Epicness937 :
the 800 series is the same architecture of the 700 series so makeing mobile and desktop 800s of different architectures would become very confusing to so minimise confusion nvidia skipped the 800 desktop gpus
This is incorrect.
The low-power 800 series (800m, 820m) used the Fermi architecture from the 500/600 series.
The higher-power 800 series (860m, 870m, 880m) used the Kepler architecture from the 600/700 series.
The mid-power 800 series (830m, 840m, 850m, 860m) used the Maxwell architecture, and were the first nvidia GPUs to use Maxwell
Yes the 860m is on the list twice. There was a Maxwell version and a Kepler version.
Maxwell was delayed due to third party fab factorries' inability to get 16nm processes working. Intel successfully managed to get 3D FinFETs working thus allowing them to shrink down to 14nm, but other fabs had trouble creating the intricate 3D structure. Samsung was the first non-Intel fab to manage it, but nvidia uses TSMC and Global Foundries. And most of their early 16nm production was taken up by Apple's processor for their iPhone/iPad (the fabs gave Apple priority because hundreds of millions of units ordered vs tens of millions of units ordered).
http://www.computerhope.com/jargon/f/finfet.htm
Nvidia eventually resorted to producing Maxwell on 28nm, which resulted in a lot more heat than it was originally designed for. 1.87 billion (Maxwell) vs 1.3 billion (Kepler) transistors, and 148mm^2 vs 118mm^2. So they initially released Maxwell only as mobile GPUs, which ran at lower voltage and clock speeds and thus generated less heat. After they further refined Maxwell to run on 28nm, they released it as the 900 series for the desktop.
The fabs have now got 16nm down, and Pascal GPUs (nvidia 10x0) are being made on 16nm.