w) f(w) = Xw y 2 2 Closed form solution exists Gradient Descent is iterative (Intuition: go downhill!) n w * w Scalar objective:

More information

Deep Learning Accelerators

Deep Learning Accelerators Deep Learning Accelerators Abhishek Srivastava (as29) Samarth Kulshreshtha (samarth5) University of Illinois, Urbana-Champaign Submitted as a requirement for CS 433 graduate student project Outline Introduction

More information

Why DNN Works for Speech and How to Make it More Efficient?

Why DNN Works for Speech and How to Make it More Efficient? Why DNN Works for Speech and How to Make it More Efficient? Hui Jiang Department of Electrical Engineering and Computer Science Lassonde School of Engineering, York University, CANADA Joint work with Y.

More information

Accelerating Ceph with Flash and High Speed Networks

Accelerating Ceph with Flash and High Speed Networks Accelerating Ceph with Flash and High Speed Networks Dror Goldenberg VP Software Architecture Santa Clara, CA 1 The New Open Cloud Era Compute Software Defined Network Object, Block Software Defined Storage

More information

Maximizing Server Efficiency from μarch to ML accelerators. Michael Ferdman

Maximizing Server Efficiency from μarch to ML accelerators. Michael Ferdman Maximizing Server Efficiency from μarch to ML accelerators Michael Ferdman Maximizing Server Efficiency from μarch to ML accelerators Michael Ferdman Maximizing Server Efficiency with ML accelerators Michael

More information

THOUGHTS ABOUT THE FUTURE OF I/O

THOUGHTS ABOUT THE FUTURE OF I/O THOUGHTS ABOUT THE FUTURE OF I/O Dagstuhl Seminar Challenges and Opportunities of User-Level File Systems for HPC Franz-Josef Pfreundt, May 2017 Deep Learning I/O Challenges Memory Centric Computing :

More information

GPUfs: Integrating a file system with GPUs

GPUfs: Integrating a file system with GPUs GPUfs: Integrating a file system with GPUs Mark Silberstein (UT Austin/Technion) Bryan Ford (Yale), Idit Keidar (Technion) Emmett Witchel (UT Austin) 1 Traditional System Architecture Applications OS CPU

More information

Application Acceleration Beyond Flash Storage

Application Acceleration Beyond Flash Storage Application Acceleration Beyond Flash Storage Session 303C Mellanox Technologies Flash Memory Summit July 2014 Accelerating Applications, Step-by-Step First Steps Make compute fast Moore s Law Make storage

More information

NVIDIA GPU CLOUD DEEP LEARNING FRAMEWORKS

NVIDIA GPU CLOUD DEEP LEARNING FRAMEWORKS TECHNICAL OVERVIEW NVIDIA GPU CLOUD DEEP LEARNING FRAMEWORKS A Guide to the Optimized Framework Containers on NVIDIA GPU Cloud Introduction Artificial intelligence is helping to solve some of the most

More information

The NE010 iwarp Adapter

The NE010 iwarp Adapter The NE010 iwarp Adapter Gary Montry Senior Scientist +1-512-493-3241 GMontry@NetEffect.com Today s Data Center Users Applications networking adapter LAN Ethernet NAS block storage clustering adapter adapter

More information

Networking at the Speed of Light

Networking at the Speed of Light Networking at the Speed of Light Dror Goldenberg VP Software Architecture MaRS Workshop April 2017 Cloud The Software Defined Data Center Resource virtualization Efficient services VM, Containers uservices

More information

Research Faculty Summit Systems Fueling future disruptions

Research Faculty Summit Systems Fueling future disruptions Research Faculty Summit 2018 Systems Fueling future disruptions Wolong: A Back-end Optimizer for Deep Learning Computation Jilong Xue Researcher, Microsoft Research Asia System Challenge in Deep Learning

More information

Mellanox InfiniBand Solutions Accelerate Oracle s Data Center and Cloud Solutions

Mellanox InfiniBand Solutions Accelerate Oracle s Data Center and Cloud Solutions Mellanox InfiniBand Solutions Accelerate Oracle s Data Center and Cloud Solutions Providing Superior Server and Storage Performance, Efficiency and Return on Investment As Announced and Demonstrated at

More information

Scaling Distributed Machine Learning

Scaling Distributed Machine Learning Scaling Distributed Machine Learning with System and Algorithm Co-design Mu Li Thesis Defense CSD, CMU Feb 2nd, 2017 nx min w f i (w) Distributed systems i=1 Large scale optimization methods Large-scale

More information

Using MRAM to Create Intelligent SSDs

Using MRAM to Create Intelligent SSDs Using MRAM to Create Intelligent SSDs Jérôme Gaysse Senior Technology&Market Analyst jerome.gaysse@silinnov-consulting.com Santa Clara, CA 1 Study context Analysis of system & application Performance modeling

More information

Toward Scalable Deep Learning

Toward Scalable Deep Learning 한국정보과학회 인공지능소사이어티 머신러닝연구회 두번째딥러닝워크샵 2015.10.16 Toward Scalable Deep Learning 윤성로 Electrical and Computer Engineering Seoul National University http://data.snu.ac.kr Breakthrough: Big Data + Machine Learning

More information

TensorFlow: A System for Learning-Scale Machine Learning. Google Brain

TensorFlow: A System for Learning-Scale Machine Learning. Google Brain TensorFlow: A System for Learning-Scale Machine Learning Google Brain The Problem Machine learning is everywhere This is in large part due to: 1. Invention of more sophisticated machine learning models

More information

Overcoming the Memory System Challenge in Dataflow Processing. Darren Jones, Wave Computing Drew Wingard, Sonics

Overcoming the Memory System Challenge in Dataflow Processing. Darren Jones, Wave Computing Drew Wingard, Sonics Overcoming the Memory System Challenge in Dataflow Processing Darren Jones, Wave Computing Drew Wingard, Sonics Current Technology Limits Deep Learning Performance Deep Learning Dataflow Graph Existing

More information

Hardened Security in the Cloud Bob Doud, Sr. Director Marketing March, 2018

Hardened Security in the Cloud Bob Doud, Sr. Director Marketing March, 2018 Hardened Security in the Cloud Bob Doud, Sr. Director Marketing March, 2018 1 Cloud Computing is Growing at an Astounding Rate Many compelling reasons for business to move to the cloud Cost, uptime, easy-expansion,

More information

Deep Learning with Tensorflow AlexNet

Deep Learning with Tensorflow   AlexNet Machine Learning and Computer Vision Group Deep Learning with Tensorflow http://cvml.ist.ac.at/courses/dlwt_w17/ AlexNet Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E. Hinton, "Imagenet classification

More information

Enabling Efficient Use of UPC and OpenSHMEM PGAS models on GPU Clusters

Enabling Efficient Use of UPC and OpenSHMEM PGAS models on GPU Clusters Enabling Efficient Use of UPC and OpenSHMEM PGAS models on GPU Clusters Presentation at GTC 2014 by Dhabaleswar K. (DK) Panda The Ohio State University E-mail: panda@cse.ohio-state.edu http://www.cse.ohio-state.edu/~panda

More information

Exploiting the OpenPOWER Platform for Big Data Analytics and Cognitive. Rajesh Bordawekar and Ruchir Puri IBM T. J. Watson Research Center

Exploiting the OpenPOWER Platform for Big Data Analytics and Cognitive. Rajesh Bordawekar and Ruchir Puri IBM T. J. Watson Research Center Exploiting the OpenPOWER Platform for Big Data Analytics and Cognitive Rajesh Bordawekar and Ruchir Puri IBM T. J. Watson Research Center 3/17/2015 2014 IBM Corporation Outline IBM OpenPower Platform Accelerating

More information

Deep Learning and Its Applications

Deep Learning and Its Applications Convolutional Neural Network and Its Application in Image Recognition Oct 28, 2016 Outline 1 A Motivating Example 2 The Convolutional Neural Network (CNN) Model 3 Training the CNN Model 4 Issues and Recent

More information

Evaluating On-Node GPU Interconnects for Deep Learning Workloads

Evaluating On-Node GPU Interconnects for Deep Learning Workloads Evaluating On-Node GPU Interconnects for Deep Learning Workloads NATHAN TALLENT, NITIN GAWANDE, CHARLES SIEGEL ABHINAV VISHNU, ADOLFY HOISIE Pacific Northwest National Lab PMBS 217 (@ SC) November 13,

More information

Profiling the Performance of Binarized Neural Networks. Daniel Lerner, Jared Pierce, Blake Wetherton, Jialiang Zhang

Profiling the Performance of Binarized Neural Networks. Daniel Lerner, Jared Pierce, Blake Wetherton, Jialiang Zhang Profiling the Performance of Binarized Neural Networks Daniel Lerner, Jared Pierce, Blake Wetherton, Jialiang Zhang 1 Outline Project Significance Prior Work Research Objectives Hypotheses Testing Framework

More information

Optimization for Machine Learning

Optimization for Machine Learning with a focus on proximal gradient descent algorithm Department of Computer Science and Engineering Outline 1 History & Trends 2 Proximal Gradient Descent 3 Three Applications A Brief History A. Convex

More information

CIS581: Computer Vision and Computational Photography Project 4, Part B: Convolutional Neural Networks (CNNs) Due: Dec.11, 2017 at 11:59 pm

CIS581: Computer Vision and Computational Photography Project 4, Part B: Convolutional Neural Networks (CNNs) Due: Dec.11, 2017 at 11:59 pm CIS581: Computer Vision and Computational Photography Project 4, Part B: Convolutional Neural Networks (CNNs) Due: Dec.11, 2017 at 11:59 pm Instructions CNNs is a team project. The maximum size of a team

More information

IO virtualization. Michael Kagan Mellanox Technologies

IO virtualization. Michael Kagan Mellanox Technologies IO virtualization Michael Kagan Mellanox Technologies IO Virtualization Mission non-stop s to consumers Flexibility assign IO resources to consumer as needed Agility assignment of IO resources to consumer

More information

Deep Learning Frameworks with Spark and GPUs

Deep Learning Frameworks with Spark and GPUs Deep Learning Frameworks with Spark and GPUs Abstract Spark is a powerful, scalable, real-time data analytics engine that is fast becoming the de facto hub for data science and big data. However, in parallel,

More information

Asynchronous Stochastic Gradient Descent on GPU: Is It Really Better than CPU?

Asynchronous Stochastic Gradient Descent on GPU: Is It Really Better than CPU? Asynchronous Stochastic Gradient Descent on GPU: Is It Really Better than CPU? Florin Rusu Yujing Ma, Martin Torres (Ph.D. students) University of California Merced Machine Learning (ML) Boom Two SIGMOD

More information

High-Performance Training for Deep Learning and Computer Vision HPC

High-Performance Training for Deep Learning and Computer Vision HPC High-Performance Training for Deep Learning and Computer Vision HPC Panel at CVPR-ECV 18 by Dhabaleswar K. (DK) Panda The Ohio State University E-mail: panda@cse.ohio-state.edu http://www.cse.ohio-state.edu/~panda

More information

How to Network Flash Storage Efficiently at Hyperscale. Flash Memory Summit 2017 Santa Clara, CA 1

How to Network Flash Storage Efficiently at Hyperscale. Flash Memory Summit 2017 Santa Clara, CA 1 How to Network Flash Storage Efficiently at Hyperscale Manoj Wadekar Michael Kagan Flash Memory Summit 2017 Santa Clara, CA 1 ebay Hyper scale Infrastructure Search Front-End & Product Hadoop Object Store

More information

Efficient Communication Library for Large-Scale Deep Learning

Efficient Communication Library for Large-Scale Deep Learning IBM Research AI Efficient Communication Library for Large-Scale Deep Learning Mar 26, 2018 Minsik Cho (minsikcho@us.ibm.com) Deep Learning changing Our Life Automotive/transportation Security/public safety

More information

Multi-dimensional Parallel Training of Winograd Layer on Memory-Centric Architecture

Multi-dimensional Parallel Training of Winograd Layer on Memory-Centric Architecture The 51st Annual IEEE/ACM International Symposium on Microarchitecture Multi-dimensional Parallel Training of Winograd Layer on Memory-Centric Architecture Byungchul Hong Yeonju Ro John Kim FuriosaAI Samsung

More information

A performance comparison of Deep Learning frameworks on KNL

A performance comparison of Deep Learning frameworks on KNL A performance comparison of Deep Learning frameworks on KNL R. Zanella, G. Fiameni, M. Rorro Middleware, Data Management - SCAI - CINECA IXPUG Bologna, March 5, 2018 Table of Contents 1. Problem description

More information

Parallel Deep Network Training

Parallel Deep Network Training Lecture 19: Parallel Deep Network Training Parallel Computer Architecture and Programming How would you describe this professor? Easy? Mean? Boring? Nerdy? Professor classification task Classifies professors

More information

Harp-DAAL for High Performance Big Data Computing

Harp-DAAL for High Performance Big Data Computing Harp-DAAL for High Performance Big Data Computing Large-scale data analytics is revolutionizing many business and scientific domains. Easy-touse scalable parallel techniques are necessary to process big

More information

Training Deep Neural Networks (in parallel)

Training Deep Neural Networks (in parallel) Lecture 9: Training Deep Neural Networks (in parallel) Visual Computing Systems How would you describe this professor? Easy? Mean? Boring? Nerdy? Professor classification task Classifies professors as

More information

Interconnect Your Future

Interconnect Your Future Interconnect Your Future Smart Interconnect for Next Generation HPC Platforms Gilad Shainer, August 2016, 4th Annual MVAPICH User Group (MUG) Meeting Mellanox Connects the World s Fastest Supercomputer

More information

NFS/RDMA over 40Gbps iwarp Wael Noureddine Chelsio Communications

NFS/RDMA over 40Gbps iwarp Wael Noureddine Chelsio Communications NFS/RDMA over 40Gbps iwarp Wael Noureddine Chelsio Communications Outline RDMA Motivating trends iwarp NFS over RDMA Overview Chelsio T5 support Performance results 2 Adoption Rate of 40GbE Source: Crehan

More information

The rcuda middleware and applications

The rcuda middleware and applications The rcuda middleware and applications Will my application work with rcuda? rcuda currently provides binary compatibility with CUDA 5.0, virtualizing the entire Runtime API except for the graphics functions,

More information

High Performance Computing

High Performance Computing High Performance Computing Dror Goldenberg, HPCAC Switzerland Conference March 2015 End-to-End Interconnect Solutions for All Platforms Highest Performance and Scalability for X86, Power, GPU, ARM and

More information

Distributed Training of Deep Neural Networks: Theoretical and Practical Limits of Parallel Scalability.

Distributed Training of Deep Neural Networks: Theoretical and Practical Limits of Parallel Scalability. Distributed Training of Deep Neural Networks: Theoretical and Practical Limits of Parallel Scalability. arxiv:1609.06870v4 [cs.cv] 5 Dec 2016 Janis Keuper Fraunhofer ITWM Competence Center High Performance

More information

Oncilla - a Managed GAS Runtime for Accelerating Data Warehousing Queries

Oncilla - a Managed GAS Runtime for Accelerating Data Warehousing Queries Oncilla - a Managed GAS Runtime for Accelerating Data Warehousing Queries Jeffrey Young, Alex Merritt, Se Hoon Shon Advisor: Sudhakar Yalamanchili 4/16/13 Sponsors: Intel, NVIDIA, NSF 2 The Problem Big

More information

World s most advanced data center accelerator for PCIe-based servers

World s most advanced data center accelerator for PCIe-based servers NVIDIA TESLA P100 GPU ACCELERATOR World s most advanced data center accelerator for PCIe-based servers HPC data centers need to support the ever-growing demands of scientists and researchers while staying

More information

Deep learning prevalence. first neuroscience department. Spiking Neuron Operant conditioning First 1 Billion transistor processor

Deep learning prevalence. first neuroscience department. Spiking Neuron Operant conditioning First 1 Billion transistor processor WELCOME TO Operant conditioning 1938 Spiking Neuron 1952 first neuroscience department 1964 Deep learning prevalence mid 2000s The Turing Machine 1936 Transistor 1947 First computer science department

More information

Maximum Performance. How to get it and how to avoid pitfalls. Christoph Lameter, PhD

Maximum Performance. How to get it and how to avoid pitfalls. Christoph Lameter, PhD Maximum Performance How to get it and how to avoid pitfalls Christoph Lameter, PhD cl@linux.com Performance Just push a button? Systems are optimized by default for good general performance in all areas.

More information

DDN. DDN Updates. DataDirect Neworks Japan, Inc Nobu Hashizume. DDN Storage 2018 DDN Storage 1

DDN. DDN Updates. DataDirect Neworks Japan, Inc Nobu Hashizume. DDN Storage 2018 DDN Storage 1 1 DDN DDN Updates DataDirect Neworks Japan, Inc Nobu Hashizume DDN Storage 2018 DDN Storage 1 2 DDN A Broad Range of Technologies to Best Address Your Needs Your Use Cases Research Big Data Enterprise

More information

Interconnect Your Future

Interconnect Your Future Interconnect Your Future Gilad Shainer 2nd Annual MVAPICH User Group (MUG) Meeting, August 2014 Complete High-Performance Scalable Interconnect Infrastructure Comprehensive End-to-End Software Accelerators

More information

SDA: Software-Defined Accelerator for Large- Scale DNN Systems

SDA: Software-Defined Accelerator for Large- Scale DNN Systems SDA: Software-Defined Accelerator for Large- Scale DNN Systems Jian Ouyang, 1 Shiding Lin, 1 Wei Qi, 1 Yong Wang, 1 Bo Yu, 1 Song Jiang, 2 1 Baidu, Inc. 2 Wayne State University Introduction of Baidu A

More information

Eleos: Exit-Less OS Services for SGX Enclaves

Eleos: Exit-Less OS Services for SGX Enclaves Eleos: Exit-Less OS Services for SGX Enclaves Meni Orenbach Marina Minkin Pavel Lifshits Mark Silberstein Accelerated Computing Systems Lab Haifa, Israel What do we do? Improve performance: I/O intensive

More information

RDMA and Hardware Support

RDMA and Hardware Support RDMA and Hardware Support SIGCOMM Topic Preview 2018 Yibo Zhu Microsoft Research 1 The (Traditional) Journey of Data How app developers see the network Under the hood This architecture had been working

More information

Asynchronous Peer-to-Peer Device Communication

Asynchronous Peer-to-Peer Device Communication 13th ANNUAL WORKSHOP 2017 Asynchronous Peer-to-Peer Device Communication Feras Daoud, Leon Romanovsky [ 28 March, 2017 ] Agenda Peer-to-Peer communication PeerDirect technology PeerDirect and PeerDirect

More information

Exploiting Full Potential of GPU Clusters with InfiniBand using MVAPICH2-GDR

Exploiting Full Potential of GPU Clusters with InfiniBand using MVAPICH2-GDR Exploiting Full Potential of GPU Clusters with InfiniBand using MVAPICH2-GDR Presentation at Mellanox Theater () Dhabaleswar K. (DK) Panda - The Ohio State University panda@cse.ohio-state.edu Outline Communication

More information

Paving the Road to Exascale

Paving the Road to Exascale Paving the Road to Exascale Gilad Shainer August 2015, MVAPICH User Group (MUG) Meeting The Ever Growing Demand for Performance Performance Terascale Petascale Exascale 1 st Roadrunner 2000 2005 2010 2015

More information

Throughput-Optimized OpenCL-based FPGA Accelerator for Large-Scale Convolutional Neural Networks

Throughput-Optimized OpenCL-based FPGA Accelerator for Large-Scale Convolutional Neural Networks Throughput-Optimized OpenCL-based FPGA Accelerator for Large-Scale Convolutional Neural Networks Naveen Suda, Vikas Chandra *, Ganesh Dasika *, Abinash Mohanty, Yufei Ma, Sarma Vrudhula, Jae-sun Seo, Yu

More information

Deep Learning on Modern Architectures. Keren Zhou 4/17/2017

Deep Learning on Modern Architectures. Keren Zhou 4/17/2017 Deep Learning on Modern Architectures Keren Zhou 4/17/2017 HPC Software Stack Application Algorithm Data Layout CPU GPU MIC Others HPC Software Stack Deep Learning Algorithm Data Layout CPU GPU MIC Others

More information

NVIDIA Think about Computing as Heterogeneous One Leo Liao, 1/29/2106, NTU

NVIDIA Think about Computing as Heterogeneous One Leo Liao, 1/29/2106, NTU NVIDIA Think about Computing as Heterogeneous One Leo Liao, 1/29/2106, NTU GPGPU opens the door for co-design HPC, moreover middleware-support embedded system designs to harness the power of GPUaccelerated

More information

Scale-Out Acceleration for Machine Learning

Scale-Out Acceleration for Machine Learning Scale-Out Acceleration for Machine Learning Jongse Park Hardik Sharma Divya Mahajan Joon Kyung Kim Preston Olds Hadi Esmaeilzadeh Alternative Computing Technologies (ACT) Lab Georgia Institute of Technology

More information

Parameter Hub: a Rack-Scale Parameter Server for Distributed Deep Neural Network Training

Parameter Hub: a Rack-Scale Parameter Server for Distributed Deep Neural Network Training Parameter Hub: a Rack-Scale Parameter Server for Distributed Deep Neural Network Training University of Washington, Microsoft Research Abstract Distributed deep neural network (DDNN) training constitutes

More information

Entropy-Aware I/O Pipelining for Large-Scale Deep Learning on HPC Systems

Entropy-Aware I/O Pipelining for Large-Scale Deep Learning on HPC Systems Entropy-Aware I/O Pipelining for Large-Scale Deep Learning on HPC Systems Yue Zhu Fahim Chowdhury Huansong Fu Adam Moody Kathryn Mohror Kento Sato Weikuan Yu Florida State University Lawrence Livermore

More information

Efficient and Scalable Multi-Source Streaming Broadcast on GPU Clusters for Deep Learning

Efficient and Scalable Multi-Source Streaming Broadcast on GPU Clusters for Deep Learning Efficient and Scalable Multi-Source Streaming Broadcast on Clusters for Deep Learning Ching-Hsiang Chu 1, Xiaoyi Lu 1, Ammar A. Awan 1, Hari Subramoni 1, Jahanzeb Hashmi 1, Bracy Elton 2 and Dhabaleswar

More information

GUNREAL: GPU-accelerated UNsupervised REinforcement and Auxiliary Learning

GUNREAL: GPU-accelerated UNsupervised REinforcement and Auxiliary Learning GUNREAL: GPU-accelerated UNsupervised REinforcement and Auxiliary Learning Koichi Shirahata, Youri Coppens, Takuya Fukagai, Yasumoto Tomita, and Atsushi Ike FUJITSU LABORATORIES LTD. March 27, 2018 0 Deep

More information

High-Performance Broadcast for Streaming and Deep Learning

High-Performance Broadcast for Streaming and Deep Learning High-Performance Broadcast for Streaming and Deep Learning Ching-Hsiang Chu chu.368@osu.edu Department of Computer Science and Engineering The Ohio State University OSU Booth - SC17 2 Outline Introduction

More information

In-Network Computing. Sebastian Kalcher, Senior System Engineer HPC. May 2017

In-Network Computing. Sebastian Kalcher, Senior System Engineer HPC. May 2017 In-Network Computing Sebastian Kalcher, Senior System Engineer HPC May 2017 Exponential Data Growth The Need for Intelligent and Faster Interconnect CPU-Centric (Onload) Data-Centric (Offload) Must Wait

More information