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

[mikeyakame]

In my personal experience Vtt needs to be higher than Vcc but only by 0.02v or so. If you raise Vtt the best thing to do is drop CPU GTLREF to around 0.62 and go from there. Raise Vtt lower GTL, drop VTT raise GTL. Sometimes you need to do both, But 470 FSB on X48 doesnt need anywhere near that Vnb on PL8. Which Divider are you using? Helps if you give that because not all dividers are stable at high multiplier. I have same problem low multi no issues, raise it and breaks most of the dividers. You have same sticks as I do, try use 12:10 divider I find its best personally if you arent already. Half Multis might not be all that stable at higher FSB too. You need to take that into account if possible. Try a full multiplier and see if same problem occurs.

With Vnb I use 1.49v LLC enabled PL7 on 12:10 divider at 481mhz. 0.63x NB GTL.

[spoof]

Maybe i'll try to lower vNB, honestly didn't try this, but it will help just with the heat, thas chipset is generation, won't help with the lockups. PL is set to 8, but i think i can lower it direct or through Pull-In's, but this is when everything is stable. For the NB GTL i have interesting problem - it can not boot while using lower GTLs than 63x and higher than 57x (i'm no sure, but between this two i have 61x 60x and one more). No just this, but using different than 67x cause my RAM to fail (or it's NB fail, but i get mass error detection it Memtest). As for the memory divider i use 12:10 too, the only way to achieve my rated speed of 1100Mhz. The most strangest thing is that when i try 1:1 it refuses to boot. Dunno why, this two boot problems, connected with the chipset are very strange for me. Apparently the problem here is in CPU, but what exactly and why, no idea.

Now when i think a bit - i had 1.42v VTT and 1.44 vCPU and it locked up, than i lowered vCPU to 1.3* and now i have no problem for 30 minutes (mostly, when it's going to lockup it take couple of minutes). I realy have to test this, thank you!

[Alien Grey]

What i don't understand is why someone can use low VTT and others don't? Is there an explanation for this?

[mikeyakame]

Depends on what Vcc and Vnb you are using! If using low voltages you don't need high Vtt! Only need high Vtt for high Vcc/Vcore.

[ci2kla]

I think that's what I'm working with. Yesterday I was completely lost as original setting won't stable anymore~
Hence I reset and start from where I begin. What I'm playing right now is clock skew. By using different clock skews in the same FSB, vNB GTL sweet spot will change.
But due to lack of time these testing still don't have a strong clue. Will post more about that soon.

Current Progress: VTT 1.37v, vCore 1.2v, FSB 471Mhz, CPU Skew: 400ps, NB Skew: 100ps, vNB GTL: +60mv, other CPU GTLs temp. on auto

[CryptiK]

Yes NB GTL Ref sweetspot changes when other parameters are changed, and it can be a very influential setting. Also, I found that I was Linpack/Orthos/Memtest stable at a certain vNB, but in games I would sometimes crash to desktop as if I was very unstable. Crysis was the best at doing this, sometimes it would only last 30 seconds before it crashed. Even increasing vNB from 1.30v to 1.45v did nothing to help this situation, nor did changing clock skews ram skews, or NB DDR Ref's or CHA/CHB Refs. I finally dropped my NB GTL Ref to -100mv, and although it only passed 2/6 tests in linpack, it was absolutely fine in crysis, no more crashes. I then bumped it up to -80mv, and that made the system totally linpack etc stable again, and was still totally stable in crysis.

I have found on my M2F, I really only need to use cpu/NB clock skew when I am over 540 - 550 FSB. Behaviour is different with quads though, some guys need to start using the skews around 450 FSB.

[mikeyakame]

Hey cryp!

If you get bored would you care to try setting CPU Margin Enhancement to Performance and see if you need any less CPU/NB clock skew with it set to that?

[Alien Grey]

Depends on what Vcc and Vnb you are using! If using low voltages you don't need high Vtt! Only need high Vtt for high Vcc/Vcore.

What if i don't have to use high Vcc and need high Vnb for that FSB?

It seems that i can't change the NB GTL from 0.67X to a lower multiplier.

Is this the point that you talk about where you have to use a CPU/NB Clock Skew to keep stability?

[mikeyakame]

Can be yes! Yeah with such a high Vnb it makes it hard to not use high Vtt. It's a double edged blade unfortunately. This is really where we near clock skew adjustments. High Vtt is just simply compensation for GTL/Skew related issues. I need a lot less Vtt on a P45 board than X48 just because I can set up skews and GTLs very finely. 475FSB on a P45 with 65nm chip I can get away with 1.26v Vtt with NB clock skew around 100-200ps, CPU clock skew around 400-500ps, same FSB on a X48 I need 1.47V Vtt real for same stability.

[CryptiK]

Hey cryp!

If you get bored would you care to try setting CPU Margin Enhancement to Performance and see if you need any less CPU/NB clock skew with it set to that?

I would but I am using my M2F which doesn't have that setting. I was doing some testing with the P5Q-D last night (which does have that setting) but it's back in the box now.

With older bioses I found it didnt really have any effect on anything, but it may have been poorly implemented in earlier versions. It really seems to affect quads more than duals, but more in the sense of simply providing stability, as it was causing quad users some major hassles if set incorrectly before BIOS 1406 IIRC (I only tried it with BIOS 1201).

[mikeyakame]

I would but I am using my M2F which doesn't have that setting. I was doing some testing with the P5Q-D last night (which does have that setting) but it's back in the box now.

With older bioses I found it didnt really have any effect on anything, but it may have been poorly implemented in earlier versions. It really seems to affect quads more than duals, but more in the sense of simply providing stability, as it was causing quad users some major hassles if set incorrectly before BIOS 1406 IIRC (I only tried it with BIOS 1201).

That makes sense indeed. I found same problem so I can only assume it affects clock skews in a major way. I'm surprised M2F doesn't have that option! Another thing is Mem OC Charger, I think that forcefully sets DRAM Clock Skews as I remember with it enabled the DRAM Clock skews for all slots were all locked when checked with MCHBAR dump or Everest.

[CryptiK]

I think that the options was designed to aid overclocking of low end cpu's (think e4300 etc). However being Asus' top of the line P45 board, the M2F was probably not expected to be paired up with low end chips, hence the option is not present (that's my take on the situation anyway).

Interesting what you said about Mem OC Charger, I always assumed it was asus-ese for ram drive strength. With my M2F, I cannot overclock the ram over 605 MHz with it set to disabled, but with it enabled, I can bench at around 650 MHz. Also with it enabled my D9's are stable at 600 MHz 5-5-5-15 @ 1.98v, whereas they need around 2.05v with it disabled. So you are saying that when it's enabled, ram clock skews when altered have no effect in reality as it presets & locks them? Any more info you can give regarding that option would be appreciated.

[mikeyakame]

I think that the options was designed to aid overclocking of low end cpu's (think e4300 etc). However being Asus' top of the line P45 board, the M2F was probably not expected to be paired up with low end chips, hence the option is not present (that's my take on the situation anyway).

Interesting what you said about Mem OC Charger, I always assumed it was asus-ese for ram drive strength. With my M2F, I cannot overclock the ram over 605 MHz with it set to disabled, but with it enabled, I can bench at around 650 MHz. Also with it enabled my D9's are stable at 600 MHz 5-5-5-15 @ 1.98v, whereas they need around 2.05v with it disabled. So you are saying that when it's enabled, ram clock skews when altered have no effect in reality as it presets & locks them? Any more info you can give regarding that option would be appreciated.

I haven't really checked with it disabled. But on P5Q deluxe with it enabled I see Dimm1/2/3/4 Fine Clock Delay at 6/2/2/4 or so. CHanging DRAM Clock Skews for any dimm slot doesn't effect those values what so ever. Those values correspond to MCHBAR registers for DRAM Fine Clock Delay which DRAM Clock skew is supposed to set as well as DRAM Address Fine Clock Delay. It's odd to say the least. Don't have one at hand to check so can't say for certain.

I wish Asus instead of labelling options with stupid names would actually call them the right thing and explain what they change. Would make life of an overclocker so much easier.

[CryptiK]

With it disabled and with ram skews on 'normal' I have Dimm 1 - 4 fine clock delay of 6-2-2-4 respectively. IIRC changing ram skews had no effect on them even when Mem OC Charger was disabled, but I may be mistaken and will re-examine this later. Changing Ai Clock Twister settings also did not effect them. The boost in ram overclocking speed and reduction in required vdimm at any given speed with it enabled led me to conclude it was drive strength related, and as such I run it disabled 24/7, but I'd really like to know exactly what it does. I also really dislike the way asus name BIOS options, perhaps they think it sounds more appealing, but come on, we have already bought the board, they shouldn't still be using marketing strategies in the BIOS of all places.

[mikeyakame]

I personally like how DFI name them. The right names and do exactly what they are named to do! Weird what you just said about them never changing. That has me thinking what the hell. Mem OC charger might do just what you think it does, with DQ drive strength stabilization. I think we'll never know what their options do exactly till we can figure out the nature of AMI 1B bios module and how to locate instructions for options.

[andest2003]

i have a q9650 on asus striker 2 extreme

if my vcore is 1.38 vcore idle 1.38 load 1.33
and vtt 1.22

i would work this out at

1.22 x 0.635 = 0.775
1.22 x 0.667 = 0.814

0.775 - 0.814 = 0.039

as its a 45nm cpu i would look for closest setting i think are -40 and -36

-40
-36
-40
-36

then nb gtl i would set to -40

is this correct

as i change vtt the calculation would change

i need to under stand this to get better clocks but makes my head hurt lol

[mikeyakame]

Yes as you change either Vtt or any voltage which uses Vtt as a termination for logical high you change GTLs. heh It's a tough one.

I'm personally not at all familiar with Nvidia's implementation of Vtt on their chipsets so I can't be of much help. It should be how Intel says to implement it for Vcc but I am not certain since NV doesn't use FSB rather it uses A hyper-transport like bus similar to AMD for Intel chips.

[spoof]

Still nothing here - i tried almost everything: lower VTT, higher VTT, vCPU, vNB, GTLs, PLL, combination, tried vCPU and VTT with 0.2 difference - still suffeting from hard lockups. Somethink interestin is, that i managed to lower my vCPU, but still this damn lockups. I'm out of any ideas. 471FSB rockstable with 6x multiplier, just rise it to 8.5x and everything goes to hell.

[Alien Grey]

Yes NB GTL Ref sweetspot changes when other parameters are changed, and it can be a very influential setting. Also, I found that I was Linpack/Orthos/Memtest stable at a certain vNB, but in games I would sometimes crash to desktop as if I was very unstable. Crysis was the best at doing this, sometimes it would only last 30 seconds before it crashed. Even increasing vNB from 1.30v to 1.45v did nothing to help this situation, nor did changing clock skews ram skews, or NB DDR Ref's or CHA/CHB Refs. I finally dropped my NB GTL Ref to -100mv, and although it only passed 2/6 tests in linpack, it was absolutely fine in crysis, no more crashes. I then bumped it up to -80mv, and that made the system totally linpack etc stable again, and was still totally stable in crysis.

I have found on my M2F, I really only need to use cpu/NB clock skew when I am over 540 - 550 FSB. Behaviour is different with quads though, some guys need to start using the skews around 450 FSB.

That's the reason why i stress the whole system. Before i used UT2004 with Prime95 i had something similar like you've been experiencing with Crysis.
My CPU is at 100% load all the time with BOINC, and before i used UT2004 together with Prime95 a 3D application crashed within minutes.
The idea to use UT2004 together with Prime95 is not my idea. Some guy posted it on a forum, don't know what forum it was, to do it like this to find full system stability.

Like i said before, i do this to find the right VTT. With the wrong VTT UT2004 will crash sooner or later while Prime95 is still running without errors.

[mikeyakame]

Cryptik!

I have a mission for you if you choose to accept it How would you feel about doing some MCHBAR dumps for me

http://www.tweakers.fr/baredit.html

You want this software to save the dumps.

All you need to do if you feel like it is this.

Change NB and CPU clock skews individual of each other without changing any other options in bios (that is the critical part), and boot into Windows load up Baredit, set MCHBAR to 32bit and click bar.txt.

CHange the file name to represent what setting you used. Do this 2 or 3 boots for each setting, and save dump each time.

Try a few different values if possible.

Ie
NB Normal, CPU 200PS
NB 100ps, CPU 200ps
NB Normal, CPU Normal

etc

Critical part is you remember which is which ! hehe. The 2 or 3 boots and dumps for same settings is to just compare against each to figure out which registers I don't recognize that are incrementing or not relevant. More dumps I have to analyze and compare the easier it will be to find what I'm looking for.

THis is just a hunch I have that NB & CPU clock skews are values stored in MCHBAR registers and not CPU MSRs or PLL Clock Gen registers.

Don't need to rush with it, and don't worry if a setting wont post just try another. Don't kill your system doing this either not even for sake of results! heh.

If its too much to ask just let me know, take all the time you need after all it's only time.

What I would like to achieve is being able to adjust these values dynamically inside Windows to aid with tuning provided they are within MCHBAR and they arent RWO (read write once) registers which require PWROK to clear.

By the way mate if you are interested in meddling with DRAM clock skews inside Windows let me know and I'll explain to you which MCHBAR register offsets to read and write to change them Baredit allows that of course too thanks to Felix's awesome handywork.

[CryptiK]

Yeah mate for sure I'll do that, sounds very interesting I have read a bit about it but I have not got into it. That program if used as you directed is not going to cause any problems will it? I assume it just reads and dumps into a text file. So I open it, select 32bit (under configuration space & mchbar drop down boxes) then click bar.txt button. Where does the txt file get saved to?

[spoof]

So far nothing - lockup or reboot, all day long. Now i decided to do this - i set FSB to 470 and multi to 6x and start to lower all the voltages. Of course, it lockup while booting windows, what isn't new. The first repeatable sing, that i found is that i lockups just before entering LogOn screen, when 3 green light on the keyboard are flashing all 3 at a time. Lowering voltages doesn't make it lockup earlyer or later. So far i'm at 1.28 vCPU, 1.41 vNB, 1.5 CPU PLL, 1.3 VTT and going down. GTLs are 65x for CPU and 67x for NB.

So, i didn't do anything with finding max FSB and then testing the cpu. I'll set all voltages to minimum, 333x8.5 and i'll rising 10mhz at i time and 1 step on required voltages at a time. That's all i can do ...
Question - how to set GTLs just to be sure, that they aren't too high? From testing so far 65x on the CPU and 67x on the NB give the most stability, but since now i'll be going from down to up, may be i need to set them to min available values - 62x CPU GTL and 52x NB GTL? Any help here?

[mikeyakame]

Yeah mate for sure I'll do that, sounds very interesting I have read a bit about it but I have not got into it. That program if used as you directed is not going to cause any problems will it? I assume it just reads and dumps into a text file. So I open it, select 32bit (under configuration space & mchbar drop down boxes) then click bar.txt button. Where does the txt file get saved to?

wherever you save it, default is baredit install folder!

Nah it wont cause any probs for reading. It just reads pci register space is all!

[mskvorc1]

Looks like I don't have much luck with my E8500. Since I have EVGA 750i motherboard I only have positive values to set up. Tried with 40/0/40/0 and 50/10/50/10 with 1.35V in BIOS, actually 1.336V in load at 9x445 with FSB VTT at 1.2V (1.18V Everest says) and nForce MCP 1.40V, but Prime crashed with first core after 2 minutes...

With VTT 1.2V I get -0.038V that's way i tried combinations 40/0/40/0 and 50/50/50/10 for GTL ref values...

[CryptiK]

wherever you save it