AWS Deep Learning AMIs Now Provide Faster Training on Volta GPUs for TensorFlow and Microsoft Cognitive Toolkit

The AWS Deep Learning AMIs for Ubuntu and Amazon Linux now include the latest versions of TensorFlow (1.5) and Microsoft Cognitive Toolkit (2.4). These frameworks provide support for NVIDIA CUDA 9 and cuDNN 7 drivers. This allows you to take advantage of the mixed-precision training supported by the V100 Volta GPUs powering Amazon EC2 P3 instances. In our earlier tests for TensorFlow 1.5 on Volta, training the ResNet-50 benchmark with synthetic ImageNet data in FP16 mode on a p3.8xlarge instance was 1.8x times faster than training with TensorFlow 1.4.1.

Latest in deep learning framework updates

The Deep Learning AMIs provide prebuilt pip binaries for the latest official versions of deep learning frameworks in separate Conda-based virtual environments. Each framework is preconfigured with the latest version of NVIDIA CUDA that it supports.

Frameworks with CUDA 9:

• Apache MXNet 1.0 (with Gluon)
• Caffe2 0.8.1
• Microsoft Cognitive Toolkit (CNTK) 2.4
• PyTorch 0.3
• TensorFlow 1.5
• Theano 1.0
• Caffe 1.0 with CUDA 8
• Keras 1.2.2 and Keras 2.1.3

The AMIs also include model serving and debugging capabilities, which are provided by the following tools:

• Apache MXNet Model Server 0.1
• TensorFlow Serving 1.4.0
• TensorBoard 1.0.0

Getting started with the AWS Deep Learning AMIs

The latest releases of the AWS Deep Learning AMIs are available in the AWS Marketplace. Our AMI selection topic helps you pick the right AMI for your deep learning project. We’ve also provided many tutorials and developer resources to help you quickly deploy your first deep learning model on AWS.

Sumit Thakur is a Senior Product Manager for AWS Deep Learning. He works on products that make it easy for customers to get started with deep learning on cloud, with a specific focus on making it easy to use engines on Deep Learning AMI. In his spare time, he likes connecting with nature and watching sci-fi TV series.