PyTorch Release 18.02
The NVIDIA container image of PyTorch, release 18.02, is available.
Contents of PyTorch
This container image contains the complete source of the version of PyTorch in
/opt/pytorch. It is pre-built and installed in the
pytorch-py3.6 Conda™ environment in the container image.
The container also includes the following:
- Ubuntu 16.04 including Python 3.6 environment
- 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.debcommand is the workaround for the known issue described below.
- NVIDIA CUDA® Deep Neural Network library™ (cuDNN) 7.0.5
- NCCL 2.1.2 (optimized for NVLink™ )
Release 18.02 is based on CUDA 9, which requires NVIDIA Driver release 384.xx.
Key Features and Enhancements
- Improved multi-GPU performance on image networks shown in
/opt/pytorch/examples/imagenet. You can run this example for multi-GPU by issuing the
python -m multiproc main.pycommand.
- Latest version of cuBLAS
- Ubuntu 16.04 with January 2018 updates
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.