AI Research on NVIDIA DGX Spark: Pushing the Frontier of Local AI on a Petascale Desktop

AI Research on NVIDIA DGX Spark is the field journal of running serious AI research on a petascale desktop. Sixty-three chapters, organized into nine arcs, walk through what actually works on a Spark and what does not: foundations, inference and retrieval, training and pretraining, fine-tuning and alignment, agentic systems, observability and evaluation, deployment and distribution, products, and a Field Kit API reference.Every chapter is a hands-on session. You see hypotheses, hardware setup, runs, failed runs, post-mortems, and the next move. The book is written for engineers and researchers who want to push frontier ideas without burning cloud GPU credits. It is honest about where Spark wins on cost and iteration speed, and where a cluster still beats it.Inside, you will find a complete Field Kit API reference, build-stats callouts pulled from real runs, the first Orionfold Arena product writeup, and reproducible methodology that ports to other local-AI rigs. The narrative is a working researcher's notebook, not a marketing tour.This book is a companion to AI Native Business, the architecture and operating-model side of local-first AI. Together they cover the full shift: how to build with frontier intelligence that fits on your desk, runs on your terms, and answers to you. Read more