Highly Parallel Implementation of NeuroJet using Graphics Processing Units

This paper presents the parallel implementation of NeuroJet, a neural network simulation program, using GPUs. To help researchers study different correlation among applications, programming models, and underlying hardware architectures, NeuroJet is examined as a case study to provide insights and findings specific to GPU programming. We demonstrate the importance of programmer effort through our parallel implementation of NeuroJet. Our study also shows that GPUs provide substantial performance improvement up to 500,000X speedup, albeit in expense of readability and maintainability of the code, which is not desirable in certain cases.