Prerequisites#
This deployment guide provides the technical framework for implementing the Red Hat AI Factory with NVIDIA, a co-engineered solution that unifies NVIDIA AI Enterprise with Red Hat OpenShift AI. It is designed for platform engineers and MLOps teams and walks the reader through the full stack deployment, including hardware and software prerequisites, installation and verification of core components like the Node Feature Discovery, NVIDIA GPU, NVIDIA Network, and NVIDIA NIM Operators, and final integration with Red Hat OpenShift AI for streamlined model serving, experimentation, and production MLOps workflows.
Hardware Requirements#
The Red Hat AI Factory with NVIDIA requires a synchronized set of hardware and software prerequisites to ensure a performance-optimized, enterprise-grade AI environment. This solution is designed for NVIDIA-Certified Systems running Red Hat OpenShift to provide a consistent experience from data center to edge.
Note
Refer to the NVIDIA RTX Pro AI Factory Reference Architecture for an example of certified hardware.
To leverage the full acceleration of NVIDIA AI Enterprise in a Red Hat AI Factory, the underlying hardware must meet the following specifications:
Compute#
NVIDIA AI Enterprise minimum hardware requirement including the use
For production workloads, it’s recommended to have a minimum of
3 control plane machines (or master) dedicated to the control plane for high availability.
2 compute node (or worker) machines dedicated to running AI workloads
For NIM workloads, refer to the specific NIM support matrix for guidance on model specific hardware planning
For other model runtimes, refer to the Red Hat support article
Networking#
NVIDIA Spectrum X Ethernet or NVIDIA Quantum Infiniband Networking
Storage#
A supported storage class that supports dynamic provisioning
NVIDIA recommends NVIDIA Certified Storage solutions
Several components of OpenShift AI require or can use S3-compatible object storage, such as AWS S3, MinIO, Ceph, or IBM Cloud Storage