NVIDIA Fleet Command deploys container-based applications on GPU-accelerated Kubernetes clusters using Helm charts. This guide will provide the information necessary to build applications compatible with Fleet Command and includes information on setting up a development environment.
Containerization involves bundling an application with all of its related configuration files, libraries, and dependencies required to run efficiently for portability. While containers are an excellent way to bundle and run applications, it is essential to manage the containers that run the applications and ensure no downtime in a deployment environment.
Kubernetes is an open-source platform for managing containerized applications. For example, if a container goes down, another container needs to start. Kubernetes helps in maintaining container life cycles without any human intervention.
Fleet Command leverages these technologies to provide a fully supported cloud-native platform that securely deploys, manages, and scales your applications across a distributed edge infrastructure.
Applications deployed on NVIDIA Fleet Command must run on the following software stack:
Ubuntu 22.04.3 LTS
CUDA 12.3.0
NVIDIA Driver 535.129.03 (Pre-compiled signed driver)
Containerd 1.7.3
Kubernetes 1.27.8
NVIDIA Container Toolkit 1.13.5
Helm 3.12.3
NVIDIA MIG Manager 0.5.1
NVIDIA GPU Operator 23.9.0
NVIDIA GPU Operator includes the NVIDIA GPU Driver libraries. For information about NVIDIA GPU Operator, refer to the release notes.
Instructions on setting up this software stack manually for development purposes are provided later in this guide, in the Development Environment section.