CPU : Vcore,VTT and GTLREF question

[Demo]

Im wondering if the VTT/GTLREF can help me stabilize my OC w/o bumping my CPU Vcore,

Currently im using 1.335v ( bios ) for 400 * 8 on my Q6600 G0, and left the CPU VTT to 1.20v and GTLREF at 63% I can run at 1.315v only 8hours prime stable,after that 1 random core gets an error.

My temps are fine and dont hit more than 63c on a scorching afternoon.

[cantankerous]

VTT is used to stabilize the CPU at high FSB. Since you are only running 400mhz a cpu VTT of 1.20 should be fine. A slight raising to 1.25 or 1.30 wouldn't hurt anything and could be tried to see if your prime session will go longer. GTLREF, especially on quad cores can play a HUGE role in keeping your cpu stable, especially at high clock speeds matched with high FSB. Though not set in stone, a good value for the GTL is 67% and the value is based on current VTT setting. You did not mention which board you are using, though if you are able to manually select a GTL value this homepage may come in mighty handy for you. It is put together by a forum member by the name of Praz here on his message board. Some info on GTL can be found here.

http://www.edgeofstability.com/artic.../gtl/gtl1.html

some values to try out can be found here.

http://www.edgeofstability.com/artic...fault_gtl.html

This was put together with the new DFI offerings in mind, but perhaps you can use some of this information to your advantage. Remember, these values are not set in stone and are just a guideline. The values can sway either way of the above listed by a few notches depending on your current setup as no two parts are identical.

GTL values, if set correctly are great as they allow you to maintain stability without using a 'brute force' attempt to keep things stable by throwing high levels of voltage through your system. If set incorrectly they can also do more damage than good stability wise. If random cores are failing try and adjust the GTL value related to your NB. If one, or a certain pair of cores seems to keep failing adjust the GTL value for that said core/pair. Again, I am unsure on what sort of flexibility your board has as I am going by my DFI which is a very hands on board in terms of bios manipulation.

GTL value manipulation plays a larger role in quad core cpus on 65nm but seem to also be of equal importance with dual core cpus on 45nm process.

[Demo]

Abit IP35-E, So the best option would be to go back to 1.315v then bump up GTL to 65% and vtt to 1.23 ? ( 63% and 1.20 are defaults ) and if i see stability increase over 24hours of priming ill try to bump down either the VTT or the GTL to see.

[Blacklash]

I wasn't able to hold 3.8 until I increased my CPU VTT to 1.2375 and CPU GTLREF to 67%.

[Demo]

Tried 1.23vtt and 65% GTLREF, core03 failed after 3 hours at 1.315v, should i gun for 67% or just bump the vcore to 1.335 ?

[Blacklash]

Tried 1.23vtt and 65% GTLREF, core03 failed after 3 hours at 1.315v, should i gun for 67% or just bump the vcore to 1.335 ?

Since you are doing 400x8, go vcore. For the record, I need 1.36v to hold 3.2GHz prime stable. Eventually, you'll need the higher CPU VTT and CPU GTLREF as you climb in clocks. I didn't need to push the GTLREF up until I was trying to get above 3.75GHz. I can do 400x9 with 1.46v, and my ram @ 1.95v. I don't need to touch any other voltage.

My comment was simply to let you know what enabled me to reach 3.8 stable was ultimately increased CPU VTT and CPU GTLREF. It was something to keep in mind if you push much higher and hit a wall.

I am doing 3.8 like so-

1:1 CPU|ram ratio to squeeze those last MHz out, I am @ 423x9.

Memory-

tRFC 42 for the best stability. I manually set my ram voltage and do not use auto. I get 4Gb of ram with 2 sticks not 4x 1Gb.

CPU Settings-

Vcore- 1.54v
CPU VTT- 1.2375
ICH- 1.60v
MCH- 1.33v
CPU GTLREF 67%
PCI-e 100 for best stability

Virtualization- disabled.

C1E and EIST are enabled. Leave them on if they don't cause problems because they will help keep your rig cool.

[Ron 61]

VTT is used to stabilize the CPU at high FSB. Since you are only running 400mhz a cpu VTT of 1.20 should be fine. A slight raising to 1.25 or 1.30 wouldn't hurt anything and could be tried to see if your prime session will go longer. GTLREF, especially on quad cores can play a HUGE role in keeping your cpu stable, especially at high clock speeds matched with high FSB. Though not set in stone, a good value for the GTL is 67% and the value is based on current VTT setting. You did not mention which board you are using, though if you are able to manually select a GTL value this homepage may come in mighty handy for you. It is put together by a forum member by the name of Praz here on his message board. Some info on GTL can be found here.

http://www.edgeofstability.com/artic.../gtl/gtl1.html

some values to try out can be found here.

http://www.edgeofstability.com/artic...fault_gtl.html

This was put together with the new DFI offerings in mind, but perhaps you can use some of this information to your advantage. Remember, these values are not set in stone and are just a guideline. The values can sway either way of the above listed by a few notches depending on your current setup as no two parts are identical.

GTL values, if set correctly are great as they allow you to maintain stability without using a 'brute force' attempt to keep things stable by throwing high levels of voltage through your system. If set incorrectly they can also do more damage than good stability wise. If random cores are failing try and adjust the GTL value related to your NB. If one, or a certain pair of cores seems to keep failing adjust the GTL value for that said core/pair. Again, I am unsure on what sort of flexibility your board has as I am going by my DFI which is a very hands on board in terms of bios manipulation.

GTL value manipulation plays a larger role in quad core cpus on 65nm but seem to also be of equal importance with dual core cpus on 45nm process.

Is there any information like this for wolfdales? This seems to be targeted to quad core cpu's.

[mad_skills]

Yes, GTL settings ought to be used only with high FSB and when pushing your cpu in high speeds compared to default frequency.

[Demo]

I see, i guess none of those settings will help me yet as im currently contented with 3.2Ghz

BTW: my tRFC value by SPD is at a 51 default at 400FSB, would lowering it to 42 make my system unstable ?

[Praz]

Is there any information like this for wolfdales? This seems to be targeted to quad core cpu's.

I'm not sure what type of info you are looking for. The principles don't change because of using a Wolfdale other then the CPU GTL target is closer to 63%-64%.

[Ron 61]

I'm not sure what type of info you are looking for. The principles don't change because of using a Wolfdale other then the CPU GTL target is closer to 63%-64%.

The charts linked above don't show cpu vtt as low as the default 1.1v for my E8400. Also I have not seen default GTL Values as high as 84/82, my boards have all defaulted to 67%. Please pardon my ignorance on this subject, I am just trying to make some sense of it.

[Praz]

The settings in the charts are applicable to the DFI boards. They don't use percentage. The settings are a theoretical numerical value of 0 - 255 which allows a very fine control of GTL voltage. 65nm processors have a default GTL voltage of 1.120 while 45nm processors are at 1.100 volts.

[g0dM@n]

The settings in the charts are applicable to the DFI boards. They don't use percentage. The settings are a theoretical numerical value of 0 - 255 which allows a very fine control of GTL voltage. 65nm processors have a default GTL voltage of 1.120 while 45nm processors are at 1.100 volts.

I thought that 0 - 255 was some sort of 256-bit code/setting, but when people speak about it, 110 value = 1.10v, so on and so forth.

So theoretically we can push up to 2.55v?

You're right about 1.100v as the default for my wolfdale (GTL setting of 110), so I don't understand why the default is set to 1.10v when the default VTT is also 1.10v. That would mean it was running 100% by default!!

[Praz]

I thought that 0 - 255 was some sort of 256-bit code/setting, but when people speak about it, 110 value = 1.10v, so on and so forth.

When people speak of the relationship like that they do not understand how the settings work on the DFI board. That value of 110 will represent a different voltage based on what VTT is set at. Just as every other value will.

So theoretically we can push up to 2.55v?

No. VTT does not scale high enough.

You're right about 1.100v as the default for my wolfdale (GTL setting of 110), so I don't understand why the default is set to 1.10v when the default VTT is also 1.10v. That would mean it was running 100% by default!!

When the GTL settings are disabled (default) the value that is displayed, 110 in this instance, is the current value. The current CPU GTL value will be approximately 67% of VTT for 65nm processors and 63% of VTT for 45nm processors. Regardless of the processor NB GTL will be approximately 67%.

[Demo]

Isnt the defualt GTL for the 65nm at 63% only ?

[jonny_ftm]

In some anandtech reviews, they stated that GTL ref voltage are only reference voltages, so no risk to degrade a CPU. But, I never could find any more info on this despite all researches. So anyone, with enough technical knowledge, can confirm the total safety of gtl ref voltages?

[graysky]

Anyone have a reference for this 63 % GTL setting for 45 nm and 67 % GTL for 65 nm?

[Rinaun]

Anyone have a reference for this 63 % GTL setting for 45 nm and 67 % GTL for 65 nm?

I too would like to find out where these numbers come from, because mine so far is showing better stability at 69%.

[Big Lar]

I think they are in the Intel white papers. I believe Praz showed and quoted from it in the X48 DFI thread if not mistaken.

Larry

[jonny_ftm]

And any confirmation on absolute safety of these settings?

[Praz]

Anyone really interested in GTLREF techniques I would suggest picking up a book or two pertaining to the design of high speed signaling circuits. GTLREF has been in use for quite some time. Intel didn't invent it. They adopted the technology and modified it.