Skip to content

AI Infra Setup

Component-by-component guides for building a large-scale GPU training environment. It covers the full sequence from host setup through the scheduler, containers, storage, and validation.

  • DLAMI Host Setup & Validation

    Deep Learning AMI-based GPU host setup and driver, EFA, and GDS validation.

    DLAMI Host Setup & Validation

  • Slurm + Pyxis + Enroot

    Scheduler and container runtime setup, plus job submission options.

    Slurm + Pyxis + Enroot

  • Container Image (NGC + EFA)

    Building a training container by adding EFA to the NGC image.

    Container Image

  • GDS (GPUDirect Storage)

    Connect FSx for Lustre directly to GPU memory (GDS over EFA) to accelerate data loading.

    GDS

  • Distributed Training Validation

    Measure a go/no-go performance baseline before full training.

    Distributed Training Validation

  • EFA Networking Deep Dive

    Inter-node communication principles and NCCL plugin validation.

    EFA Networking Deep Dive