cuda sorting 800million integers travel between gpu ram and host ram

Channel:
Subscribers:
42
Published on ● Video Link: https://www.youtube.com/watch?v=0yvEQ6xs7ko



Duration: 0:48
10 views
0


gpu ram 4GB, which not enough for 800m integers. I used managedAllocation option, so the cuda would switch between host and gpu memory, which caused fully use of PCIe bandwidth. (right low corner of the video)