Do mobile to desktop cpu's scale the same way as mobile to desktop gpu's?
Yes, and no. The question you're asking is rather vague, so I'll try to cover all of the bases.
CPUs and GPUs scale the same way in the sense that they scale with clock speed, and scale through architecture.
Where their scaling differs is with the code that they run. GPUs run what is called "embarrassingly parallel" code, where you can have a very large number of "cores" and the scaling will be nearly 1:1 with the number of cores.
CPUs on the other hand often run code that can only run on a single thread. In this instance, more cores wouldn't scale performance up at all. Well, not exactly, since you'll essentially always have more than one thread running at a time, but it's going to be close enough to zero that adding more cores would be about the least efficient thing you could do in the search of boosting performance. Another caveat is that cores are generally paired with additional shared cache, so you may have a core sitting there not being utilized, but because it came with some extra L3, you might have a workload that could benefit from that extra cache.
I looked at mobile gpus and noticed if mobile gpus were used in desktops there would be better performance using less energy and producing less heat.
This is sort of true. High end GPUs typically aren't too off in efficiency from lower end SKUs from the same family and same platform (e.g. desktop GPU vs. desktop GPU), so using a "lower power" GPU won't help you in this case. On the other hand, mobile GPUs are going to be lower leakage parts, that were tested before packaging and deemed to use less power. They may also be clocked lower, with a higher priority placed on power or efficiency.
Is this the same for cpu's?
With all else equal? It's the same as the GPU scenario I've outlined. They're just better binned.
Now they'll also have different clock speeds as well, and today's mobile CPUs have more aggressive dynamic frequency and voltage scaling.
I've read some on cpu's and I can't fully tell if this is the same.
Examples: gpu-2x mobile faster than 1x desktop
GTX 780 tdp 250(desktop)
3977 gflops
-
2x gtx 780m tdp 2x100=200(mobile)
5056 gflops
The primary difference here is with different clock speeds and binning, as I outlined above. Also, in this case, the 780 is going to have a lot of additional circuity devoted to double precision floating point operations, whereas the 780m will be much more focused towards single precision FLOPS. The 780 would perform much more efficiently, and also much faster, than the 780m when it comes to DP-FLOPS.
Is it the same if you used 2 laptop cpu's in a desktop?
i7-4870HQ tdp 47--- 9526-passmark(mobile)
i7-4790- tdp 84---10223-passmark(desktop)
2x i7-4870HQ tdp 47x2=94?)passmark 9526x2?=19052?
So does this mean that 2 mobile cpu's if down clocked? could outperform 1 desktop cpu at the same energy/heat/other?
A disclaimer: I have absolutely zero faith in Passmark. Their scores are mostly useless.
In this case, yes, two HQs would outperform the 4970. But there are some things to keep in mind -- there is no support for two separate CPUs for either the HQ model or the 4790. Multiple sockets is a feature only allotted to the Haswell-E platform. So you'd need two discrete i7-4870HQ systems.
This is only the case because Passmark is a multithreaded benchmark. Also, in reality, 2xi7-4870HQ would be quite a bit away from "19052," and would probably be closer to ~17000, or perhaps lower. Most multithreaded CPU code isn't going to scale 1:1; it'd be more like 80%.
Would multiple intel atoms be better than 1 laptop/desktop?
performance/price/other?
Definitely no on price. Atom is in a rough place right now, and it is much more expensive for its performance than something like a Haswell Celeron. Intel dropped the ball on Atom's price efficiency for the current generation.
Performance per watt would be dependent on the code you are running, but the answer is "probably," so long as the code you're running is multithreaded. If it's not, well, throwing more Atoms at the problem would do you no good at all.
Also it seems 2x mobile gpu is faster than 1x desktop. would multi tegra k1's in a laptop be better
all around for performance/heat/power consumption than 1-2 mobile gpus?
For multithreaded code? Uhh... probably more efficient, yes. But definitely more expensive.
Not all multithreaded code is created equal, though. Performance is a fairly complicated equation. You have branches, loops, adds, divides, vectors, scalars... there are so many different instructions, so many different types of computations... GPUs aren't fit to handle many of these things -- they're very special purpose. What they can do well, they do much better than a CPU. What they can't, well, it's the opposite. If performance were just a matter of FLOPS, we'd all be using GPUs.
Also consider that programming for a GPU is much more difficult, fewer people know how to do it, and those people are often paid much more. CUDA Jesus will make 300K (or something like that), whereas C++ Average Joe is going to make < $100K.