I developed some image-processing code on an 8600, then got a new 9800 GX2 to run it on. The code runs at about 80% of what I need it to; most of its time is spent in CUFFT calls, which of course I can’t rewrite for speed.
My questions: does CUDA see the GX2 as two devices, or one? Since I haven’t modified the code to specifically use two devices, does that mean I’m only using half of my 9800 GX2?
(I’m using CUDA 1.1 on WinXP, if you’re curious; XP sees the GX2 as two graphics adapters.)