O'Reilly logo

Stay ahead with the world's most comprehensive technology and business learning platform.

With Safari, you learn the way you learn best. Get unlimited access to videos, live online training, learning paths, books, tutorials, and more.

Start Free Trial

No credit card required

IBM PowerAI: Deep Learning Unleashed on IBM Power Systems Servers

Book Description

Abstract

This IBM® Redbooks® publication is a guide about the IBM PowerAI Deep Learning solution. This book provides an introduction to artificial intelligence (AI) and deep learning (DL), IBM PowerAI, and components of IBM PowerAI, deploying IBM PowerAI, guidelines for working with data and creating models, an introduction to IBM Spectrum™ Conductor Deep Learning Impact (DLI), and case scenarios.

IBM PowerAI started as a package of software distributions of many of the major DL software frameworks for model training, such as TensorFlow, Caffe, Torch, Theano, and the associated libraries, such as CUDA Deep Neural Network (cuDNN). The IBM PowerAI software is optimized for performance by using the IBM Power Systems™ servers that are integrated with NVLink. The AI stack foundation starts with servers with accelerators. graphical processing unit (GPU) accelerators are well-suited for the compute-intensive nature of DL training, and servers with the highest CPU to GPU bandwidth, such as IBM Power Systems servers, enable the high-performance data transfer that is required for larger and more complex DL models.

This publication targets technical readers, including developers, IT specialists, systems architects, brand specialist, sales team, and anyone looking for a guide about how to understand the IBM PowerAI Deep Learning architecture, framework configuration, application and workload configuration, and user infrastructure.

Table of Contents

  1. Front cover
  2. Notices
    1. Trademarks
  3. Preface
    1. Authors
    2. Now you can become a published author, too!
    3. Comments welcome
    4. Stay connected to IBM Redbooks
  4. Chapter 1. Introduction to artificial intelligence and deep learning
    1. 1.1 Deep learning
      1. 1.1.1 Artificial intelligence milestones and the development of deep learning
    2. 1.2 Neural networks overview
      1. 1.2.1 A brief history about neural networks
      2. 1.2.2 Why neural networks are an important subject
      3. 1.2.3 Types of neural networks and their usage
      4. 1.2.4 Neural network architectures
      5. 1.2.5 Difference between a classical and deep neural networks
      6. 1.2.6 Neural networks versus classical machine learning algorithms
    3. 1.3 Deep learning frameworks
      1. 1.3.1 Most popular deep learning frameworks
      2. 1.3.2 A final word on deep learning frameworks
  5. Chapter 2. Introduction and overview of IBM PowerAI
    1. 2.1 What is IBM PowerAI
      1. 2.1.1 Contents of IBM PowerAI (IBM PowerAI Release 4)
      2. 2.1.2 Minimum hardware requirement for IBM PowerAI
    2. 2.2 Why IBM PowerAI simplifies adoption of deep learning
    3. 2.3 IBM unique capabilities
      1. 2.3.1 NVLink and NVLink 2.0
      2. 2.3.2 Power AI Distributed Deep Learning
      3. 2.3.3 Large Model Support
    4. 2.4 Extra integrations that are available for IBM PowerAI
      1. 2.4.1 IBM Data Science Experience
      2. 2.4.2 IBM PowerAI Vision (technology preview)
      3. 2.4.3 IBM Spectrum Conductor Deep Learning Impact
  6. Chapter 3. IBM PowerAI components
    1. 3.1 IBM PowerAI components
      1. 3.1.1 IBM PowerAI support and extra services from IBM
      2. 3.1.2 IBM Power Systems for deep learning
      3. 3.1.3 Linux on Power for deep learning
      4. 3.1.4 NVIDIA GPUs
      5. 3.1.5 NVIDIA components
      6. 3.1.6 NVIDIA drivers
      7. 3.1.7 IBM PowerAI deep learning package
      8. 3.1.8 Libraries
      9. 3.1.9 Frameworks
      10. 3.1.10 Other software and functions
    2. 3.2 IBM PowerAI compatibility matrix
  7. Chapter 4. Deploying IBM PowerAI
    1. 4.1 IBM PowerAI V1.4 setup guide
      1. 4.1.1 About this chapter
      2. 4.1.2 Preparing to install IBM PowerAI V1.4
      3. 4.1.3 IBM Power System S822LC for High Performance Computing initial setup
      4. 4.1.4 Installing Ubuntu 16.04.x
      5. 4.1.5 Installing IBM PowerAI V1.4
    2. 4.2 Testing IBM PowerAI V1.4
      1. 4.2.1 First test: TensorFlow test program
      2. 4.2.2 Utilization of a multilayer perceptron on a sample data set
      3. 4.2.3 Using Caffe with MNIST
      4. 4.2.4 Using Caffe with TensorFlow
    3. 4.3 Setting up IBM PowerAI V1.5.0 on a POWER L822SC for High Performance Computing server
      1. 4.3.1 Deep learning software packages
      2. 4.3.2 System setup
      3. 4.3.3 Installing the deep learning frameworks
      4. 4.3.4 Tuning recommendations
      5. 4.3.5 Getting started with machine learning and deep learning frameworks
      6. 4.3.6 Uninstalling machine learning and deep learning frameworks
    4. 4.4 IBM PowerAI V1.5.0 setup guide for POWER AC922
      1. 4.4.1 Deep learning software packages
      2. 4.4.2 System setup
      3. 4.4.3 Installing the deep learning frameworks
      4. 4.4.4 Tuning recommendations
      5. 4.4.5 Getting started with machine learning and deep learning frameworks
      6. 4.4.6 Uninstalling machine learning and deep learning frameworks
  8. Chapter 5. Working with data and creating models in IBM PowerAI
    1. 5.1 Knowing your requirements and data
    2. 5.2 Why is it so important to prepare your data
    3. 5.3 Sentiment analysis by using TensorFlow on IBM PowerAI
      1. 5.3.1 Example data set
      2. 5.3.2 How the code is structured
      3. 5.3.3 Data preparation
      4. 5.3.4 Model creation
      5. 5.3.5 Using the model
      6. 5.3.6 Running the code
    4. 5.4 Word suggestions by using long and short term memory on TensorFlow
      1. 5.4.1 Our data set
      2. 5.4.2 Overall structure of the code
      3. 5.4.3 Data preparation
      4. 5.4.4 Model creation
      5. 5.4.5 Training
      6. 5.4.6 Using the model
      7. 5.4.7 Running the code
      8. 5.4.8 Final considerations
  9. Chapter 6. Introduction to IBM Spectrum Conductor Deep Learning Impact
    1. 6.1 Definitions, acronyms, buzzwords, and abbreviations
    2. 6.2 Benefits of IBM Spectrum Conductor Deep Learning Impact
    3. 6.3 Key features of Deep Learning Impact
      1. 6.3.1 Parallel data set processing
      2. 6.3.2 Monitoring and Optimization for one training model
      3. 6.3.3 Hyperparameter optimization and search
      4. 6.3.4 IBM Fabric for distributed training
      5. 6.3.5 IBM Fabric and auto-scaling
      6. 6.3.6 DLI inference model
      7. 6.3.7 Supporting a shared multi-tenant infrastructure
    4. 6.4 DLI deployment
      1. 6.4.1 Deployment consideration
      2. 6.4.2 DLI single-node mode
      3. 6.4.3 DLI cluster without a high availability function
      4. 6.4.4 DLI cluster with a high availability function
      5. 6.4.5 Binary files installation for the high availability enabled cluster
      6. 6.4.6 A DLI cluster with a high availability function installation guide
    5. 6.5 Master node crashed when a workload is running
    6. 6.6 Introduction to DLI graphic user interface
      1. 6.6.1 Data set management
      2. 6.6.2 Model management
      3. 6.6.3 Deep learning activity monitor and debug management
    7. 6.7 Supported deep learning network and training engine in DLI
      1. 6.7.1 Deep learning network samples
      2. 6.7.2 Integrating with a customer’s network in DLI
    8. 6.8 Use case: Using a Caffe Cifar-10 network with DLI
      1. 6.8.1 Data preparation
      2. 6.8.2 Data set import
      3. 6.8.3 Model creation
      4. 6.8.4 Model training
      5. 6.8.5 Model validation
      6. 6.8.6 Model tuning
      7. 6.8.7 Model prediction
      8. 6.8.8 Training model weight file management
  10. Chapter 7. Case scenarios: Using IBM PowerAI
    1. 7.1 Use case one: Bare metal environment
      1. 7.1.1 Customer requirements
      2. 7.1.2 IBM solution
      3. 7.1.3 Benefits
    2. 7.2 Use case two: Multitenant environment
      1. 7.2.1 Customer requirements
      2. 7.2.2 IBM solution
      3. 7.2.3 Benefits
    3. 7.3 Use case three: High-performance computing environment
      1. 7.3.1 Customer requirements
      2. 7.3.2 IBM solution
      3. 7.3.3 Benefits
    4. 7.4 Conclusion
  11. Appendix A. Sentiment analysis code
    1. Sentiment analysis with TensorFlow
    2. How the code is organized
    3. Sentiment analysis code
    4. Model and training
    5. Using the model
  12. Appendix B. Problem determination tools
    1. Logs and configuration data gathering tools
    2. Troubleshooting pointers for Linux on Power
    3. Solving a RAID failure
  13. Related publications
    1. IBM Redbooks
    2. Online resources
    3. Help from IBM
  14. Back cover