Is really just going to bottle neck? (assuming 4 16x pcie cards.)
I really want to build a compact sim/render farm taking advantage of the cuda cores using red shift or octane (probably Octane since v4.0 is out and looks to be significant imporvements).
Eh, I think you need to find out whether your apps need PCIe bandwidth for the GPUs. This cluster works with mining rigs because they don't care about PCIe bandwidth, but maybe rendering and simulation does.