How to use textures and arrays in CUDA?

Why use CUDA in image processing?

• CUDA as it is can offload processing of image processing steps to the GPU. While this can be a considerable boost in performance in terms of processing time, the big issue is ~ is it worth it? CPU processing of image processing apps can certainly run in real-time with average fps of 21~30. Other than offer a constant fps while processing on the GPU, this is not a real reason to use GPU which certainly leaves the usage of GPGPU in a grey area. So if we can do image processing in real-time on CPU then where does that leave GPGPU? Do we really need that extra performance boost to do what the CPU already can accomplish? Hi Justin , thanks for your answer. GPGPU is certainly not limited to image processing but I'd like to limit this question in terms of Image processing. A lot of research publications have been focusing on the performance gains that graphics processing units offer and by this alone I'm not sure what benefit it may have no matter how complex the application may be.

• Answer:

  The usage of GPGPU isn't just for image processing. GPGPU is a great candidate for anything that implements SMID (single instruction, multiple data). Perhaps your image processing projects have not been complex enough for you to see the benefit...but yes, the performance is well worth it as you add more and more and more complex shaders.