CPU-GPU Speed Comparison

Performance Evaluation of CPU-GPU and CPU-only Algorithms for Detecting Defective Tablets through Morphological Imaging Techniques

Role: MS student, Sole Developer.
Tools and skills used: NVIDIA GPUs, CUDA C, C++, Visual Studio.


This project is the enhancement of the work presented in “Implementation of SCADA System for Unsought Tablets Detection through Morphological Image Processing, Proceedings of the 12th IEEE International Multi-topic Conference, ISBN: 978-1-4244-2823-6, pp. 493-500”.

The purpose of this project is to speed-up the system process via implementing the image processing part of above referenced project on GPU and CPU both. A 44.94x speed improvement in CPU-GPU algorithm as compare to CPU-only version was determined. Then GPU algorithm was further optimized to reduce global memory traffic and a 9.76x additional improvement in speed of processing the image was observed. Overall GPU-CPU algorithm runs 54.9x faster than a CPU-only algorithm.

Hasan Baig
Hasan Baig
Assistant Professor