i'm just stress testing a new everyday overclock I set up earlier as we type. It's one of those "hey I got a cool idea I wonder how it will work in practice overclocks", and so far so good, got through 9mins of prime95 first try, made a few little adjustments and trying again, if im lucky it may just need a couple of changes as I bed it in over the next few days if I notice any irregularities in system behaviour or random errors in windows event viewer. first impressions is for 120mhz less cpu and 1 less multiplier, 20mhz less ram, 43mhz more fsb, and change to higher divider is that the performance is already more consistent and at least as good from some quick linpack and everest testing.

1.38v vTT set in bios ( 1.42v in readout ), 0.63x cpu gtl ref, 0.60x nb gtlref, -20mv dram controller ref, -20mv dram ch a ref, -10mv dram ch b ref, 1.3625v vcc, 1.49v vnb, 1.075v vsb, 1.58v vcpu_pll, 1.50v vsb_pll, 2.24v vdimm (2.28v in readout), common performance level of 7, with a single phase pull in cha ph1 ( pl = 6 ) and and all the others cha ph2-5 and chb ph1-5 all using pl = 7, cha clk skew of -250ps, chb clk skew of -150ps, divider 5:6 333mhz bclk, 7x cpu multi, 1139mhz ddr2 5-5-5-15-45-5-2N, 474mhz fsb, 3318mhz cpu, 102mhz pci-e, dram static read enabled, ai clock twister stronger.

with these current settings I'm finding the higher vtt along with lower vcc, vtt being roughly set inbetween vcc and vnb I'm able to get away with quite a low gtl ref multi for both cpu and nb, and from what I can tell the overall system response in terms of how fast the screen is drawn, bootup time from initializing the boot roms till vista desktop, application performance consistency, and a few other visual and response related things i use to make fine adjustments to specific bus values if needed (sometimes with diff frequencies and voltages delays between mch / pci, or cpu / mch and take a little adjustment to correct).

to say there is a starting value other than the specifications of the hardware is extremely difficult unless you are using all the other intel specified nominal voltages. it's always better to add a little more vTT and obtain the same gtl reference target voltage by means of a smaller multiplier or reducing the amount of mV to the current multiplier in place. this leaves a little more headroom to make fine adjustments if the need arises while you are bedding it in for a few days. i dont think i've ever just made major changes without going back over the next week or two and make minute changes every so often...the craziest part is I've become so used to it I just make adjustments based on a hunch and they usually sort the problem out.