GTC Silicon Valley-2019 ID:S9339:Profiling Deep Learning Networks
Poonam Chitale(NVIDIA),David Zier(NVIDIA)
Learn how to use NVIDIA tools to understand GPU utilization in a deep learning model. Traditionally, profiling is program optimization, including tasks like removing unnecessary code and addressing gaps and wait times. Profiling for deep learning networks is an end-to-end workflow from data input to inference. It involves understanding the data pipeline, the neural network structure and post processing.