Setup

Requirements

Ensure your environment meets the following requirements:

• NVIDIA Docker
• MXNet 18.12-py3 NGC container or newer
• NVIDIA-DALI 0.5.0 -- included in the MXNet container
• Python 3.5 -- included in the MXNet container
• CUDA 10 -- included in the MXNet container
• cuDNN 7.4.1 -- included in the the MXNet container
• (optional) NVIDIA Volta or Turing GPU (see section below) -- for best training performance using FP16

For more information about how to get started with NGC containers, see the following sections from the NVIDIA GPU Cloud Documentation and the Deep Learning Documentation:

• Getting Started Using NVIDIA GPU Cloud
• Accessing And Pulling From The NGC Container Registry
• Running MXNet

Training using mixed precision with Tensor Cores

Hardware requirements

Training with mixed precision on NVIDIA Tensor Cores, requires an NVIDIA Volta-based or Turing-based GPU.

Software changes

For information about how to train using mixed precision, see the Mixed Precision Training paper and Training With Mixed Precision documentation.