https://docs.google.com/viewer?a=v&...VlbmluZ3RoZWJlYXN0fGd4OjdhNTI3NzQ0MTJjYzY3MDE
See the paper. I don't necessarily agree with the paper, but it is kind of obvious that gaming PCs are a terrible waste of power if you are using it for anything other than gaming. Most people would be better off with a cheap power efficient PC (even something like Skylake T which is hardly slow along with a power efficient mobo and DC power?) and then use the gaming PC for just games. Then again, if you only use your $1000+ gaming PC for just games, isn't that basically a really expensive console?
If you did that, then you've got two PCs running. How's that an advantage? You're using as much extra power doing that as just wasting it through the idle GPU. The added crap will make up a lot of the difference (such as a KVM), which is going to be a single light bulb's worth anyway, unless the comparison is to a NUC or similar.
The best solution would be Optimus on the desktop, for the dGPU problem.
I use my integrated Xeon E3-1245V3 graphics because even in internet browsing, email, and stupid stuff a descrete graphics card is adding an additional 100w or so more.
Not with Maxwell, and an ad-blocker. I don't break 100W without either synthetic testing, compiling, or gaming, much less 100W just from the GPU. My old GTX 460? Er...it sort had idle and current hog modes only, despite being similar in TDP . W/o blocking ads, I can absolutely believe their web browsing numbers in the paper, though. How people manage like that, I do not know.
GPU-limited gaming tends to get me to around 300W total (80+ Platinum PSU, B85M Pro4, Xeon E3-1230V3, GTX 970, U2414H). I have little doubt that my motherboard is the weakest link in idle power consumption (it's a run of the mill 3+1 phase VRM setup, IIRC). Highest I recall was just over 320W, though usually being closer to 280W.
I think they were excessive with their Fig 14 improvements, in that an i5 and GTX 970/980, with a decent mobo and PSU, and any monitor normal people buy, would have gotten them 95% of the way to using a Pentium, without the minimum FPS sacrifices (which they do not document).
That said, I do agree with their conclusion,
which is that non-techies don't have access to solid information on the subject of any given component's energy efficiency.