blender internal CUDA?

Does the internal blender render engine uses Open GL CUDA or does it all works with the CPU? is there a way to switch in between these two?

I’m building a new computer to render a very heavy animation, I would like to know if I should buy a new video board or maybe should I aim in other components or characteristics? can someone point me out to the best characteristics of a rendering computer, or if anyone know a web page that talks about this can you send me the links… thank you

BI use CPU only, Cycles render use GPU for rendering if properly configured and on selected hardware (modern NVIDIA, > 550 series for example).
Edit:
It big question, many ppl working on that and still do not know. Render algorithms very complex and different, some setup can give 10+ increase for selected scene, but you change mesh a little and performance go way down (bad branch prediction, inefficient cache size overflow/cache line misalignment).

Most robust solution are multi core CPU with big caches, as Xeon /Opteron, very expensive. More common just middle- high good CPU with enough RAM.

GPU based setup still exotic, moving target, you can get 50+ times (no kidding) increase from single average GPU or get worse then CPU performance.

If you have lot of money, go Xeon/Opteron server hardware, if risky try GPU. Otherwise, usual common setup is equally good for rendering or other tasks.

I would gladly change all my materials to cycles, but there are just too many… is there atool to batch change them all or an easier way?

Not sure it exist, material system too different, i saw some thread related to that with script that help to quickly set some default values, but i fear you must recreate all them from scratch.

Minchia… that’s “shit” in sicilian… thanx for the help …

what do yo think if use two i7 3930k 3.2ghz 6 core

That will be very cool, especially if ECC RAM can be used (not sure it can support ECC RAM), you get guaranteed fast render of any scene, animation data, smoke/particle simulation, not only pure pixel render accelerration.

I forgot another solution is modern Cloud services like Amazon EC2, never use it myself but it can be good for example for big one-time final render, you not need to setup hardware, it have some fault tolerant feature. You still need good workstation for interactive modeling of course.

You do know that EEC stands for Extended Error Correction and is a parity feature and not a feature that increases performance?

@arexma
Sure, it increase final performance reducing you need to re-render 4k/8k frames with artefacts because of RAM fault, at least single bit flips (most common). Other solution is render more times same frame and compare (majoritarian redundancy?) but with ECC you get almost same with no time penalty.