GTL Ref Tweaking Guide

[negev]

Hrmm, I think my chip just doesn't want to do 3.5ghz, either that or gtls need adjusting without core#1 being at a -5 offset.

vcore 1.41 (1.4 real)
vtt 1.35
PLL 1.58
SB 1.57
NB 1.67
NB step +3
CPU GTL 10,5,10,10

1:3 divider to take memory out of the equation.

At 438x8 (3.5ghz) I get a rounding or calc error in prime after about 2 mins every time.

I have tried:

vcore up to 1.43v
vtt 1.45
PL 1.62
NB step +4

All with no difference. if NB is less than 1.67 the machine bluescreens, but at 1.67 I get calc errors which makes me think that the NB is now stable but the cpu needs tweaking.

I have tried adjusting GTLs:

9,4,9,9
11,6,11,11
12,7,12,12
13,8,13,13
14,9,14,14
15,10,15,15

I think this is as far as I went, initially these changes made no difference but the last two caused the machine to hardlock almost instantly when running prime.

I am wondering if higher gtl settings are needed, but that I might need to also adjust something else to get it stable. In the past hardlocks have nearly always been related to vNB, i'm wondering if increases the cpu gtls like this means I also need to adjust NB GTL or lower vNB.

Failing that, what is a good methodology for adjusting individual GTLs? I'd rather have some kind of method than just random changes...

[cadaveca]

do you really need 1.35 vtt negev?

[RAYTTK]

what is a good methodology for adjusting individual GTLs? I'd rather have some kind of method than just random changes...

My point Exactly. Hopefully saaya will come through with some solid Quadcore results

[negev]

@cadaveca

1.35 with this bios is only 1.25 real

@RAYTTK

fingers crossed! I'm actually curious to know if GTL refs only affect a specific core. I.e. if GTL0 is wrong, would that always cause errors in core #0 only or could it cause errors in the other cores as well? If the GTL for an individual core makes a difference only to that core then we could use affinity in task manager and fine-tune them one by one.

[cadaveca]

GTL0 and 2 should be the same core, 1 and 3 the other.

And negev, I understand that, however I have been finding that with quite a few 45nm chips(about 40% of the 40 or so I've played with) that increase in VTT merely creates instability, with C0, C1 and E0 quads and duals. Alot of times I am finding that memory is the limiting factor, so with you with 1700mhz 2x2, I'd say drop the divider and try for more.

Not to say that the memory isn't capable of more, but I'm running @ 450FSB with 4x1gb(1800mhz), no VTT changes. NB volts definately much lower than what you are using...

But the other chips, the remaining 60%...seem to require MORE than what you are giving..as you say, it's only 1.25v...so I don't understand how you came to use that value...

GTL tweaking adjsuts the "window" in which the signal is read. As you adjust VTT, vcpu, and vNB you change this window...maybe you've made the window to big...don
t forget that GTL cna be tweaked DOWN as well...

[negev]

GTL0 and 2 should be the same core, 1 and 3 the other.

But I have four cores... what about the other two?

And negev, I understand that, however I have been finding that with quite a few 45nm chips(about 40% of the 40 or so I've played with) that increase in VTT merely creates instability, with C0, C1 and E0 quads and duals. Alot of times I am finding that memory is the limiting factor, so with you with 1700mhz 2x2, I'd say drop the divider and try for more.

You missed this in my post: "1:3 divider to take memory out of the equation."

Not to say that the memory isn't capable of more, but I'm running @ 450FSB with 4x1gb(1800mhz), no VTT changes. NB volts definately much lower than what you are using...

Interesting, what board are you using? If it's the blackops can you post all your settings please?

But the other chips, the remaining 60%...seem to require MORE than what you are giving..as you say, it's only 1.25v...so I don't understand how you came to use that value...

Several other people used this for their setting, Xello got up to about 4ghz i think using 1.35 in bios (1.25 real).

GTL tweaking adjsuts the "window" in which the signal is read. As you adjust VTT, vcpu, and vNB you change this window...maybe you've made the window to big...dont forget that GTL cna be tweaked DOWN as well...

Yeah I tried one notch down on each ref.. maybe i should go down further

[RyderOCZ]

But I have four cores... what about the other two?

You are controlling all 4 cores (0,1,2,3) Core 0 and 2 are on the same die, Core 1 and 3 are on the other.. therefore you only have 2 GTL refs to change. 2 cores run on each GTL ref.

Hope that helps.

[negev]

So GTL 0,1 is for Die1 which is also Core 0,1 and GTL 2,3 is for Die2 which is also Core 2,3

correct?

[cadaveca]

Negev,

vcore 1.325
vtt 1.2
PLL 1.509
SB 1.50
NB 1.50

but this is with QX9650/Q9650, and maybe because q9450 has less cache, maybe FSB of cache is neutered on your chip.

10,5,10,10 cpu GTL has never worked for me. Sure, it can create a bootable situation, but never stability. But I always keep VTT low...

you should not need to tweak GTL or VTT, IMHO, until over 1.3625v cpu. each step above is going to require miniscule adjustments, and this can be offset by adjsuting other voltages to fall in line with cpu volts.

Also, more than 1.6v NB is going to cause issues. You should NOT need that much @ 425FSB...more like 1.4v...I can even run 450fsb dual, 4x1, @ 1.375v bios.

GTL 0 and 1 should be different dies in quad, AFAIK. when I say core, I mean piece of silicon, not cpu.

[negev]

And thats on a blackops?

[negev]

no way are those settings on a blackops, i can't even post at 435fsb using that (with all GTLS at 0).

and 425fsb i need 1.53v, 1.5v is unstable. for 438 i need 1.67

[RAYTTK]

I'm actually curious to know if GTL refs only affect a specific core. I.e. if GTL0 is wrong, would that always cause errors in core #0 only or could it cause errors in the other cores as well?

I went down this road a few times myself, If core 1 failed prime i went into bios & changed the value up & tried again but it just turned into another waste of time.
I see we are getting some VIP members joining this thread maybe one of these xtreme Guru's will guide the blind.
Im now running 10 x 375 with no vtt increase & GTL's on default 2 hrs prime stable

[cadaveca]

no way are those settings on a blackops, i can't even post at 435fsb using that (with all GTLS at 0).

and 425fsb i need 1.53v, 1.5v is unstable. for 438 i need 1.67

Yep, blackops. On air too, with fan blowing away from nb block. Maybe you just got a bad cpu...becuase I do not feel I have a cherry board or nothing...but you'd have to ask Saaya about that one.

[negev]

Well my settings at 425 aren't too dissimilar from xello's and he has a QX9650.

I'm pretty sure the voltages im running at the moment, my cpu needs in order to be stable.

I tried vtt at 1.2 (bios) - prime locked instantly
vtt at 1.5 (bios) - no difference, prime errors within 2mins

[cadaveca]

OK, so you upped VTT lot, and is slightly more stable, which means that GTL tweaks must be the way for you. Adjusting VTT changes GTL window, but I think becuase you have NB volts so high, alot of GTL tweak is nessecary, and hard to do when you have both NB and cpu GTL to worry about.

Your chip IS NOT qx9650, and my QX9650 works with lower cpu volts than listed above, so pls don't make that comparison. Xello's QX9650 isn't an exceptional sample or anything, it's run of the mill.

Personally, I think you should try default VTT and play with GTLs...find your max stable @ default gtl, and then go a few MHZ above and adjust GTL to suit. I mean, if things were great for you, we wouldn't be having this conversation, so maybe a new approach is a good idea?

[negev]

Changing VTT from 1.35 to 1.42 (in bios) made no difference whatsoever. Lowering it made the system lock almost instantly when running prime.

I don't think 425x8 on a q9450 is that bad, I just want to push it higher and it seems like GTL tweaks are whats needed.

Max stable at default GTL was about 400x8 I think, after that i have to change them and 10,5,10,10 is what a lot of people use and it works at this fsb.

[Ryan12]

I know we are talking more so about quad cores here... But i did find on my E8400 that no matter what i set the NB at whilst trying to overclock it at around (9x450), that the NB only would like to be at 1.56, AND CPU PLL would only like 1.547, anything else at this clock speed just failed prime immediately...

But that aside, i just found that going higher and higher with volts didnt work at all (except cpu).... But i have no experience with quads.

Also negev i have a mate which has a Q9450 and he has a xfx 790i and he claims he runs his at 3.8ghz on water cooling... Ill see what his settings are and see what he might have done differently

[ol'norton]

There is so much of this I don't understand, but I'll start with this

Core1 GTL Ref is higher than the rest of the Cores by default, and is linked to NB Gtl!

How are they linked? When raising one do you need to raise the other, in which case, why have seperate controls.

I have all the voltages on default & are stable @ 9X 440 FSB Memory @ 1760 7-7-7. with all GTL ref voltages on default.
At 445fsb I pass 3 or 4 out of 5 passes on Linpack.
If I find the "magic" GTL settings for my E8500, it should stabilise 445 FSB & maybe higher. Is that right?

[negev]

Also negev i have a mate which has a Q9450 and he has a xfx 790i and he claims he runs his at 3.8ghz on water cooling... Ill see what his settings are and see what he might have done differently

1. Create new motherboard with $$$ pricetag and 200,000 BIOS options to confuse everyone and disguise the fact that it's closely rivalled by an alarm clock

2. ?????????

3. PROFIT!!


Kidding, I love my BO really

[Ryan12]

Mmmm my mate overclocked this E8400 on his xfx 790i to 4.8 under 1.5 volts easy.... now i cant overclock it to even 4.5 even with 1.6... Such a joke...
Degrading chip you ask? I would have thought that if it hadn't of come straight from his motherboard to my blackops from just sitting at 4.8

[negev]

If anyone is in south east london and willing to let me try my Q9450 in a different board I'd be very interested and even willing to pay for the time involved.

Sooo many times I've heard people say they couldn't get the results they wanted with this board using parts that clock like mad on other boards.... I don't have another board to use as a frame of reference so I can't really give an opinion of how true this is.

[Xello]

Xello's QX9650 isn't an exceptional sample or anything, it's run of the mill.

Correction, it sucks needs 1.45v and not a single bit less for 4ghz (100% stable) Still i use EIST so not too worried about it dying any time soon..

By the way i need 1.58v NB for 421fsb, 1.67v for 450fsb, to be 100% stable. Funny thing is the inbetween value from what i remember (i posted about it somewhere) actually makes it unstable. I firmly believe that there are tons of chips like Q9450's out there that will run higher clocks than my hugely expensive QX chip with alot less volts But i'm happy with 4ghz anyways.

[negev]

Xello, have you tried your cpu on another board? Would be very interested to see settings for 4ghz on something else..

*cough*CLASSACTION*cough*

[Xello]

Xello, have you tried your cpu on another board? Would be very interested to see settings for 4ghz on something else..

*cough*CLASSACTION*cough*

Nah, i'm not generally into doing this more than once every 3-4 years, i wish i had the time to make it a proper hobby

[ol'norton]

Yeah, my QX9650 needed 1.67 NB to get 450FSB stable, but giving it 1.7 (NB watercooled) didn't help me get any higher.
The same chip on my P5Q3 Deluxe managed 490FSB stable with 1.70 NB. (NB on air)