The MOSFETs are under that heatsink with the arrow on them above, right after where the +12V CPU 4-pin power connector is situated.
Everything which comes before the CPU socket area and after the +12V 4-pin needs to be supplied, cooled and regulated well for best results.

Remember: IC's hate heat and anything which initiates or aggravates its onset.

Right now I'm trying to get stable oc's and by that I mean days without failure under dynamic stress to then result in no immediate degradation as witnessed once before, trying to find a sweet spot. For me, it's very important to establish 100% stability as much as I can after all the initial basic testing before calling it stable to anyone else... boasting isn't one of my characteristics, but honesty and genuineness to the best of my ability is, to give accurate feedback reports, as I want stability the same as a factory oc attains. I can go for 1hr Prime, mislead and get everyone in trouble trying to copy, but that doesn't fit inside the stability criteria for me as I oc systems at the 3 places I work at professionally and so I have a very stern criteria to meet: facts. Usually everything and then Linpack compiled for AMD systems to run dynamic loads at dynamic problem sizes for up-to 7 days, but this time, Linpack isn't running for some odd reason and I've not given it another try yet.

For those who can attain similar MHz as this chip I have, they can try imitating the settings I use. Saves time, a lot of work and frustration.

Lowest I could achieve fully stable so far for 5 days Prime/Memtest testing on and off, cold, warm, reboot cycling, changing MHz and changing them back again, no freezes, no overheating, no lockups or any performance oddities (SuperPi and some similar mathematical computation based software including Prime95 shows instability in one other way: the computation values be abnormally slow, adrift of proportional scaling) at 2.6G CPU, 2.4G IMC and 1066 RAM is the following ->
(click for full size image: todays testing ongoing at typical temps - idles 31-33C, ambient case 20C, CPU fan <2170/3100RPM, 215W AC load):



I have tried lower IMC VID, it failed by passing 3hrs P95 and suddenly freezing.
Second lowest CPU voltage I tried hard stability testing is 1.264V and that locked up the system after 3hrs Prime. 1.272V is the next I've tried and this is fully stable for many days testing yet but there is a lower region in between that 1.264-1.272V which maybe just as stable, I've not tried yet.

If you really want to find best MHz/V then play around with VIDs/HT ref. Different VIDs allow different voltage range combinations. I could only get 1.27V with 1.225VID CPU or lower. If I want lower volts than that (not auto) then I have to drop VID further and set voltage manually.