Training Time delay neural networks with Parallel and GPU Computing ?

I’m trying to speed up the training of my 'timedelaynet' by using the GPU support that ÷ get from the parallel computing toolbox . Although I use the same network structure for both, when I compare the performance in case of CPU vs. GPU, CPU achieves better prtformance. and The training using GPU takes longer time than GPU. any explaination?

Answer 1

[Anshika Chaurasia]

Hi,

Performance of GPU code is dependent on the algorithm used, data size of the problem and the GPU hardware used. Significant performance gain over CPU is seen when the algorithm is computationally intensive and the data size is large enough.

Refer to the link given below to know why GPU not outperform CPU in some cases:

Why does my GPU not outperform my CPU/another GPU? Troubleshooting Steps - MATLAB Answers - MATLAB Central (mathworks.com)

Hope it helps!