when using GPU's from the Quadro family, setup of the displays will be done using NVIDIA's Mosaic control panel. No need to change your application. Same for AMD EyeFinity GPU's.
From a software point of view, I don't see why you'd be forced to use CUDA in this case. While compute shaders would allow you to use fancy techniques like this, there is no need to use such a technique just because you're running on multiple high resolution displays. In fact, I've found that rendering directly to the main buffer, instead of to an offscreen
Fbo, is often faster in a multi-display setup. Just don't use the convenience methods (like
gl::drawSolidRect(), etc.) but always use your own
Vbo's and reduce the number of draw calls as much as you can by using instancing. This will help maintaining a high frame rate (even in single-display setups). Also try to reduce overdraw, make sure to use culling where applicable, etc. You'll find that Quadro cards have plenty of power and bandwidth to run your application in high resolution.