How to set up GTL Ref Values for 45nm & 65nm

[Cibic]

Thanks gentlemen. Very insightful. I'm actually thinking "leave well enough alone" where I'm at, which is basically my sig settings below.

vCore =1.356
pLL = 1.590
vTT = 1.378
vNB = 1.56
vdram=1.90 (kingston 2x2 hyper X 1600MHz @1350MHz 7-7-7-21

GTL ref. +50/+50/+50/+50 with NB ref. +40

LLC=enabled

These settings pass an Intel Burn @max load for 10 rounds with the cores reaching upper 60's Celsius.

My current 3DMarkVantage:



Maybe I need to offset cores 1 and 3 from 0 and 2.

My max stable on this build was 460x9 but I wasn't comfortable with the high voltages and upper 70's Celsius core temps with IBT.
But that test is rigid and doesn't apply to gaming. I truly expected a better bench and 3D score on H2O. I gave up and settled on a
24/7 set-up.

Thanks guys! -keith

You certainly don't need an offset. IBT is good but you can pass even unstable. There's nothing like prime95 blend for at least 4h. If that passes and for utmost stability try prime95 blend among with GPU Caps Viewer> Fur Rendering at 1920x1200. If it crashes try setting NB gtl on auto and raise NB a notch. You need to run that for atleast an hour

[keithallenlaw]

You certainly don't need an offset. IBT is good but you can pass even unstable. There's nothing like prime95 blend for at least 4h. If that passes and for utmost stability try prime95 blend among with GPU Caps Viewer> Fur Rendering at 1920x1200. If it crashes try setting NB gtl on auto and raise NB. You need to run that for atleast an hour

OK, maybe its time I try this prime 95 test. I lead myself to believe
from others that IBT was the most stringent.

Thanks Cibic! -keith

[keithallenlaw]

One more note. I get tearing when I do the 3DMark test with the performance test because my monitor is
a 2560x1600 and most of the rendering test are at a much lower rez. I believe 3DMark test are 1280x1024.
Thats not even near a 16:10 ratio. You say Prime @ 1920x1200? At least that is 16:10 and the same as my
ratio. I read that trying to render at a different aspect ratio than your native actually lowers your score. Dunno...

[Cibic]

One more note. I get tearing when I do the 3DMark test with the performance test because my monitor is
a 2560x1600 and most of the rendering test are at a much lower rez. I believe 3DMark test are 1280x1024.
Thats not even near a 16:10 ratio. You say Prime @ 1920x1200? At least that is 16:10 and the same as my
ratio. I read that trying to render at a different aspect ratio than your native actually lowers your score. Dunno...

Prime doesn't stress graphics card, i ment run GPU Caps Viewer, "Fur Rendering" at your max resolution supported by vga and monitor along with prime95 blend.
Note that your cpu, ram must be stable to proceed to this step.

[keithallenlaw]

Prime doesn't stress graphics card, i ment run GPU Caps Viewer, "Fur Rendering" at your max resolution supported by vga and monitor along with prime95 blend.
Note that your cpu, ram must be stable to proceed to this step.

Got it. Thanks!

[gugle]

CryptiK, with the Maximus II Formula and P5Q-Deluxe
you used to do the same GTL over on your e8400?

[OnDborder]

How would you tune in a dual core? Are the cores 0/2, 1/3, 1/2, 0/1 on a e8600 running a vtt of 1.35?

[Cibic]

-_-

[OnDborder]

From the reading, I would have thought it would be a -45, being a 45nm. I came up with 45 too.
Would that be 45 across all 4 settings or the sum of 45 on 2 settings? Like -35/-10/-35/-10 etc..
I'll try prime blend also.

[radaja]

does any one know how the gtls on the dfi=ut-p35-t2r operate?is it percentages or mv?

[Cibic]

From the reading, I would have thought it would be a -45, being a 45nm. I came up with 45 too.
Would that be 45 across all 4 settings or the sum of 45 on 2 settings? Like -35/-10/-35/-10 etc..
I'll try prime blend also.

-_-

[OnDborder]

Depending on my fsb voltage, I'm having luck with -40's or -45's straight across. And +50 on the nb.

[Cibic]

Motherboards know best

[CryptiK]

CryptiK, with the Maximus II Formula and P5Q-Deluxe
you used to do the same GTL over on your e8400?

Only did relatively brief stability testing on the P5Q-Deluxe as I just bench on it and run the M2F 24/7, but I was around the same Vref on both boards.

I run +60/+20/+60/+20 on the M2F and 0.680x 0/2 & 1/3 on the P5Q-Deluxe, both with a Vtt of 1.25v giving a vref of 0.85v on both boards.

does any one know how the gtls on the dfi=ut-p35-t2r operate?is it percentages or mv?

No idea, theres a table somewhere you can DL that explains them IIRC. I'm pretty sure mikeyakame DL'd it for his DFI although its an X48 maybe give him a pm and see.

I think that's a myth along with offsets. But you can try -45 -45 -45 -45, and if u get load drop then try +45 +45 +45 +45. I'm eager to know your results too.

Wafer thickness in nm is not what to set your GTL Refs to. 65nm and 45nm do behave differently, but whatever works best for you.

[Cibic]

Motherboard makers were waiting for you to tell them how tune GTL refs

[CoY_82]

newbie here in xs, but not in oc'ing.. =) i have been reading this thread from page 1 to eternity and i tried to dig in to the equations and computations to get the applicable gtl ref for my system... problem is, i've tried using the formula .8 / vtt/fsbterm value = gtl ref from 1.3vtt - 1.4vtt with the vnb set @ 1.4 (to rule out nb getting choke coz of hi fsb) and the cpu pll @ 1.52

tested stable setup (see sig) is 450x9 @ 1.26vcore, 470x9 1.28vcore, 500x8 1.28vcore. tried 533x7.5 but i cant get it stable even if i set the vnb to 1.42, i know the p5q-d is capable of running 1600fsb so i think it doesnt need too much vnb to go around 2000fsb... when i try to add around 10fsb to either one of the stable settings posted, it wont stable, even if i bump the cpu pll to 1.58, vnb to 1.44, vcore to 1.35 (bios) 1.32 (cpu-z)

settings are pl=10 no pullin enabled, 333 strap 1:1, pcie clock @ 101. as far as i can see those settings should be good anough for, let's say 4.2~4.3ghz... but i cant... only the 470x9 1.28vcore is stable but when i go to 480x9, my system goes quack quack (as occt says lol!) everytime i run occt 2hr medium set..

any inputs on where i should start checking for the right gtl? any inputs also on my settings?

math hater here, my nose bled so much while i was trying to understand the other formulas that's why i ended using the easiest one, the .8 / vtt/fsbterm = gtlref

help... =(

[CoY_82]

i was able to make 510x8 stable... i used the gtl calculator from seban (nice work dude!), set my vtt/fsbterm to 1.3 and set the gtls to 635/675, however, the vnb that was computed was 1.46 and i set the nb gtl to .620 following .9 / 1.46 = .620... temp rised on the nb, have no home for my system right now lol! nb temps reached 45 max temp during 2 hours occt custom medium test

still working on the 533x7.5... anyone tried putting a fan on their p5q-d's nb? like a 40mm fan? or how's like antec spotcool for my nb?

thanks for all the guys who shared their knowledge about gtls and stuff... even though my nose bled non stop while i was reading some of the "more mathematical" approach... lol! thanks! will post screenies once i got some stable settings around 520fsb++...

hey cryptik, could you post your settings for your e8400 e0? i'll try to check out some settings that can boost my e8400 e0 =) i know this baby can do more than just 4ghz =)

[GrossmeisterB]

Hello everyone!

I'm trying to get my Q6600 (G0) stable, and I'm working on the GTLs now!
I had an E8400 before, and it was quite "easy" to tune it with the GTLs, because I changed both 0/2 and 1/3 in the same way.
Now I will have to change them seperatly, right???
I d/loaded Seban's calculator, but how does it help me?
I have an Asus P5Q-E, so I enter the GTL-REF in "0.670"-values...
So, how do I do this now?? The BIOS say "0.630" is default vor 0/2 and "0.670" is default for 1/3...so, where to start now?? I'm aiming for 3800-4000MHz, so a FSB of about 445MHz max. if possible...

Thanks in advance for your help!!

[Demo]

Is GPU caps viewer safe to run at extended times, since Furmark will burn your card.. more likely the VRMs..

[Cibic]

I really doubt that, but anyhow software that torture vga's accomplish what is needed within a hour IMO.

[mikeyakame]

does any one know how the gtls on the dfi=ut-p35-t2r operate?is it percentages or mv?

Sorry for replying so late.

DFI GTLs aren't either. They are mapped values against Vtt voltage points vs 160 steps from roughly 0.523xVtt to 0.83xVtt.

each value between 0-160 is roughly equivalent to 0.0015v - 0.0020v.

For example 1.06v Vtt GTL table:

CPU GTL 1/2, 0 = 0.556v, 160 = 0.816v
CPU GTL 0/3, 0 = 0.545v, 160 = 0.843v
NB GTL, 0 = 0.586v, 160 = 0.861v

1.50v Vtt GTL Table:

CPU GTL 1/2, 0 = 0.783v, 160 = 1.046v
CPU GTL 0/3, 0 = 0.773v, 160 = 1.069v
NB GTL, 0 = 0.824v, 160 = 1.102v


To correctly calculate them and roughly extrapolate between 2 Vtt voltage maps, you'll need to print the Vtt mapping tables out that DFI have hidden away on their website and have a calculator handy for figuring out the rough voltage u need for your Vtt, then find its numerical equivalent in the tables, and if Vtt is inbetween 2 points just roughly calculate what inbetween value you need.

I have the VTT GTL tables printed out and stored in a folder so I can refer to them when making changes. You have so much precision in GTL voltage adjustment it's mind blowing. It's worth spending the little bit of extra time to calculate what values you should need based on guesstimation.

I find I can run really low Vtt voltage if I spend the extra time, and I haven't had a GTL related Prime95 or Linpack failure that I couldn't completely eliminate yet. As long as Vtt is set reasonably with respect to FSB frequency and CPU you can get it stable with fine adjustments to GTLs, that no other board will ever do.

http://dfics.dfi.com.tw/dfi_cs/LTX48/X38X48_vtt.xls

Here you go. Have fun with it

[radaja]

thanks for the info mikey!

[gugle]

here it difficult to stabilize my Q6600 at 3.6ghz with the P5Q-E

Bios - P5Q-E - Q6600 VID 1.2500v

CPU Voltage - 1.41250v
CPU GTL 0/2 - 0.630
CPU GTL 1/3 - 0.630
CPU PLL - 1.54v
vFSB - 1.30v
vRam - 2.00v
vNB - 1.26v
NB GTL - 0.630
vSB - 1.20v
vPCI-E - 1.60v

CPU Clock Skew - 200 ps
NB Clock Skew - 200 ps

vcore CPU-Z
Min - 1.392v
Max - 1.416v


WC = Block GTZ + Pump DP-1200 +
Rad. Swiftech 320

amb. +/- 31ºC
water idle 33.2ºC
water full 35.3ºC

Real Temp TJ Max = 100ºC
Min - 43/45/38/37 ºC
Max - 63/65/61/61 ºC

[mikeyakame]

here it difficult to stabilize my Q6600 at 3.6ghz with the P5Q-E

Bios - P5Q-E - Q6600 VID 1.2500v

CPU Voltage - 1.41250v
CPU GTL 0/2 - 0.630
CPU GTL 1/3 - 0.630
CPU PLL - 1.54v
vFSB - 1.30v
vRam - 2.00v
vNB - 1.26v
NB GTL - 0.630
vSB - 1.20v
vPCI-E - 1.60v

CPU Clock Skew - 200 ps
NB Clock Skew - 200 ps

vcore CPU-Z
Min - 1.392v
Max - 1.416v


WC = Block GTZ + Pump DP-1200 +
Rad. Swiftech 320

amb. +/- 31ºC
water idle 33.2ºC
water full 35.3ºC

Real Temp TJ Max = 100ºC
Min - 43/45/38/37 ºC
Max - 63/65/61/61 ºC

Start with around 1.45v CPU and 1.22-1.28v Vtt and CPU0/3 Gtl around 0.67x, CPU1/2 GTL around 0.64-0.65x. NB GTL around 0.67x

[jas420221]

Can someone explain with the GTL refs for the cores (.63 and .67 were are stock on P5Q) are different when its stated GTLref is .67 of FSB?