GTLREF and FSB VTT Voltage Tweaking

Status
Not open for further replies.

Aaron

Active member
GENERAL
Joined
Nov 2, 2006
Messages
5,690
*** This is all done at your own risk. I, any user on this forum, this forum, or MSI will NOT be held responsible for your actions! ***


I'm sure you've heard about GTL reference voltages.

I've decided to write a simple guide (it's not that simple any more... lol) so, people can understand how to tweak these voltages, and get those high FSB &/or CPU clock speeds.  I'll get right down to business...

A quick breakdown of exactly what A/GTL+ [Advanced Gunning Transceiver Logic] does:

[quote author=Intel Q9000 Datasheet]
Most processor Front Side Bus signals use Gunning Transceiver Logic (GTL+) signaling technology.  This technology provides improved noise margins and reduced ringing through low voltage swings and controlled edge rates.  Platforms implement a termination voltage level for GTL+ signals defined as VTT.  Because platforms implement separate power planes for each processor (and chipset), separate VCC and VTT supplies are necessary.  This configuration allows for improved noise tolerance as processor frequency increases.  Speed enhancements to data and address busses have caused signal integrity considerations and platform design methods to become even more critical than with previous processor families.
[/quote]

Rather optional: read this article on how the values work: http://www.thetechrepository.com/showthread.php?t=87 |  It's a great explanation and shouldn't require you to do any more research into the matter.  However, if you still don't understand how GTLREF works, and how VTT effects it: Google is your friend.


You can tweak the GTLREF and VTT (AKA FSB VTT voltage/FSB Terminator voltage) via the Cell Menu in the CMOS setup.

There are many different GTLREF voltages.  There are GTLREF values all over mainboards.  The ones we want however, are for each CPU die (only P45 boards have the settings to signal each die (so far?)) or both CPU dies (every board excluding P45), and the ICH/NB.  You could guess the names.  They are pretty obvious, and sometimes named differently on each board, but they all do the same thing, generally speaking.

The P45 boards have quite an edge.  You can sometimes get a flaky core that needs more voltage than the rest. You can fix that problem with the individual die settings.

NOTE: You will NOT find these options present on a budget/low-end board.  Do not ask if it will have the options in a future BIOS release - it won't.


According to Intel's datasheets: the nominal voltages for GTLREF are 2/3 of the VTT voltage.  However, I've found better results from setting ~63% on the CPU [Yorkfield and Kentsfield] and ~67% on the NB [780i, P35 and X48].

You use a simple formula to calculate what each value will be:

Code:
a * b = c

a = FSB VTT voltage
b = percentage
c = GTLREF setting

For example:

1.37v * 67% = 0.9179v - Note how precise my result is. The more precise you are, the better chance you will have of getting in the correct margin. Vnoise must also be taken into account. It's +/-10%.  However, I doubt noise is this high on high end boards.  Call it +/-3-5% if you have a high end board with a good power phase.

NOTE: If upon tweaking, nothing is improved, you may need to go up/down a few notches.

<hr>

After further investigation to Intel's datasheets.  There is a maximum the GTLREF should be set to both in an above and below manner.  The minimum and maximum should be:

Min:

Code:
0.550 * VTT = min

Max:

Code:
0.725 * VTT = max

[For those of you that are quick minded, that means don't go below 55% or over 72.5%]

These GTLREF voltages are terminated on-die, going TOO HIGH will most likely mean the death of your CPU!  Going TOO LOW is guaranteed instability!

<hr>

Do not hesitate to tweak these values, but do not go beyond the limits!  They can sometimes give you great results.  On the other hand, they can also make your results worse (yes, worse).  That means you need to tweak the values more.  Please keep that in mind.

Post your results in here - I'd like to see them.

Good luck with your tweaking!

If you have anything to add or change here, don't hesitate to contact me!

<hr>

A sidenote for P45 users:

MSI Tech.  11/03/2008  This is decision of our engineers which work directly with Intel. 
End User  11/03/2008  They also recommend you don't go about 1.5 volts on the CPU, but this mobo will do it. Is there a technical reason why the GTL Ref voltage will not to 67%. All technical documents from Intel suggest 72% is the limit. 
MSI Tech.  11/03/2008  No, since this is recommended by Intel. 
End User  11/03/2008  I can not set the GTL Ref voltage to 67%, the highest it will go is 63%. This may be limiting my overclocking on my quad core. It there anyway to change this? Thanks

james1701 contacted me via PM and shared this info with me.  It appears MSI refuses to put the GTLREF voltage over the limit of 63%.
 
very fine!  :agrees:
Finaly a User who speak about GTL?s.

I want to add on:

Some overclocking experiments shows, that CPU?s with Quad Core in a general rule need 1% - 3% (68-70%) more GTL.

And Cpu?s with Dual Cores needs 3 - 4% less GTL. Also 63 - 67%

The NB GTL Reference Voltage should be around 67%.-


But i want to underscore AaronYuri :
all GTL voltage me be agreed!!
Otherwise you have blue screens, or the system wont boot! Demage should be not excluded!

 
DJRamses said:
Some overclocking experiments shows, that CPU?s with Quad Core in a general rule need 1% - 3% (68-70%) more GTL.

And Cpu?s with Dual Cores needs 3 - 4% less GTL. Also 63 - 67%

That's why I mentioned Vnoise in the post above. :yes:

AaronYuri said:
1.37v * 67% = 0.9179v - Note how precise my result is. The more precise you are, the better chance you will have of getting in the correct margin. Vnoise must also be taken into account. It's +/-10%. However, I doubt noise is this high on high end boards. Call it +/-3-5% if you have a high end board, with a good power phase.

DJRamses said:
all GTL voltage me be agreed!!
Otherwise you have blue screens, or the system wont boot! Demage should be not excluded!

Do you mean all GTLREF voltages on the mainboard must be identical, or...?
 
GTL will be DIFFERENT for the Intel quad cores released up to this point.  That's because they're not NATIVE quad cores.  They're TWO dual cores that are on the same chip but they communicate with each other through the northbridge.

I hope this helps whoever was asking if GTL should all be the same.  :)

It's the whole point of why there is a GTL setting in the first place.  Because even with the best quality control, the two dual cores on a quad processor won't be identical, and when you overclock, those differences get magnified to the point where you need to manually modify the GTL stuff.  If you have a dual core, there's probably no need to mess with the GTL unless you bought a $10 power supply from the back of a van.  :-D

You should explain ringback to everyone Aaron.  :)  Because you have an afternoon to waste right?
 
autumnale said:
Because you have an afternoon to waste right?

Hahaha! :lol_anim: Sadly, I have a life to live! ;) Google is everyone's friend. Someone else can give the answers, I'm sure.

If you got a crap PSU, you shouldn't be overclocking anyway.

Explaining it in more simpler terms: During the transition from the start to the termination point, there's a period where noise will come into play; this causes a "skew" on the signal. Altering the GTLREF (which atone for the signal's voltage) causes this signal's skew to change. What you're basically trying to find is a "sweet spot" where the transition from start to termination is as clean as possible.

And now we return back to reality! :lol_anim:
 
Hi AaronYuri and Autumnale,

I'd like to pick your brains if you don't mind.  Both of you have made excellent posts here and on the extremesystems.org forum and have helped me very much with setting up my P7N Diamond.  I've learned a tremendous amount on the finer aspects of CPU OC'ing.

This thread about GTLRef and FSB VTT tweaking is especially valuable.  But would either or both of you care to elaborate more on the relationship between Vcore and VTT?  I've seen references (from both of you and other posters) to http://edgeofstability.com/articles/dfi_p35/gtl/gtl3.html and statements declaring that VTT should not exceed Vcore.  From the link, I've come to understand that Intel processors have a safety mechanism to improve stability, but that raising VTT above Vcore will effectively disable the mechanism.

Code:
Another issue that most users are unaware of is the relationship between VCORE and VTT. To reduce the amount of ringing at the driver,
Intel has added a weak pull-up device to the output buffer as mentioned above. This device turns on at the beginning of a low-to-high signal
transition, substantially reducing the impedance mismatch between the output buffer and the transmission line. As a result, the amount of
overshoot and ringback is significantly reduced. The source terminal of the pull-up device is connected to the core voltage supply. This causes
the logic high voltage to rise above the GTL termination voltage for one cycle. After one bus cycle, the pull-up device is turned-off and the output
will stabilize at VTT if the output remains in the logic high state. But when we raise VTT above VCORE we have effectively removed the pull-up
device from the circuit.

Therefore can anyone clarify how much of a threshhold exists between VTT and Vcore?  Also, does this rule refer to the absolute minimum Vcore during operation (net of Vdrip and Vdroop under full load), or just the average or max Vcore?

On my P7N Diamond / Q6700 combination, I have VTT (FSB Term Volt) set at 1.293V.  My CPU VID is 1.2625V (per CoreTemp).  With cell menu Vcore set to +0.1250V, my idle Vcore varies between 1.344V and 1.352V as per CPU-Z (~0.04V Vdrip).  At load, Vcore varies between 1.296V and 1.304V as per CPU-Z (~0.05V Vdroop).  The voltages fluctuate frequently but mostly center around 1.352V idle and 1.304V load (Prime 95 Small FFTs).

As such, my Vcore is typically well above my VTT.  But for brief periods, at it's absolute lowest point, my Vcore falls to 1.296V, just 0.003V above my VTT of 1.293V.  Assuming, the readings from CPU-Z are accurate, is this sufficient to not cause any conflicts with the safety mechanism?  I suppose I could lower VTT further but I've managed to run 12 hr+ Prime 95 Small FFTs under the current setup and am ready to FREEZE my cell menu tweaks.

Should I consider my settings stable or continue experimenting with lower VTT (to avoid conflicting with the pull-up safety device)?  In any case (yes, you both do have lives and may not have time to respond), thanks for all your valuable explanations.
 
Someone of you have problem with GTL on P45 NEO2-F?? I dont have 63-67% VTT Voltage results on CPU GTL REF0 and REF1.

Please help if someone know how to set this.
 
Tygrys, open a separate topic up. This thread is for results not troubleshooting.

Guessing from this statement:

I dont have 63-67% VTT Voltage results on CPU GTL REF0 and REF1.

You didn't quite understand the post. You must calculate the values yourself, you can't just set a percentage.

Stooper, not sure about the VTT and Vcore relation--I haven't looked into it. It is advised that you keep VTT below the Vcore, however. Judging from that statement it means while it won't effect stock speed stability; it'll definitely hold you back in overclocking.

Like I said with the GTL adjustments: you're looking for that "sweet spot" where the skew is spot on. If you do not achieve that skew, you do not achieve the higher clock speeds.
 
AaronYuri said:
Stooper, not sure about the VTT and Vcore relation--I haven't looked into it. It is advised that you keep VTT below the Vcore, however. Judging from that statement it means while it won't effect stock speed stability; it'll definitely hold you back in overclocking.

Like I said with the GTL adjustments: you're looking for that "sweet spot" where the skew is spot on. If you do not achieve that skew, you do not achieve the higher clock speeds.

I see.  I think I'm happy with my OC to 3.4GHz.  OCCT, Prime95, and Orthos all seem to be happy too  :-P) Whether or not my VTT is conflicting with the pull-up safety mechanism, the system seems rock solid.  So I'm going to consider the system stable and FREEZE my volt tweaks.  I think I can push OC higher but don't want to raise volts/temps any further.

Thanks again.
 
i wonder is this some sort of 'safety' mechanism in GreenPowerCenter for P45 platinum boards which prevents the GLTref to be set over 65.5% and NB GLT REF over 62% ?
in BIOS it seems GTLref is limited on 63-64% max
 
Dwarden said:
i wonder is this some sort of 'safety' mechanism in GreenPowerCenter for P45 platinum boards which prevents the GLTref to be set over 65.5% and NB GLT REF over 62% ?
in BIOS it seems GTLref is limited on 63-64% max

Intel's datasheets state that 72.5% is the maximum.  We don't exactly control how a BIOS is written either.  So the limits are obviously of MSI's own will.
 
Dwarden said:
i wonder is this some sort of 'safety' mechanism in GreenPowerCenter for P45 platinum boards which prevents the GLTref to be set over 65.5% and NB GLT REF over 62% ?
in BIOS it seems GTLref is limited on 63-64% max

That's unfortunate that the P45 prevents CPU GTL Ref from exceeding 65.5%.  As mentioned, >67% appears to be very relevant for OC'd quad core processors.  As for the "safety" mechanism I'm referring to, it's unrelated to GTL Ref limits (neither CPU nor NB) in mobo BIOS settings.  Rather, it pertains to VTT considerations for Intel processors...

After reviewing the article on edgeofstability.com, I think I've found the answer to my own question.  When transitioning from low signal (logical 0) to high signal (logical 1), the source voltage is connected to Vcore supply.  This occurs on the leading edge of the state transition, thus aiding to "pull-up" the voltage to high state.  So long as Vcore is higher than VTT, then the signal will be pulled "above and beyond" VTT, in turn helping to reduce overshoot and resulting ringback.  On the next cycle, the signal will settle back down to VTT (if state remains high).  Clearly, this design is only effective if Vcore > VTT.

In contrast, if Vcore is **lower** than VTT, then the "pull-up" mechanism will be insufficient for pulling the signal past VTT voltage level.  This, in and of itself, isn't an issue as the signal will still transition to high state.  However, there will be greater likelihood of overshoot and, in turn, ringback.  Hence, maintaining the VTT < Vcore relationship is an important guideline for improving stability, but it isn't an absolute necessity.  A system may operate very stable despite violating the guideline.  An important factor appears to be the quality of the power supply and amount of noise that occurs.
 
stooper said:
That's unfortunate that the P45 prevents CPU GTL Ref from exceeding 65.5%.  As mentioned, >67% appears to be very relevant for OC'd quad core processors.  As for the "safety" mechanism I'm referring to, it's unrelated to GTL Ref limits (neither CPU nor NB) in mobo BIOS settings.  Rather, it pertains to VTT considerations for Intel processors...

After reviewing the article on edgeofstability.com, I think I've found the answer to my own question.  When transitioning from low signal (logical 0) to high signal (logical 1), the source voltage is connected to Vcore supply.  This occurs on the leading edge of the state transition, thus aiding to "pull-up" the voltage to high state.  So long as Vcore is higher than VTT, then the signal will be pulled "above and beyond" VTT, in turn helping to reduce overshoot and resulting ringback.  On the next cycle, the signal will settle back down to VTT (if state remains high).  Clearly, this design is only effective if Vcore > VTT.

In contrast, if Vcore is **lower** than VTT, then the "pull-up" mechanism will be insufficient for pulling the signal past VTT voltage level.  This, in and of itself, isn't an issue as the signal will still transition to high state.  However, there will be greater likelihood of overshoot and, in turn, ringback.  Hence, maintaining the VTT < Vcore relationship is an important guideline for improving stability, but it isn't an absolute necessity.  A system may operate very stable despite violating the guideline.  An important factor appears to be the quality of the power supply and amount of noise that occurs.

Where's a clapping smiley when you need one?
 
AaronYuri said:
Intel's datasheets state that 72.5% is the maximum.  We don't exactly control how a BIOS is written either.  So the limits are obviously of MSI's own will.

well i was bit shocked this same problem exist even in the performance BIOSes and it most likely explains the 'OC walls' these MSI boards have ...
 
I partially understand the principles behind GTL referance voltage. Orthos wil run for 2 sec on auto and 20min after I messed around a bit, the thing I don't understand is that my MSI P43 Neo-F has CPU GTL Ref0 and CPU GTL Ref1.
Leaving those settings on auto Ref0 will be 63% of VTT, but Ref1 has a much higer percentage(didn't calculate) of VTT. Where does this difference come from?
Since I have a budget motherbord and a E7300. I should not be able to tweak the individual cores, but what if I can do that, why is there a big difference between ref0 and ref1?

I'm running bios 1.6 and 8.5x415 with 1.36Vcpu and 1.25VTT.
 
I am not sure of the differences between those two settings because I have no experience with the board.  However, because 0 is 63%, I'm guessing that's for your CPU die, and because the other is a higher percentage [67-69%], it's probably for the ICH.

This being said, take it with a grain of salt, as only MSI could really say what is truly what.
 
After reading some articles again, I answered part of my own question. Ref0 is for data and Ref1 is for adresses. I have yet to find out if the diff in % of VTT between ref0 en ref1 should be as they are on my board.
 
Leaving those settings on auto Ref0 will be 63% of VTT, but Ref1 has a much higer percentage(didn't calculate) of VTT.

Razkin, I am curious about one thing:  How do you know that there is a difference between those values, when they are set to AUTO?  The values do not appear in BIOS as far as I know, so which tool do you use to monitor the AUTO Settings?  If I am not mistaken, Green Power Center does not support your Board and Dual Core Center (at least as far as I remember) does not display GTL Voltage Settings.  I am asking, because when I check the GTL Auto Settings with Green Power Center on my P45 Platinum, the values are always identical:



...always 63.45% of VTT, to be more precise.
 
Status
Not open for further replies.
Back
Top