More GPU's or more nodes?

[WSurvey]

Hello,

Cost aside, if one wanted to decrease processing times for aligning photos and building dense clouds, or any time consuming operations, would one go about getting more licences to run network processing or start throwing more graphics cards at the problem? Assuming each extra computer has useless (integrated) graphics but good core i7 processor.

Also second question. I'm processing on an i7-3770 with 16Gb and no graphics card (just motherboard integrated GPU) and that's the best I've got in my office. Am I missing out on much speed by not having a good, single, stand-alone graphics card like GTX 970?

Cheers.

[Thibaud Capra]

Having a standalone graphics cards would drastically reduce the computing time, except for photos alignment that is only processed on the CPU, unless you're running PhotoScan 1.3.x

Of course, you'll need to use OpenCL to make it useful and relevant! (As far as I know the CUDA usage is now optimized further more on PhotoScan for nVidia graphics cards)

TL; DR: GPU helps a lot.