NDPipe is a deep learning system designed to enhance both training and inference performance by embracing the concept of near-data processing within storage servers. At its core, NDPipe utilizes an innovative architecture that distributes storage servers equipped with cost-effective commodity GPUs across a data center. NDPipe is composed of two main elements: PipeStore (storage server equipped with a low-end GPU for near-data training and inference) and Tuner (training server that manages distributed PipeStores) This artifact appendix provides a way to emulate NDPipe, and also provides the source code and scripts for a better understanding of the evaluation in our paper. Please refer to the README file at https://github.com/dgist-datalab/NDPipe.