I'm no expert on this - by no means, so somebody please correct me if im wrong. From what I understand, the more cores - the better the computer can handle multi-tasking. Some servers run virtual desktops for a whole lot of thin clients, this means that all software for a whole bunch of people doing all kinds of stuff, runs on the server - the "thin clients" only show the desktop screen and sends keystrokes and mouse-movement/-clicks etc. to the server where everything is processed. I guess in these cases the multiprocessors with a bunch of cores come in handy. Some software, like Photoscan, can take advantage of all available cores. Although - I think Alexey Pasumansky wrote somwhere here that at some stage in the mesh triangulation (87% into build geometry?) only one core is used exclusively.
The GPU is better suited for the calculation types in most of the steps on the way from pics to textured model, while some stages has to be processed on the CPU. Sure it will run faster on an extremely fast CPU, but overall the stages where the CPU does the work represents a small percentage of the whole workflow in Photoscan. Thats why you should spend your money on some good GPUs, instead of an expensive CPU - if Photoscan is your main reason to invest.
If I bring up the perfomance monitor while working in Photoscan, I can see that all cores are working hard most of the time. Now, if there where more of them I guess they would be working just as hard at the same speed as the others. Most of this workload is shared with the GPU, but the majority of calculations, while processing in Photoscan, seems to fit the GPU-processors better. I understand it as if the GPU is better at smaller, quick math-pieces, while the CPU has to deal with the hard stuff.
Again - somebody, please correct me if I'm wrong. I know I have a lot to learn
Added: Heres what was said in that other thread:
Hello Andy,
At 87% of mesh generation phase mesh is being triangulated at this step really doesn't use all the cores. It may take quite a long time depending on the dense point cloud quality (and number of points in the cloud).