Bubble sort was not able to be sped speed up as fast as quick/merge/heap sort O(n * logn) but compared to serial bubble sort we ran on matrix & average pc, there was very noticeable difference. Our overall approach felt a lot like the reduction lab where we called the kernel multiple times to get the answer. We also realized that the optimal solution will be different on different hardware platforms.While doing research, we found out there are even better parallel sorting algorithms (radix, rank sort, parallel merge sort, hyper quick sort). We actually found the CUDA sample code for parallel merge sort (It was for older version of CUDA). When we profiled it and ran it, it turned out to be much slower (1 – 3 seconds) than serial merge sort for N = 500,000. However, the CUDA quick sort given was just as fast as the serial quick sort (potentially faster by few milliseconds).