GTC 2020: An Overview of GPU Recurrent Neural Network Performance

GTC 2020 S21301
Presenters: Jeremy Appleyard,NVIDIA
We’ll focus on performance of recurrent neural networks (RNNs) at the sub-framework level. There are several distinct methods one can use to implement an RNN, each with advantages and disadvantages, and the right choice will depend on both the target GPU and the network hyperparameters. We’ll cover these methods in detail and give approximate hyper-parameter ranges for each method.

Watch this session
Join in the conversation below.