Does it suck your dick, too?
In all seriousness though, are the graphics just being offloaded from the CPU to the card? Is that what's going on there? Like you basically have a separate supercomputer that is processing your graphics. Just seems like there would be a bottleneck since the card isn't actually outputting anything externally.
Or is that an accessory to a graphics card?
it must hah for that price.
there are three kinds of workstation gfx cards, essentially: youve got mathematical computation (protein folding and the like), youve got active graphics, and youve got massive pixelcount displays (display walls)
the quadro is the king of active graphics currently, which includes CREO, autocad, anything that has any kind of geometric modelling that is updated in real time with loads of polygons and a good amount of active math. the fullscale gpu, either the hawaii chip for amd or the kepler chip for nvidia, is much better than a cpu because it has a ridiculous number of active threads that can modify the cards memory, and graphics is much more about memory swaps then it is actual computation.
amd has the best chips for display walls, 6 active displays is ridiculous, and 4 cards will drive 24 total. more ridiculous.
the tesla, however, and intel makes the xeon phi cards, are accelerators, not actual graphics cards. these cards arent meant to output anything actively, hence the lack of any video out ports, but are meant to be dedicated computational engines. when you are doing animation renders, for instance, which are almost exclusively a cpu process due to raytrace, the tesla has software that allows the cpu to transfer multithreaded tasks over to the accelerator and benefit from the architecture of a gpu when the dedicated gpu wouldnt have been able to deal with it due to lack of drivers. (this also allows you to not bog down the dedicated gpu with a x-hour render...) these are the cards that trump all else in terms of passive calculation, anything that doesnt rely on any immediate i/o. you could use it as a graphics engine as well, i suppose, but thats not really its purpose... an accelerator ends up being about effectively 10 times faster than any multithreaded gpu, give or take.
the tesla would cause a bottleneck, thats true, if it were actually calculating things for an active output to a screen. youd have a terrible latency, so the cards wouldnt work for something like gaming (even though it uses the same architecture, kepler is the same chip as the titans, 780s, etc) but since renders, for instance, only write to storage, it doesnt really make that much of a difference, and pcie doesnt slow down the transfer enough to make a difference.