Comment 11 for bug 627779

The corruption goes away over time which points to a caching or buffering problem. Also if you cat /dev/urandom to the screen while doing the test then the corruption is more readily seen as occurring in equal sized vertical lines of pixels. I have not counted them but they seem about 32 pixels long. I did a dd of 64 byte of zero to the screen and got a black line about the same length. Since the fb is 2 bytes deep this makes sense.