TensorFlow Release 18.02

The NVIDIA container image of TensorFlow, release 18.02, is available.

TensorFlow container image version 18.02 is based on TensorFlow 1.4.0.

Contents of TensorFlow

This container image contains the complete source of the version of NVIDIA TensorFlow in /opt/tensorflow. It is pre-built and installed as a system Python module.

To achieve optimum TensorFlow performance, for image based training, the container includes a sample script that demonstrates efficient training of convolutional neural networks (CNNs). The sample script may need to be modified to fit your application. The container also includes the following:

• Ubuntu 16.04
  Note:

  Container image 18.02-py2 contains Python 2.7; 18.02-py3 contains Python 3.5.

• NVIDIA CUDA 9.0.176 including:
  • CUDA^® Basic Linear Algebra Subroutines library™ (cuBLAS) 9.0.282 Patch 2 which is installed by default
  • cuBLAS 9.0.234 Patch 1 as a debian file. Installing Patch 1 by issuing the dpkg -i /opt/cuda-cublas-9-0_9.0.234-1_amd64.deb command is the workaround for the known issue described below.
• NVIDIA CUDA^® Deep Neural Network library™ (cuDNN) 7.0.5
• NVIDIA^® Collective Communications Library ™ (NCCL) 2.1.2 (optimized for NVLink™ )
• Horovod™ 0.11.2

Driver Requirements

Release 18.02 is based on CUDA 9, which requires NVIDIA Driver release 384.xx.

Key Features and Enhancements

This TensorFlow release includes the following key features and enhancements.

• Updated OpenSeq2Seq example to include latest bug fixes
• Latest version of cuBLAS
• Ubuntu 16.04 with January 2018 updates

Known Issues

• cuBLAS 9.0.282 regresses RNN seq2seq FP16 performance for a small subset of input sizes. This issue should be fixed in the next update. As a workaround, install cuBLAS 9.0.234 Patch 1 by issuing the dpkg -i /opt/cuda-cublas-9-0_9.0.234-1_amd64.deb command.
• The broadcast and reduce (but not all_reduce) functions in the tf.contrib.nccl module cause an error when executed as part of a graph. This issue should be fixed in the next update. The multi-GPU training example script nvidia-examples/cnn/nvcnn.py includes a workaround for the nccl.broadcast function so that the script still runs correctly.
  Note:

  The Horovod example script nvidia-examples/cnn/nvcnn_hvd.py is not affected by this issue.

• Some Python 3 codes may encounter errors when handling text strings containing non-Latin characters. This can be fixed by setting an environment variable with the following command:
  Copy

  Copied!

              
  
              
  $ export LC_ALL=C.UTF-8
          

  This issue should be fixed in the next update.