GTL Ref Tweaking Guide

[saaya]

Hey guys,

Been playing with Blackops and GTL Ref settings today and studdied the docs from Intel a bit. Its 10pm and i need to get some sleep, some quick points.

1. Core0 and Core2 are in the same chip? can anybody confirm?
2. Core1 and Core3 are in the same chip? can anybody confirm?
3. i meassured CPU GTL Ref volts, and the table i posted earlier is a bit off
4. 1 step is pretty much exactly 7mv
5. the BIOS maps VTT incorrectly! default vtt for 45nm cpus is 1.10v, which is applied by default, but its displayed as 1.20v. so when you use a 45nm cpu and set vtt to 1.30v its actually only 1.20v. That makes some fsb overclocks ive seen here even more amazing since they were actually with .10v vtt less!
6. this will be fixed in G33 thatll come out soon
7. Core1 GTL Ref is higher than the rest of the Cores by default, and is linked to NB Gtl!
8. Core0 2 and 3 was 720mv GTL Ref by default for me (which is 1.100v vtt x 0.655 instead of the official spec of x 0.635, its only slightly off tho, and it varies by board and cpu, so nothing special)
9. Core 1 has 760mv GTL Ref by default
10. Core1 always has 6 steps higher default GTL Ref, so when you set
    +10+10+10+10
    its actually
    +10+16+10+10
    this is per intel spec! so it should be like this on all boards...
    default per intel spec is
    Core0 0.635 x VTT
    Core1 0.700 x VTT
    Core2 0.635 x VTT
    Core3 0.635 x VTT
11. every board will have slightly different GLT Ref settings, with the same cpu, some boards will have a default GTL Ref +/- 30mv which is around 4 steps.
12. GTL Ref of Core0 2 and 3 should always be within +/- 5mv though since they are related according to intel spec

I made a little guide to show where to meassure GTL Ref voltages and how to calculate GTL Ref voltages easily for a given VTT

http://depositfiles.com/files/8755756
(kostenlos download -> download)



Here are some good guides explaining GTL Ref:
BlazingPC
EdgeOfStability

[negev]

Ok im confused.

How do you know what GTL voltage you want in the first place? Other peoples settings? Black Magic/Voodoo?

Also, your spreadsheet says "enter current VTT setting or measured VTT in highlighted field" but you said that the vtt settings is +0.10v higher than the real vtt, so won't this affect the calculation?

[Ryan12]

Ok im confused.

How do you know what GTL voltage you want in the first place? Other peoples settings? Black Magic/Voodoo?

Thats my exact question as well... I assume we just need trial and error and see which works best like any other voltage i guess?

[negev]

Well the attachment documents measuring points for reading the voltage with a multimeter, although I'm not sure what kind of heatsink you could use to actually take the measurement, as the points are right next to the cpu socket..

[saaya]

Ok im confused.

How do you know what GTL voltage you want in the first place? Other peoples settings? Black Magic/Voodoo?

Thats my exact question as well... I assume we just need trial and error and see which works best like any other voltage i guess?

yes, every cpu looks a slightly different GTL ref for each core, and every board varies slightly, so its a trial and error thing. but you can learn a lot from looking at other peoples results since all 45nm dualcore cpus are about the same, all 45nm quads are about the same etc

ill see if i can collect some settings from people to give you guys some more guidance.

but its called fine-tuning, if there would be a way to know what setting is best for your cpu, then the bios would configure this automatically and there wouldnt be a need to fine tune
btw, i dont think any other board allows this fine steps actually of only 0.007v
i remember my p5e3 had 4 options for cpu GTL, and it was applied for all 4 cores

Also, your spreadsheet says "enter current VTT setting or measured VTT in highlighted field" but you said that the vtt settings is +0.10v higher than the real vtt, so won't this affect the calculation?

sure, you gotta keep in mind that real vtt is lower than what it says in bios.
so if you set 1.2v in bios, enter 1.1 in the spreadsheet to see which GTL Ref ratio results in what GTL Ref voltage.

Basically this lets you guys adjust the FSB signal window.
you can increase the amplitude by increasing vtt, and you can adjust the center of the wave signal to compensate for the signal rising or falling too fast, at least thats how i understood GTL Ref so far

Well the attachment documents measuring points for reading the voltage with a multimeter, although I'm not sure what kind of heatsink you could use to actually take the measurement, as the points are right next to the cpu socket..

temps were actually quite ok as you can see

but yeah, joking aside, you need a slim heatsink like the TRUE or a waterblock. or you solder wires to it:


actually measuring gtl ref all the time doesnt make that much sense, every step is 7mv, so all you need to check is what the default voltages with your cpu and your board are at stock speeds with stock vtt and with increased vtt, and from that you know all you need to know already. so you can put a small heatsink on there just to meassure the voltages with a DMM and then after your done taking your notes, put the big heatsink back on there

[Ryan12]

Makes sense, thanks heaps saaya... Never seen good support support from a rep on forums before.. hopefully it will help me tune in me E8400 to some good clock speeds with lower voltages!

[RAYTTK]

@ Negev love the black magic voodoo line pissed myself laughing,
@ saaya thanks for taking some time for this.
Sure there are a few tutorials about what GTLS do but Blackops relevent GTL refs are whats required

[saaya]

Makes sense, thanks heaps saaya... Never seen good support support from a rep on forums before.. hopefully it will help me tune in me E8400 to some good clock speeds with lower voltages!

your welcome
ive personally seen litle gain from GTLref tweaking, but some people got very nice results with it and tweaked an extra 20mhz fsb on quads out of their boards

@ saaya thanks for taking some time for this.
Sure there are a few tutorials about what GTLS do but Blackops relevent GTL refs are whats required

mhhh what do you mean?
GTL Ref is all the same on any board, we just have more options to fine tune it than other boards

heres a good guide, although a bit confusing:
http://www.edgeofstability.com/artic.../gtl/gtl1.html

[Ryan12]

Where everywhere i read people just keep saying up the voltages blah blah blah... Where just playing around with aegis panel and prime95 for hours... It's totally the right mixture of voltages rather then just MORE volts... Theres someone on here that keeps correcting ppl to say its the right 'mixture' and now im getting somewhere with this board, but i must say its a good learning curve
Also - does the south bridge volts change alot, what does the volts actually reference to?

[Xello]

Hey guys,

Been playing with Blackops and GTL Ref settings today and studdied the docs from Intel a bit. Its 10pm and i need to get some sleep, some quick points.

[LIST=1][*]Core0 and Core2 are in the same chip? can anybody confirm?[*]Core1 and Core3 are in the same chip? can anybody confirm?

Sounds right, 0 and 1 are the first cores on each die.

[*]the BIOS maps VTT incorrectly! default vtt for 45nm cpus is 1.10v, which is applied by default, but its displayed as 1.20v. so when you use a 45nm cpu and set vtt to 1.30v its actually only 1.20v. That makes some fsb overclocks ive seen here even more amazing since they were actually with .10v vtt less!

Always wondered about this, it agrees with Raja too who told me 1.35v vtt gives 1.27v real.

[saaya]

Also - does the south bridge volts change alot, what does the volts actually reference to?

its the main sb supply voltage... as i said before, ive never seen this influence anything but people pushing high fsbs said it helps, and people running raid said it helps them when ocing which makes sense...

Sounds right, 0 and 1 are the first cores on each die.
Always wondered about this, it agrees with Raja too who told me 1.35v vtt gives 1.27v real.

ok, thanks for confirming!
and yes, im a bit annoyed that vtt thing was fixed already

FYI, i meassured CPU GTL REF voltages with an 8600 and created a table with the ACTUAL GTL REF dividers:
http://www.xtremesystems.org/forums/...d.php?t=201022

ill check with a quad tomorrow

[RAYTTK]

mhhh what do you mean?
GTL Ref is all the same on any board, we just have more options to fine tune it than other boards

I mean some of the guides and forum threads ive read have been for ASUS 790i or have had more numbers in the equation, like the black ops stops at 0.740 but others go as high as 0.905 and TBH ive read that much its turned my brain to mush. I drive a chemical tanker truck for a living so going to work at 5am & tweaking this board till 12pm is making me sleep deprived lol. Im still hoping for a simple if you put VTT @ x volts you need to set GTL's to + whatever
What quadcore are you going to use?

[negev]

Also very interested in quadcore results.

What's interesting in the graph that HDCHOPPER posted (in the other thread about the same subject) is that the margin for error in GTL refs drops very close to 0 as you approach 1800Mhz fsb, but seems to bounce back up again after it... this tends to suggest this it would be very difficult to tune GTL refs to 1800Mhz but easier at slightly higher speeds..

[RAYTTK]

What's interesting in the graph that HDCHOPPER posted (in the other thread about the same subject) is that the margin for error in GTL refs drops very close to 0 as you approach 1800Mhz fsb, but seems to bounce back up again after it... this tends to suggest this it would be very difficult to tune GTL refs to 1800Mhz but easier at slightly higher speeds

I cant get past 450 FSB on this QX with an 8x multi, If I up to 9 its a no go but maybe I should try 10 x 400 FSB and set all the GTL;s to +16? what do you think?

[negev]

I cant get past 450 FSB on this QX with an 8x multi, If I up to 9 its a no go but maybe I should try 10 x 400 FSB and set all the GTL;s to +16? what do you think?

I would say the Blackops SAS team will crash through your window and assassinate everyone in sight if you try this.

Seriously, I have no idea, I was just commenting on the graph. The curve seems to go up at 1800Mhz, but I doubt it goes up very high before coming back down even lower... perhaps slightly above 1800Mhz is a good spot. This is entirely theoretical - if you cook your cpu or accidentally create a new self-replicating army of warrior blackops replicators that spit poisonous venom and sing Queen songs I am not taking responsibility or running into any dangerous areas with a naquada bomb.

[RAYTTK]

Tried It, It didnt like it
Reset OC Gear. Had enough now watching tropic thunder

[Sailindawg]

It's totally the right mixture of voltages rather then just MORE volts

I think that's a pretty succinct statement regarding GTL to stabilize an overclock. After reading forum posts and other articles, that's pretty much sums up how I have begun to look at using GTL tuning. It's a very time consuming and slower approach, but an approach that should yield stable clocks without over volting the cpu unnecessarily.

@Saaya

ill check with a quad tomorrow

Any chance you could run a 65 nm quad as well?

I cant get past 450 FSB on this QX with an 8x multi, If I up to 9 its a no go but maybe I should try 10 x 400 FSB and set all the GTL;s to +16?

I'm running a Q6600 and I'm currently priming away. I'm running at 9x380, using 1.3375 Vcpu (BIOS) and have my GTL's set at +14 for core 0&2 and +16 for cores 1&3. In my case running all cores at +13, while stable for 9x363, @ 380 core 3 would fail within 15 minutes. Setting core 3 to +14, would get one hour, +15 would make core 0 fail after 4 hours. I ended up setting core 0 & 3 both to +16 and I'm passing P95 for 3 hours so far.

The funny thing is that I can pass Intel Burn test 20 passes with GTL's all set to +14. I like to be P95 stable for at least 7 hours. I have never had a problem overclocking using that stability criteria.

The one question I have, is how high is too high for a GTL setting?

I understand that VTT should be much over 1.4V, but not so sure how high GTL's can go. I'm being stingy on the voltage purposefully to try to tune stability with GTL's.

[saaya]

I mean some of the guides and forum threads ive read have been for ASUS 790i or have had more numbers in the equation, like the black ops stops at 0.740 but others go as high as 0.905 and TBH ive read that much its turned my brain to mush. I drive a chemical tanker truck for a living so going to work at 5am & tweaking this board till 12pm is making me sleep deprived lol. Im still hoping for a simple if you put VTT @ x volts you need to set GTL's to + whatever
What quadcore are you going to use?

ive seen those settings, and those 0.905 settings are actually the voltage, not the ratio afaik.
and sure, id like to shed some light on this as well to make gtl ref tweaking easier
just be patient and ill continue to udpate this thread with my findings

ill try as many cpus as i can get ahold of

Also very interested in quadcore results.

What's interesting in the graph that HDCHOPPER posted (in the other thread about the same subject) is that the margin for error in GTL refs drops very close to 0 as you approach 1800Mhz fsb, but seems to bounce back up again after it... this tends to suggest this it would be very difficult to tune GTL refs to 1800Mhz but easier at slightly higher speeds..

interesting... can you link me?

I cant get past 450 FSB on this QX with an 8x multi, If I up to 9 its a no go but maybe I should try 10 x 400 FSB and set all the GTL;s to +16? what do you think?

whats the highest fsb with 9x multi for you?

I understand that VTT should be much over 1.4V, but not so sure how high GTL's can go. I'm being stingy on the voltage purposefully to try to tune stability with GTL's.

VTT 1.55v wouldnt boot my 8600
VTT 1.50v works, but i havent had it there for a long time yet
Regarding GTLs, i have no idea... there is no MAX rating for GTL voltages and i havent heard of anbody killing their cpus with high GTL...

but come to think of it... some people actually killes their cpus with relatively low vtt... maybe they tweaked their gtls to high volts and THAT actually killed their cpus and not the VTT?
ill play around some more and will see what i can find.

in any case, i think its unlikely that gtl will kill a cpu...
im sure its possible, but i havent heard of cpus scaling with higher and higher gtl volts. a few steps up or down is usually all you need, so theres no sense in going really high or low which might damage the cpu...

[RAYTTK]

whats the highest fsb with 9x multi for you?

9 X 433 is the highest, Ram @ 1734mhz 7's pl7

[negev]

interesting... can you link me?

http://www.thetechrepository.com/att...1&d=1169078346

[Praz]

http://www.thetechrepository.com/att...1&d=1169078346

That graph is an example of how the margin can change as FSB changes. It's not a literal representation. Because of reflections, harmonics, jitter and countless other influences the error window at one FSB setting may be completely different at another setting.

Saaya

Really nice thing you're doing here for the users of this board. Definitely raises the bar of manufacturer support.

[saaya]

9 X 433 is the highest, Ram @ 1734mhz 7's pl7

but with 10x multi the max fsb is suddenly only 380?
then its def not the fsb holding you back but the cpu itself...
but how come you got it stable 600mhz higher on a more basic board with a weaker analog pwm? that doesnt make any sense at all...
are you sure your using the same vcore and all?
what vcore did you use on the gigabyte p35 and what vcore on blackops?
could it be that the gigabyte board overvolted vcore?
even then a 600mhz difference is very weird...

http://www.thetechrepository.com/att...1&d=1169078346

ahhh yeah i saw that one on blazingpc... very interesting...

That graph is an example of how the margin can change as FSB changes. It's not a literal representation. Because of reflections, harmonics, jitter and countless other influences the error window at one FSB setting may be completely different at another setting.

Saaya

Really nice thing you're doing here for the users of this board. Definitely raises the bar of manufacturer support.

thanks, doing my best
yeah, a lot of things play into this, but there should be common trends, and im pretty sure the 450fsb thing is a common trend.
im really surprised how the tolrance level increases beyond 450fsb though... could it have to do with a higher PL setting at those high fsb speeds maybe?

[RAYTTK]

but with 10x multi the max fsb is suddenly only 380?
then its def not the fsb holding you back but the cpu itself...
but how come you got it stable 600mhz higher on a more basic board with a weaker analog pwm? that doesnt make any sense at all...
are you sure your using the same vcore and all?
what vcore did you use on the gigabyte p35 and what vcore on blackops?
could it be that the gigabyte board overvolted vcore?
even then a 600mhz difference is very weird...

FSB is changing with the mem dimms ie, last night managed 9 x 440 on white dimms. The reason I tried the Gigabyte was to determine if it was the cpu holding me back, which is what i thought so I was very surprised to get 4.2 on it, Vcore 1.5v NB 1.63 which yes ive tried on blackops. I thought maybe it was a low ram 800mhz on P35 that helped so i have tried 800 mhz a stick in blackops but no luck . Look Im not saying the Gigabytes a better board you know It was red hot & totally insufficient for 24/7. I think this chip just isnt that good a clocker past the 3.7 mark. Im running 11 x 340 now with 4gb of ram just going to chill for a week or so play some games etc. Eagerly await your results for Quadcores, Meanwhile i keep checking ebay for a cheap E8500.
Again I thank you for continued effort in helping make sense of the GTL;s

[saaya]

your welcome
hmmm well im still puzzled how you got a way higher cpu clock on that board than the blackops, which has a way beefer, faster clocked pwm... hmmm
if it would be fsb, ok, i could imagine that board has a better default setting for gtl settings or higher vtt by default etc, but higher cpu clocks is a thing that depends on the pwm, and maybe BIOS support... hmmm maybe thats it, BIOS support for 65nm hmmmm but im surprised if it would make a 600mhz difference... very weird... ill check with the BIOS engineers

[negev]

any news on 45nm quad results yet saaya?

I'm trying to get to 3.5ghz, and I think GTL refs need adjusting, either that or my cpu just wont go that far