amd a6-5400k (stock is 3.6ghz@1.35v, turbocore is off)
msi fm2-a55m-e33
4gb kingston 1600mhz 1.35v
intel g2 80gb
antec ar350 350w 80plus / picopsu-80 + delta 84w brick (no efficiency rating)

amd athlon ii x3 440 (3ghz@1.12v, 800mhz@0.768v)
gigabyte ga-ma785gm-us2h
3gb ddr2 667mhz (4 sticks)
intel g2 80gb
asus hd5450 1gb ddr3
antec ar350 350w 80plus

measurements are made a the wall and with a kill-a-watt ez clone.
....................................x3..........a6+ar350........a6+ar350.........a6+ar350..........a6+pico
..............................3ghz@1.12v.......stock..........@2.8ghz........3.6@1.156v........@2.8ghz
....................................................................................................................
idle..............................50w............28w.....................................................23w
occt cpu........................100w...........64w.....................................................46w
occt linpack....................109w...........67w.............55w...............58w...............47w
occt gpu.........................67w...........88w.................................64w................61w
occt psu (cpu+gpu)............115w...........98w.............81w...............85w..............crash!
ssf4 bench.......................93w...........81w.............67w...............71w................69w
pcsx2.............................87w............................................................................
dolphin..........................89w.............................................................................

a6 system off power is 3w with either psu, and for some reason peaks at 70w in the bios.
from memory the x3 system without the gpu idles at 42w(ar350) and 34w(pico).

the pico might have to go, 5w at idle is a smaller difference than i expected, the brick i'm using though doesn't even have an efficiency rating so a level v one would definitely help. emulation is too slow at 2.8ghz and i don't think the pico combo can handle it running undervolted and at 3.6ghz.
i'm not sure why the pico combo can't seem to do more than ~70w at the wall, but either the brick is crap or it's related to the picopsu-80s low 12v output rating.
i'm also thinking about undervolting the nb(if possible) and hoping it links to the gpu, cause as you can see it uses a big chunk of the overall power!

(If you put the table inside code tags you can use a mono-spaced font for the table. I took the liberty to put your values in a table formatted this way.)
Thanks for the values! It's interesting to see that a reduction from 1.35V to 1.156V only¹ reduces max power by 13%. Going with 2.8 GHz (which is close to the 2.7 GHz of the 35W A6-4400M which is also a two-module, 192 GPU units APU) changes max power by less than 5%. With a difference of 53W between max load and idle, this is might put the underclocked APU in a hypothetical 50W TDP range². Compared to the A6-4400M this might mean that the GPU frequency difference (496-685 MHz vs 760 MHz for the 5400k) might account for up to 15W.

I ordered an A10-5800K and an Asus mainboard for my family's computer and I'll do some similar testing over Christmas. And then I'm back to waiting for a decent FM2 mini-ITX board which supports undervolting...

¹ I had hoped for more when reducing the VCore by 14%. Isn't power supposed to scale with the square of the voltage?
² I assume 80% efficiency for the PSU. This gives us 98W*80% = 78W real load power consumption at stock frequency and voltage. Substracting the 65W TDP we get 13W used by other components. 81W at 80% efficiency equal 65W. If we substract the 13W for other components we get about 50W for the APU alone when at 2.8 GHz.

thanks for the code tip!

the nb voltage is adjustable but it seems it isn't linked to the gpu.
i did notice the gpu voltage changing though and after checking it i found it to have 3 power states: 304mhz@0.912v, 608mhz@1.199v and 760mhz@1.274v. from the numbers i could tell that the gpu uses a significant amount of power so i jumped into the bios and underclocked it to 608mhz, what happened after is the interesting part.
for some reason the gpu no longer changes states and stays locked at 608mhz, but more importantly it remains at the lowest voltage of 0.912v!
some results (pico):

it seems stable enough, no stressing done yet, too lazy.

it sucks that bios undervolting seems to be uncommon among fm2 motherboards and i sincerely hope that it won't become the trend.
according to some undervolting done by another user on the xbmc forums, an offset of about -0.2v would have worked out pretty well for this processor.

Thanks a lot for these numbers! I think they make clear that undervolting the GPU is definitely worth it. A reduction of 10W AC for the 192 graphics units version in the gpu test might translate to 20W AC for the 384 units APUs. (i.e. a max power reduction of about 16W using my naive efficiency calculation).

I think that if I hope to put an undervolted A10-5800K in a M350 and use it for gaming, I need to be able to undervolt the GPU, too. I guess GPU voltage might also be a point where the A10-5700 differs from the A10-5800K to reach the 65W TDP.