Step-by-step guide for configuring the NVIDIA RTX PRO 4500 Blackwell Server Edition on Red Hat AI workloads, including Red Hat OpenShift and OpenShift AI. Covers installing Node Feature Discovery and the NVIDIA GPU Operator with specific ClusterPolicy parameters (driver version 580.126.16, open kernel modules), verifying GPU detection via nvidia-smi, and deploying NVFP4-quantized models using Red Hat AI Inference Server 3.3 with vLLM. Performance benchmarks using GuideLLM show peak throughput of 3,515 tok/s for 8B models and 666 tok/s for 32B models on dual GPUs. Also demonstrates creating hardware profiles in OpenShift AI and running distributed LLM fine-tuning with Kubeflow Trainer.

15m read timeFrom developers.redhat.com
Post cover image
Table of contents
Optimized for Red Hat AIConfigure the RTX PRO 4500 Blackwell Server Edition on Red Hat AI EnterpriseRun Red Hat AI inferencePerformance validationRed Hat OpenShift AISummary and next steps

Sort: