This is a demo to compare the computation ability of single thread CPU, multi thread CPU and GPU.
The task is to extract the an image's local high gradient point. A local pixel block has size 7 * 7. We'll find the point that has the maximum image gradient of each 7 * 7 pixel block. An image with size 739 * 458 is provided

OpenCV > 3.0
CUDA > 7

git clone https://github.com/zhaozhongch/GPU_vs_CPU.git
mkdir build
cd build
cmake ..
make

Single CPU example

cd build
./find_gradient

Multi CPU example

cd build
./find_gradient_multi_thread

GPU example

cd build
./find_gradient_cuda

They'll all get similar image output like Those circles just show the where the local high gradient point is in every 7 by 7 pixel block.
On my ThinkPad-X1-Extreme laptop with CPU Intel(R) Core(TM) i7-8850H CPU @ 2.60GHz and GPU GTX 1050, the single CPU computation time is about 100ms, multi CPU computation time is about 50ms, GPU computation time is about 400us.