NVIDIA GPU Infrastructure Fundamentals

A Practical Map of NVIDIA’s AI Infrastructure Stack

NVIDIA GPU Infrastructure Fundamentals is a structured guide to the hardware, software, and operational practices that turn individual GPUs into dependable AI platforms. Vivian Aranha writes for people who need to understand both the accelerator and the environment around it: the CUDA model, the libraries, the cluster network, the storage path, and the tools that keep production inference running.

What the Book Covers

The book is organised into three parts. It begins with foundations of accelerated AI infrastructure, explaining how AI, machine learning, and deep learning workloads map to parallel hardware. From there it introduces CUDA, NVIDIA’s software stack, and the architecture choices behind A100, H100, L40S, and B200 GPUs.

Part two shifts to building and operating GPU infrastructure. Topics include GPU sharing with MIG and vGPU, monitoring with DCGM, scheduling with Kubernetes and Slurm, high-throughput storage and network paths, GPUDirect Storage, RDMA, InfiniBand, and DPU offload with BlueField and DOCA.

Part three looks at production AI platforms and exam readiness. It follows the AI lifecycle through MLOps tools such as Airflow, MLflow, and Kubeflow, then covers inference serving with Triton, ONNX, and TensorRT, and delivery with NVIDIA NGC and Kubernetes.

Who Will Find It Useful 🧠

The material is aimed at infrastructure engineers, AI platform teams, cloud and data centre professionals, and technical learners who need a coherent view of NVIDIA’s accelerated computing ecosystem. It is not a collection of disconnected product notes; the chapters build a working model of how compute, memory, interconnect, storage, and software interact in AI deployments.

Why This Subject Matters ⚙️

AI systems are no longer just models. They are fleets of accelerators, networks, storage tiers, schedulers, and observability tools. Understanding the full path from CUDA kernel to production inference helps teams make better decisions about capacity, performance, reliability, and cost. This book treats that path as one system.

Inside the Digital Edition

This ebook edition from Digital Delights delivers the complete text in a convenient digital format. It is a useful reference for anyone studying NVIDIA-based AI infrastructure or working with GPU-accelerated platforms in production.