NVIDIA at TensorFlow World 2019

Learn how NVIDIA GPUs and TensorFlow are helping developers dramatically accelerate their AI-based applications.

NVIDIA logo
NVIDIA
•
Nefi Alarcon
•4 min read•intermediate•
--
•View Original
BERTDeep LearningKerasTensorFlow

Overview

The article discusses NVIDIA's participation at TensorFlow World 2019, highlighting advancements in TensorFlow 2.0, TensorRT, and automatic mixed precision for enhancing AI application performance on NVIDIA GPUs. It outlines key sessions and hands-on tutorials aimed at developers to accelerate their AI-based applications.

What You'll Learn

1

How to eliminate input/output bottlenecks in deep learning applications using DALI

2

How to enable mixed precision training to improve performance on Volta GPUs

3

How to optimize TensorFlow applications using TensorRT with Keras APIs

4

How to deploy trained models using INT8 quantization in TensorRT

5

Why running TensorFlow at scale on GPU clusters is beneficial for large AI applications

Key Questions Answered

What are the benefits of using TensorRT with TensorFlow 2.0?
TensorRT is tightly integrated with TensorFlow 2.0, providing high performance for deep learning inference through a simple API. It allows developers to optimize applications easily using Keras APIs, enhancing performance on NVIDIA GPUs.
How can automatic mixed precision improve training performance?
Automatic mixed precision can provide up to 3x training performance improvement on Volta GPUs by allowing models to use lower precision calculations without sacrificing accuracy, thus speeding up the training process significantly.
What is the significance of the Jasper model in speech recognition?
The Jasper model is an end-to-end convolutional neural acoustic model that achieves state-of-the-art word error rates on the LibriSpeech dataset, simplifying the process of building competitive automatic speech recognition systems.
What techniques are covered for scaling TensorFlow on GPU clusters?
The article discusses experiences running TensorFlow at scale on GPU clusters like the DGX SuperPod and Summit supercomputer, focusing on design considerations and practical applications such as BERT and AI+HPC.

Key Statistics & Figures

Performance improvement with automatic mixed precision
up to 3x
This improvement is specifically noted for training performance on Volta GPUs.

Technologies & Tools

Some links below are affiliate links. We may earn a commission if you make a purchase.

Framework
Tensorflow
Used for building and training machine learning models.
Framework
Tensorrt
Used for optimizing deep learning inference performance.
Tool
Dali
Used to accelerate data loading and processing in deep learning applications.

Key Actionable Insights

1
Implementing DALI can significantly enhance the efficiency of data pipelines in deep learning applications.
By eliminating I/O bottlenecks, developers can ensure that their models receive data faster, which is crucial for maintaining high training speeds and overall performance.
2
Utilizing automatic mixed precision can drastically reduce training times while maintaining model accuracy.
This technique is particularly effective on Volta GPUs, making it a valuable strategy for developers looking to optimize their training processes.
3
Leveraging TensorRT for inference can lead to substantial performance gains in TensorFlow applications.
By integrating TensorRT with TensorFlow 2.0, developers can optimize their applications more effectively, ensuring they make the most of their GPU resources.

Common Pitfalls

1
Neglecting to optimize data pipelines can lead to significant slowdowns in training.
Developers often overlook the importance of I/O efficiency, which can bottleneck the entire training process, especially in large-scale applications.

Related Concepts

Deep Learning Optimization Techniques
Automatic Mixed Precision
GPU Performance Tuning
Speech Recognition Technologies