Parallel Neural Network Training using Multiple GPUs

Photo by Harrison Broadbent on Unsplash

Deep learning is quickly becoming the most powerful and ubiqutous tool within machine learning, performing well in a vast array of applications1. However, for problems requiring many hidden layers for accurate calculations, the training of these neural networks can quickly become very computationally expensive. In this project, I learned how to significantly speed-up neutral network training by using CUDA to perform calculations on a GPU and MPI to perform these calculations in parallel across multiple Tesla K80 devices.


  1. https://thenextweb.com/artificial-intelligence/2020/01/02/2010-2019-the-rise-of-deep-learning/ ↩︎

Jeremy Binagia
Jeremy Binagia
Applied Scientist

Related