ProxyTSPRD Profiles

Dataset Image

download Download


ProxyTSPRD profiles are collected using NVIDIA Nsight Systems version 2020.3.2.6-87e152c and capture computational patterns from training deep learning-based time-series proxy-applications on four different levels: models (Long short-term Memory and Convolutional Neural Network), DL frameworks (Tensorflow, PyTorch), datatypes (FP64, mixed-precision), and single-node dense GPU platforms (Nvidia™ DGX A100 and DGX-2 V100). The model uses spatio-temporal data from a dynamical system for model training and inference. The dataset is divided into two parts: (1) lightweight CSV files generated using `nsys stats -f csv`, and (2) raw QDREP files as generated using `nsys profile --kill=none -t cuda,osrt,cudnn,cublas`.