My application performance is mainly affected by what I’m doing in my shaders. These shaders really do a very large amount of work (maths, access the whole GPU memory and so on).
I use low resolutions most of the time (800x600 for instance).
Does GLSL scale up well with SLI in this situation?
If I were to use transform feedback for instance, disabling rasterization, would I see a good increase in performance with SLI (2X?), assuming I am processing a huge amount of vertex data ?
It’d be worth detailing a bit more about what you’re doing in your shader. The “access the whole GPU memory” thing especially. Generally speaking, GPUs are not so good at this if you’re spending a large portion of your time waiting on random memory reads to come back. If so, worth considering if you can refactor your problem to increase locality of reference.