HowTo for building 4 to 8-GPU systems?

I’m interested in building a 4 to 8-GPU system with server-grade (ECC) CPU and RAM, and consumer-level (e.g. GTX 780) GPUs – the results of one device can be checked on another, however if the logic of the whole process gets messed up, that’s not good.

Are there any up-to-date recipes for building such systems: which components to use, etc.? I’m on a budget, so I’d like to avoid needless mistakes when building the hardware.

I’m looking to do about the same thing. I’ll keep looking through the forums and some blogs. I’ll check back here with any info.

I’d like a system that works much the same way as the Amazon GPU cluster. My intention is to configure it for myself but also host in the same way they do.