I have 2 9800 GX2 cards which should show up as 4 devices using DeviceQuery. I am using CUDA 2.3. When SLI is enabled 4 devices appear but CUDA apps run slower. When it is disabled only 2 devices are available and the apps run faster. Can anyone explain why the 4 devices don’t appear when SLI is disabled as I would prefer to disable it seeing as there appears to be a performance increase. My program runs on a single GPU at ~100msec with SLI disabled and this doubles when I enable it. I wish to utilise all 4 devices and not have to loose the apparent speed advantage. Any help would be greatly appreciated.