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

[mikeyakame]

practically speaking the benefits are being able to dial in fsb frequencies that you normally wouldn't be able to as a result of them needing gtl reference voltages outside of nominal values, and stabilizing high fsb frequencies, usually 450mhz and above. anywhere below this, and all you'll be doing adjusting the gtlref is slightly reducing the vcc/vnb/vtt voltages. not usually worth the effort unless you have a reason for changing them. only area below 450fsb I'd suggest using gtl ref votlages explicitely is around the 425-440mhz fsb range as they are very vital for proper stability.

[ahmad]

practically speaking the benefits are being able to dial in fsb frequencies that you normally wouldn't be able to as a result of them needing gtl reference voltages outside of nominal values, and stabilizing high fsb frequencies, usually 450mhz and above. anywhere below this, and all you'll be doing adjusting the gtlref is slightly reducing the vcc/vnb/vtt voltages. not usually worth the effort unless you have a reason for changing them. only area below 450fsb I'd suggest using gtl ref votlages explicitely is around the 425-440mhz fsb range as they are very vital for proper stability.

I am trying to run at 540MHz+ FSB, I can get it in windows and do everything, but it won't pass 300%+ of memtest.

So are you telling me I could get it super stable if I tweak them, or I could just dump more vfsb and vnb and solve my problems that way. Cause if its just going to give me 400% or 500% stability in memtest, then I am not going to waste my time..

[mikeyakame]

a combination of all the above is your best bet.

[Amurtigress]

Hello,

I did some tweaking last weekend and the result is rather interesting.

My ASUS P5Q3 Deluxe board and a Q9300 C1 processor was previously set to the following, 7 hours Orthos stable:

FSB: 460
GTL ref 0/2: 0.69x
GTL ref 1/3: 0.67x
NB GTL ref: AUTO

At this old configuration, the CPU GTL references have been extremely sensitive to what value I was setting. Give or take 0.02 and I got Orthos errors pretty quickly.

Then I tried setting the NB GTL reference manually since I realized it's importance through this thread.

New config:

FSB: 460
CPU GTL ref. 0/2: 0.63x
CPU GTL ref. 1/3: 0.63x
NB GTL ref: 0.68x

The new thing is that the manual GTL setting allows for a FAR wider range of CPU GTL references. 0.69/0.67 still work fine but I can easily go down to 0.63 now while previously 0.665 was failing.

I haven't tried just how far I can go, mostly because I feel a lot more comfortable with the once again decreased sentitivity of my system, which already goes beyond what I had on the Maximus Formula board.

One observation with NB GTL on AUTO was that like, as soon as you would increase or decrease one of the CPU GTL references by 0.01, the respective other die would start erroring out as compared to the previous attempt. overall very random looking. It is also interesting that with NB GTL @ AUTO the 0/2 ref. neededto be higher than the 0/2 reference, which is very much against the BIOS' default which is 0.63 for 0/2 and 0.67 for 1/3.

Also, the default value of 0.63x for the NB GTL doesn't work stable here, you have to go to about 0.68.

All of this unfortunately doesn't help with the board's FSB wall at around 470. Neither does the needed CPU core voltage decrease through te changed NB GTL ref.

^.^Maybe that info is useful!

I hope to replace my Q9300 C1 with a 9550 E0 next week. More testing coming up then!



PS: Currently trying to reduce the FSB term. voltage. There is not much space tho, if I am staying above the core voltage.

[xarot]

Hi,

I made a little CPU GTL ref calculator, based on the formula on the front page. Hopefully at least someone finds it useful. Please note that I'm NOT a programmer, I can only do basic programs like this

Over here: http://pumppu.info/~dakotha/gtlref/gtl.zip

[R104_for]

Ty

[CryptiK]

@ xarot - nice work, that may assist the mathematically challenged users work it out. You should not that is the correction factor for the 0.667x GTL, as you leave the 0.635x GTL as is, and only subtract ~40mv from the 0.667x GTL.

Also it should be noted that is only a starting point, and the correct GTL Ref setting really depends on the FSB you are aiming for and the vFSB (aka Vtt) you are using. For example when I use 1.16 vFSB, I have my 0.635x GTL at +50mv and the 0.667x GTL at +10mv, but if I use 1.2 vFSB, I have my 0.635x GTL at -70mv and my 0.667x GTL at -30mv.

I am trying to run at 540MHz+ FSB, I can get it in windows and do everything, but it won't pass 300%+ of memtest.

So are you telling me I could get it super stable if I tweak them, or I could just dump more vfsb and vnb and solve my problems that way. Cause if its just going to give me 400% or 500% stability in memtest, then I am not going to waste my time..

No one can tell you exactly how many % coverage you will pass in memtest. You have to try it and see. However, if you can pass Orthos small FFT and Lipnpack, which are primarily cpu stressors, but fail memtest at ~300%, I'd say you need to increase vdimm slightly, perhaps vNB slightly too. Try that before you go GTL tweaking.

Windows memtest is actually very hard on the ram and NB and I actually needed 2 notches more vdimm to pass it, although I was Linpack & Orthos Large & Small FFT stable.

[mskvorc1]

After reading all this, please help me a little. I'm currently running mine E8500 @ 3825 MHz @ 450 MHz FSB. Mine CPU VTT is 1.20V in BIOS (1.18V Everest reports and mine vCore is 1.302V idle & load (made pencil mod). My nForce SPP is 1.4V. I've been able to hit 4037 MHz with vCore of 1.376V and VTT @ 1.25V and nForce SPP @ 1.45V. If I want to reach 4 GHz, but with lower vCore how should I start with GTLVref values. I can set four of them in mV, but don't know at what GTLVref values to start, I mean with what lanes and at how many mV (positive or negative).

Please, help me.

[Amurtigress]

Hi,

I made a little CPU GTL ref calculator, based on the formula on the front page. Hopefully at least someone finds it useful. Please note that I'm NOT a programmer, I can only do basic programs like this

Over here: http://pumppu.info/~dakotha/gtlref/gtl.zip

Hello,

I just tried your program. However, my server 2003 x64 Windows says it's not written for this machine type.

This means your program must be containing 16 bit code...hence it won't run on any 64 bit windows. Maybe you could look into this and see if your compiler can generate all 32 bits code. Many overclockers prefer 64 bit OSses, even if it is just for supporting more than 4 GB of RAM.

Other than that, this tool could be very useful! Thanks for posting

[HawaiianSupermn]

Anybody successfully run 500fsb on the Evga 790i FTW analog board with a 9650? I've been beating the gtlrefs silly but all I can muster is 3 minutes in windows and it suddenly restarts. The Ultra board did it easy breezy, no probs.

[xarot]

@ xarot - nice work, that may assist the mathematically challenged users work it out. You should not that is the correction factor for the 0.667x GTL, as you leave the 0.635x GTL as is, and only subtract ~40mv from the 0.667x GTL.

Also it should be noted that is only a starting point, and the correct GTL Ref setting really depends on the FSB you are aiming for and the vFSB (aka Vtt) you are using. For example when I use 1.16 vFSB, I have my 0.635x GTL at +50mv and the 0.667x GTL at +10mv, but if I use 1.2 vFSB, I have my 0.635x GTL at -70mv and my 0.667x GTL at -30mv.

Yes I know. My program was mainly for those who have difficulties calculating the values. You can do the very same things with a pencil and a piece of paper, or whatever you like.

Hello,

I just tried your program. However, my server 2003 x64 Windows says it's not written for this machine type.

This means your program must be containing 16 bit code...hence it won't run on any 64 bit windows. Maybe you could look into this and see if your compiler can generate all 32 bits code. Many overclockers prefer 64 bit OSses, even if it is just for supporting more than 4 GB of RAM.

Other than that, this tool could be very useful! Thanks for posting

I recompiled the code using another compiler and it should work now. I actually noticed yesterday that it didn't work in my Vista x64. Thank you for posting.

[Deathspine]

Thanks for the update, now works in my Vista 64.

[-n7-]

I pretty much have a headache from reading thru this thread...

I am trying to get 450 FSB stable with a:
Q9550 @ 7x450 (to ensure not hitting max CPU for now)
P5Q Deluxe bios 1406
2x2 GB RAM set all very loose to ensure its out of the equation

I am finding P95 large FFTs seems to be the fastest way to show instability due to VTT?

Small FFTs takes forever to error out, yet Large will error almost immediately if VTT isn't right.
Or with Large P95 freezes eventually if not right.
Blend seems somewhere in between; it will eventually error, but doesn't as fast as Large.

Of course Intel Burn Test also fails, but it's hard to get results immediately.
It either errors or the program justs crashes & stops running completely

I'm going insane though, because i cannot seem to figure out what VTT to use?

I've tried auto GTLs with 1.3v VTT, & it's instant fail.
If i keep increasing VTT up to 1.4v, it takes longer & longer to fail or crash, but it's still not stable.

Surely i don't need such high levels?

I've tried lower values of VTT + higher GTLs, & higher VTT + lower GTLs, yet it still fails.

I cannot seem to figure out the magic combination.

I tried your settings truehighroller:

Vcore: 1.40 leave it and mess with the ones below and see if you are stable then try to back this down,

cpu gtl 0/2: auto set these to .63
cpu gtl 1/3: auto .... .63
Vpll: 1.60 set this to 1.56
vfsb: 1.46 try 1.22 like mine is
vdram: 2.10 Overvolting right now drop it to 2.02v and see if it is ok.
NB gtl: auto
Vnb: 1.38 Set this to 1.30v
Vsb: 1.10 Set this to 1.20v as it is overvolting your south bridge set like you have it right now "bad".

VPcie: 1.50 same as the sb overvolting, set this to 1.6

But those results in instant failure in Large FFTs at a mere 7x450.

Anyone have any ideas to help a GTL n00b?

[ahmad]

No one can tell you exactly how many % coverage you will pass in memtest. You have to try it and see. However, if you can pass Orthos small FFT and Lipnpack, which are primarily cpu stressors, but fail memtest at ~300%, I'd say you need to increase vdimm slightly, perhaps vNB slightly too. Try that before you go GTL tweaking.

Windows memtest is actually very hard on the ram and NB and I actually needed 2 notches more vdimm to pass it, although I was Linpack & Orthos Large & Small FFT stable.

Thanks for your suggestions.

[CryptiK]

I pretty much have a headache from reading thru this thread...

I am trying to get 450 FSB stable with a:
Q9550 @ 7x450 (to ensure not hitting max CPU for now)
P5Q Deluxe bios 1406
2x2 GB RAM set all very loose to ensure its out of the equation

I am finding P95 large FFTs seems to be the fastest way to show instability due to VTT?

Small FFTs takes forever to error out, yet Large will error almost immediately if VTT isn't right.
Or with Large P95 freezes eventually if not right.
Blend seems somewhere in between; it will eventually error, but doesn't as fast as Large.

Of course Intel Burn Test also fails, but it's hard to get results immediately.
It either errors or the program justs crashes & stops running completely

I'm going insane though, because i cannot seem to figure out what VTT to use?

I've tried auto GTLs with 1.3v VTT, & it's instant fail.
If i keep increasing VTT up to 1.4v, it takes longer & longer to fail or crash, but it's still not stable.

Surely i don't need such high levels?

I've tried lower values of VTT + higher GTLs, & higher VTT + lower GTLs, yet it still fails.

I cannot seem to figure out the magic combination.

I tried your settings truehighroller:



But those results in instant failure in Large FFTs at a mere 7x450.

Anyone have any ideas to help a GTL n00b?

You have tried the basics, but if it errors out in large FFT's at even 7 x 450 I'd be leaning towards a ram/northbridge related problem. You are running 4 x 2GB sticks, you may need 1.4v+ on the NB to get that stable at 450 FSB. Try increasing vNB to 1.4 - 1.45v and see if there is an improvement. Leave the multi low for the test to eliminate the CPU. Once you get 450 FSB stable, you can work the cpu back up to the desired speed.

To properly find the max stable cpu clock you should pull 2 sticks and just run 4GB till you have that sorted.

[-n7-]

You have tried the basics, but if it errors out in large FFT's at even 7 x 450 I'd be leaning towards a ram/northbridge related problem. You are running 4 x 2GB sticks, you may need 1.4v+ on the NB to get that stable at 450 FSB. Try increasing vNB to 1.4 - 1.45v and see if there is an improvement. Leave the multi low for the test to eliminate the CPU. Once you get 450 FSB stable, you can work the cpu back up to the desired speed.

To properly find the max stable cpu clock you should pull 2 sticks and just run 4GB till you have that sorted.

2x2 GB RAM set all very loose to ensure its out of the equation

Thought of that before i even put the chip in.

Yesterday i basically went from 1.2 vtt to 1.4 vtt trying pretty much all the GTL values (0.6 to 0.75 basically) for each vtt voltage increment.

I eventually just gave upon the finesse approach as no combination was working.

1.42 vtt with auto GTLs seems stable @ 450, but even going to 1.44 does not gain anything; maybe 452 at most; more testing needed.

Seems this board really hates anything over 450, at least for stability.

I can boot way higher, but anything over 450 won't pass IBT or P95 Blend/Large.

[truehighroller]

Thought of that before i even put the chip in.

Yesterday i basically went from 1.2 vtt to 1.4 vtt trying pretty much all the GTL values (0.6 to 0.75 basically) for each vtt voltage increment.

I eventually just gave upon the finesse approach as no combination was working.

1.42 vtt with auto GTLs seems stable @ 450, but even going to 1.44 does not gain anything; maybe 452 at most; more testing needed.

Seems this board really hates anything over 450, at least for stability.

I can boot way higher, but anything over 450 won't pass IBT or P95 Blend/Large.

What CME option are you set to? I noticed with mine set to performance or compatible that I was getting random errors. Just a thought. Maybe you should try 0803 BIOS revision and see if your issues go away. It could also be you already jacked your gtl settings so high for awhile that maybe now your chip doesn't like lower settings any more. Just a guess as well. Mine will not run lower then what it is set to now and I read some where that after awhile per Intel that if you run higher settings that your chip will no longer like the lower settings recomended by Intel.

[-n7-]

What CME option are you set to? I noticed with mine set to performance or compatible that I was getting random errors. Just a thought. Maybe you should try 0803 BIOS revision and see if your issues go away. It could also be you already jacked your gtl settings so high for awhile that maybe now your chip doesn't like lower settings any more. Just a guess as well. Mine will not run lower then what it is set to now and I read some where that after awhile per Intel that if you run higher settings that your chip will no longer like the lower settings recomended by Intel.

I'm using 1406, & CME is at default, which is "Optimized"

I've only had this CPU running for a day now, & i don't like using higher voltages, but from what i've gathered, up to 1.45 VTT should be safe?

Or maybe i was reading wrong.

But i highly doubt anything was damaged, as it was unstable @ 450 with lower VTT from the very first time i stresstested.

I would love to get things stable with lower VTT, but thus far, that seems impossible at 450 FSB.

I wish i could do the voltages you are running...what bios are you using?

I know you were having all kinds of issues with anything higher than 803, but i thought i'd read it was fixed for you know in 1406?

[truehighroller]

I'm using 1406, & CME is at default, which is "Optimized"

I've only had this CPU running for a day now, & i don't like using higher voltages, but from what i've gathered, up to 1.45 VTT should be safe?

Or maybe i was reading wrong.

But i highly doubt anything was damaged, as it was unstable @ 450 with lower VTT from the very first time i stresstested.

I would love to get things stable with lower VTT, but thus far, that seems impossible at 450 FSB.

I wish i could do the voltages you are running...what bios are you using?

I know you were having all kinds of issues with anything higher than 803, but i thought i'd read it was fixed for you know in 1406?

I have mine set to compatible and my issues have stopped now while playing far cry 2. Try compatible and see if it acts better and yes it is supposed to be fixed now....... I don't care what any one says those voltages aren't safe.

[cadaveca]

Voltages are only not safe when excess current between loading stages cannot be handled correctly, so as example, just increase vcpu without VTT change could be problematic, but to not increase VTT with vcore would be more likely to have caused damage as something must eat the excess current from one voltage set to the next...


example:

vcore 1.25, VTT 1.1, Vchipset 1.25,

increase only vcore to say 1.45v,

vtt circuit must "eat" excees voltage from 1.45v back down to 1.1v of vtt


...that's what kills.

[DemonEyez]

I have mine set to compatible and my issues have stopped now while playing far cry 2. Try compatible and see if it acts better and yes it is supposed to be fixed now....... I don't care what any one says those voltages aren't safe.

As explained earlier, as long as vTT is higher than vCore it's not that big a deal running a 'high' vTT

Look at that other topic here on Xtreme. There's a guy running a vTT of 1.6v for over a month and his rig is still up and running.
http://www.xtremesystems.org/forums/...d.php?t=198614

[truehighroller]

As explained earlier, as long as vTT is higher than vCore it's not that big a deal running a 'high' vTT

Look at that other topic here on Xtreme. There's a guy running a vTT of 1.6v for over a month and his rig is still up and running.
http://www.xtremesystems.org/forums/...d.php?t=198614

I guess. I won't argue if you want to then jack them up that is fine but, Mine runs fine set lower I would think you would rather shoot low then high. Doesn't it affect your temps as well? Correct me if I'm wrong, I'm guessing it does.

[DemonEyez]

If it runs fine than that's great and your correct in saying you don'nt need to change it, I fully agree.
But my post was a reaction on your statement saying "I don't care what any one says those voltages aren't safe"
Now that's just an assumption, and untill now no one has come with hard evidence that it is so dangerous as some people are claiming, fuelling the whole myth around vTT.

You are correct that with higher vTT the temps will rise.
I've had mine at 1.35v vTT and 1.33v vCC and dual priming at 4.5GHz gave me load temps of around 68 degrees, so watercooling is somewhat of a must have :P

[truehighroller]

I thought I had noticed my temps got higher when messing with it but, wasn't sure. I mean if you can and it helps then jack it up but, I just feel safer with low ones that's all. I could probably get my 4Ghz mark with jacking mine up but, it wouldn't be 24/7 so it. I will leave mine where it is as I am higher then most people any way with my chip where it is.

[mikeyakame]

vtt shouldn't really raise temps all that much, since its not actually applied directly to the cpu itself like vcc. vss/vtt are termination voltages and as such would at most add a little more heat, but nothing near adding an extra 0.05v of vcc would do. the only thing it would heat up would be the voltage regulation mosfets!