NVIDIA/deepops

Tools for building GPU clusters

What it solves

DeepOps simplifies the complex process of deploying and managing GPU-accelerated server clusters. It automates the setup of the necessary software stack—including drivers, container runtimes, and cluster managers—so that researchers and engineers can focus on running AI workloads rather than manual infrastructure configuration.

How it works

DeepOps uses Ansible playbooks to orchestrate the installation and configuration of GPU environments. Depending on the user's needs, it can:

  • Deploy full Kubernetes clusters using Kubespray for containerized application management.
  • Set up Slurm clusters for traditional batch job scheduling and resource management.
  • Install essential NVIDIA components (drivers, Docker, and NVIDIA Container Runtime) on single machines.
  • Deploy the NVIDIA DGX software stack on supported systems.

Who it’s for

It is designed for system administrators and ML engineers who are managing on-prem data centers, NVIDIA DGX systems, or any bare-metal environment requiring high-performance GPU clusters for AI training and inference.

Highlights

  • Flexible Deployment: Supports full cluster orchestration or modular installation of specific components like KubeFlow or NFS storage.
  • Multi-Scheduler Support: Provides automated paths for both Kubernetes (container-centric) and Slurm (job-centric) environments.
  • DGX Optimization: Specifically tailored for NVIDIA DGX systems and DGX OS.
  • Broad OS Compatibility: Validated for Ubuntu 22.04/24.04 LTS and NVIDIA DGX OS 6/7.

Related

  • Project
    dell/omnia

    An open-source deployment toolkit using Ansible to automate the provisioning of servers, schedulers like Slurm and Kubernetes, and AI frameworks for compute clusters.

  • Project
    NVIDIA/gpu-operator

    A Kubernetes operator that automates the management and provisioning of NVIDIA GPU software components, including drivers and runtimes, to simplify GPU node scaling.

  • Project
    dstackai/dstack

    A unified control plane for GPU provisioning and orchestration that streamlines AI development, training, and inference across any GPU cloud, Kubernetes, or on-prem cluster.

  • Project
    gpustack/gpustack

    An open-source GPU cluster manager for AI model serving and instance provisioning that orchestrates inference engines across hybrid environments.

  • Project
    nebius/soperator

    A Kubernetes operator that automates the deployment and management of Slurm clusters for AI training and HPC, providing high reliability and efficient GPU utilization.