The original image:

After the negation:

The results:

As you can see from Assignment 1, it took the serial program 230 milliseconds to negate this image of equal size. Now, seeing as this is mighty GTX 960, we thought we would not iterate through the process once, but 1000 times! That would mean that the serial implementation would take approximately 230 seconds to complete. So how fast did the GPU do it in?

We saw the GPU smash 1000 iterations in 22 milliseconds. That's over 10,000 times faster! Clearly, image processing begs to be worked on by parallel processors. The massive throughput of the 1024 cuda cores, which can operate on thousands of pixels at the same time reduces the time, really beating the CPU without much of a sweat. Here is the NSIGHT performance analysis:

In closing, we're interested in optimizing our image processor to handle 2D arrays without serialization. That way, we can extent our GPU power to the rest of the methods, without worrying about columns and rows within a 1D array. All in all, we learned a lot and look forward to applying our knowledge further in the coming weeks!