2D Gausian Convolution algorithm is implemented that works on very large images. Gausian filter is often used for image down-sampling. Implementation is robust and seperable. That means, the two convolution can be seperated into two 1D convolutions. Breaking a single multi dimensional Gausian convolution into two 1D convolutions significantly improved the performance. Computation time of 2D convolution is O(nk^2) where n is the number of pixels in the image and k is the size of gausian kernel. Reducing this to two 1D convolutions yield a computation time of O(2nk). Thus provided a dramatic performance improvement. The algorithm is implemented on CPU and GPU using CUDA C. Benefit of parallel execution for this algorithm is monitored profiling the convolution on GPU verses CPU.