Simplifying HPC Workflows with NVIDIA NGC Container Environment Modules

Originally published at:

Many system administrators use environment modules to manage software deployments. The advantages of environment modules are that they allow you to load and unload software configurations dynamically in a clean fashion, providing end users with the best experience when it comes to customizing a specific configuration for each application. However, robustly supporting HPC and deep…


I tried to set the NGC_IMAGE_DIR variable to path that had all .sif file from NGC container. However, when I run singularity with just file name it try to look at .sif file in home directory or working directory. Do I need to explicitly append the $NGC_IMAGE_DIR to sif file ?



Hi Vijay,

The NGC container environment modules expect the SIF files in $NGC_IMAGE_DIR to have specific names.

You didn’t mention a particular NGC container. For instance, the tensorflow/20.06-tf2-py3 module expects the SIF file to be named nvcr.io_nvidia_tensorflow:20.06-tf2-py3.sif. (These are the file names you would get if you used to NGC Container Replicator tool to download the container image files.)

Alternatively, you can customize the environment modules for your site to use a different SIF file name. The relevant line, again for the TensorFlow container, is
local image = "nvcr.io_nvidia_tensorflow:20.06-tf2-py3.sif"