Gayan,
Thanks for the reply. I thought that address translation should be taking care of the 1D 8192 limit, is that not true?
Also, I don't get allocation problems for the 16*16*9 case only for the 128*128*9 case, BUT I also don't get problems for a 32*32*9 case (which > 8192) but I do for a 64*64*9 case.
I was hoping one of the AMD guys could shed some light.
So if you are using scatter at all then 8192 is the biggest limit you can use? That seems very limiting.