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

[LuckySevn]

try 1.46/1.48v vtt with 0.63 - 0.65x for cpu, and nb try 0.61-0.63x
you'll just have to find the sweet spot for your fsb frequency and your particular board.

normally i find if i use the right vtt voltage, then i can just set 0.65x for cpu and 0.63x for nb!

Thanks for the heads up man, I'll try it and report back! =D

[vox]

Hello ppll!

After some weeks of test to get my settings prime aproved i can only do Small ffts 15mts and sometimes more... but i cant get my Volts lower than this need your help!!!

Q9550 Asus P5Q3 DLx Gskill 2x 2gb DDR3 1333 CL7

CPU Ratio Setting: 8.5
FSB Frequency: 475
PCIE Frequency: 100
FSB Strap To North Bridge: AUTO
DRAM Frequency: 1584 !!!!
DRAM CLK Skew on Channel A1: Auto
DRAM CLK Skew on Channel A2: Auto
DRAM CLK Skew on Channel B1: Auto
DRAM CLK Skew on Channel B2: Auto
DRAM Timing Control: Manual !!!!

CAS Latency: 7
DRAM RAS to CAS Delay: 7
DRAM RAS Precharge: 7
DRAM RAS Activate to Precharge: 18
RAS TO RAS Delay: 6
Row Refresh Cycle Time: 60
Write Recovery Time: 10
Read To Precharge Time: 6
Read to Write Delay: Auto
Write to Read (s): 6
Write to Read (d): 6
Read to Read (s): Auto
Read to Read (d): Auto
Write to Write(s): Auto
Write to Write(d): Auto
Write to Pre Delay: Auto
Read to Pre Delay: 6
Pre to Pre Delay: Auto
All Pre to Act Delay: Auto
All Pre to REF Delay: Auto
DRAM Static Read Control: Disabled
DRAM Dynamic Write Control: Disabled
DRAM Read Training: Disabled
DRAM Write Training: Disabled
MEM OC Charger: DISABLE
Ai Clock Twister: auto
AI Transaction Booster: AUTO

CPU Voltage: 1.40
CPU GTL Voltage Reference (0/2): Auto
CPU GTL Voltage Reference (1/3): Auto
CPU PLL Voltage: 1.70
FSB Termination Voltage: 1.36
DRAM Voltage: 1.94
NB Voltage: 1.32
NB GTL Voltage: Auto
SB Voltage: 1.20
PCIE SATA Voltage: 1.60

Load-Line Calibration: Enabled
CPU Spread Spectrum: Disabled
PCIE Spread Spectrum: Disabled
CPU Clock Skew: AUTO
NB Clock Skew: AUTO

Before i get to other values i want to get Prime Stable on this.

Any input helps! tks!

vox

[spoof]

Hmm ok. Try using 1.3725 vcore, 1.22 Vtt, 0.67x GTL Ref, 1.55 vNB at 455 FSB. Also try advancing you ram skews to around 250 - 300ps if you have them.

Also, can you raise GTL Ref above 0.67x? I have seen some guys with C0 stepping 9550's needing around the 0.68x - 0.73x mark for 4GHz, but some also need around 1.42v load for stability at 4GHz.

Ok, i have this - 1.4 vCPU, 1.54 CPU PLL, 1.36FSBt with x67GTLREF, vNB 1.57, DRAM CLK Screw A/B to Normal.
So, the idea was, that with more vNB, since i have 4 RAM blocks, i'll fix the mo****f****** lockup. Well, it still do it - just freeze. Is there anything more what i can do? What about SB voltage?

[CryptiK]

Hello ppll!

After some weeks of test to get my settings prime aproved i can only do Small ffts 15mts and sometimes more... but i cant get my Volts lower than this need your help!!!

Q9550 Asus P5Q3 DLx Gskill 2x 2gb DDR3 1333 CL7

CPU Ratio Setting: 8.5
FSB Frequency: 475
PCIE Frequency: 100
FSB Strap To North Bridge: AUTO
DRAM Frequency: 1584 !!!!
DRAM CLK Skew on Channel A1: Auto
DRAM CLK Skew on Channel A2: Auto
DRAM CLK Skew on Channel B1: Auto
DRAM CLK Skew on Channel B2: Auto
DRAM Timing Control: Manual !!!!

CAS Latency: 7
DRAM RAS to CAS Delay: 7
DRAM RAS Precharge: 7
DRAM RAS Activate to Precharge: 18
RAS TO RAS Delay: 6
Row Refresh Cycle Time: 60
Write Recovery Time: 10
Read To Precharge Time: 6
Read to Write Delay: Auto
Write to Read (s): 6
Write to Read (d): 6
Read to Read (s): Auto
Read to Read (d): Auto
Write to Write(s): Auto
Write to Write(d): Auto
Write to Pre Delay: Auto
Read to Pre Delay: 6
Pre to Pre Delay: Auto
All Pre to Act Delay: Auto
All Pre to REF Delay: Auto
DRAM Static Read Control: Disabled
DRAM Dynamic Write Control: Disabled
DRAM Read Training: Disabled
DRAM Write Training: Disabled
MEM OC Charger: DISABLE
Ai Clock Twister: auto
AI Transaction Booster: AUTO

CPU Voltage: 1.40
CPU GTL Voltage Reference (0/2): Auto
CPU GTL Voltage Reference (1/3): Auto
CPU PLL Voltage: 1.70
FSB Termination Voltage: 1.36
DRAM Voltage: 1.94
NB Voltage: 1.32
NB GTL Voltage: Auto
SB Voltage: 1.20
PCIE SATA Voltage: 1.60

Load-Line Calibration: Enabled
CPU Spread Spectrum: Disabled
PCIE Spread Spectrum: Disabled
CPU Clock Skew: AUTO
NB Clock Skew: AUTO

Before i get to other values i want to get Prime Stable on this.

Any input helps! tks!

vox

If that is not an E0 stepping, good luck going higher on that voltage, most top out at around 4GHz @ ~1.42v (real load). You can just try various GTL Refs, start with 0.63x on each one and move up to ~ 0.70x on each one. Whichever gives the longest time in orthos without an error is the best setting. However if you can only pass 15mins without an error, I doubt you can get 100% stable with GTL tweaking. id say your cpu is at its limit.

However you could try increasing the vNB, 1.32 is pretty low for DDR3 and that FSB with a quad. Those boards aren't really built for it so you may need to push it. Try 1.40 vNB.

I also assume you have tested the cpu & ram/NB separately using small FFT, large FFT, bootable memtest & windows memtest? What do you pass/fail? Prime is not the only test you need to run, and even with prime you need to run small & large FFT's. Try to give us as much info as possible or else its very hard to know what to suggest to you ok.

Ok, i have this - 1.4 vCPU, 1.54 CPU PLL, 1.36FSBt with x67GTLREF, vNB 1.57, DRAM CLK Screw A/B to Normal.
So, the idea was, that with more vNB, since i have 4 RAM blocks, i'll fix the mo****f****** lockup. Well, it still do it - just freeze. Is there anything more what i can do? What about SB voltage?

Try pulling out 2 ram modules and see if that makes a difference. if it does its your NB at fault.

Have you tested the ram/NB with bootable memtest & windows memtest?

[LuckySevn]

try 1.46/1.48v vtt with 0.63 - 0.65x for cpu, and nb try 0.61-0.63x
you'll just have to find the sweet spot for your fsb frequency and your particular board.

normally i find if i use the right vtt voltage, then i can just set 0.65x for cpu and 0.63x for nb!

Hmm...tried VTT 1.28v ~ 1.52v (BIOS) with x0.62 ~ x0.67 for CPU; x0.57~x0.67 for NB.

Orthos fails under 3min if CPU GTL set to x0.63 and up, only x0.62 gives me the longest times. For NB anything below x0.60 does not boot at all. x0.63 seems to gives me the longest times in Orthos but still fails with any VTT, vCore and PLL setting. Pushed and lowered NB 1.35v ~ 1.55v (BIOS) to no avail.....

Oh yea and I think the degradation has started....just 3 days ago 4.45GHz was stable with 1.425v vCore, ran for over a month without problems and was passing Orthos/Prime for 3~4 hours. Then it started needing more volt gradually, until now, it needs 1.4625v to boot; 1.475v to get in Vista; and 1.4815v to be stable to run Crysis...

[truehighroller]

Hmm...tried VTT 1.28v ~ 1.52v (BIOS) with x0.62 ~ x0.67 for CPU; x0.57~x0.67 for NB.

Orthos fails under 3min if CPU GTL set to x0.63 and up, only x0.62 gives me the longest times. For NB anything below x0.60 does not boot at all. x0.63 seems to gives me the longest times in Orthos but still fails with any VTT, vCore and PLL setting. Pushed and lowered NB 1.35v ~ 1.55v (BIOS) to no avail.....

Oh yea and I think the degradation has started....just 3 days ago 4.45GHz was stable with 1.425v vCore, ran for over a month without problems and was passing Orthos/Prime for 3~4 hours. Then it started needing more volt gradually, until now, it needs 1.4625v to boot; 1.475v to get in Vista; and 1.4815v to be stable to run Crysis...

Throw the chip back to the wolves and get another one.

[gugle]

how I can to improve these values?
PLL, GTL's, vNB

Q6600 VID 1.2500v
450x8 - 3600mhz
4x1Gb DDR2- 1080

CPU Voltage	1,41250v
CPU GTL Voltage Reference (0/2) 0.630 -- correct?
CPU GTL Voltage Reference (1/3) 0.670 -- correct?
CPU PLL Voltage	1,52
FSB Termination Voltage	1,28
DRAM Voltage 2.10
NB Voltage	1.18
NB GTL Reference 0.630 -- correct?
SB Voltage	1.10
PCIE Voltage 1.50
____________________________________	
	
Load-Line Calibration	[Enabled]
CPU Spread Spectrum	[Disabled]
PCIE Spread Spectrum	[Disabled]
CPU Clock Skew	[Auto]
NB Clock Skew	[Auto]
CPU Margin Enhancement [Optimized Mode]
____________________________________	
	
Advance CPU Settings:	
CPU Ratio Setting:	
C1E Support:	[Disabled]
Max CPUID Value Limit:	[Disabled]
Intel Virtualization Tech:	[Disabled]
CPU TM Function:	[Disabled]
Execute Disable Bit:	[Disabled]

[spoof]

Try pulling out 2 ram modules and see if that makes a difference. if it does its your NB at fault.

Have you tested the ram/NB with bootable memtest & windows memtest?

Nope, did it again with two modules. I'll test other two. I'have tested with both, nothing (didn't complete the whole test, but i'll do it again). Anything else? Maybe the MB?

[CryptiK]

It may be the board but I doubt it. I'm actually leaning towards it being your cpu. Try decreasing the cpu speed until you find stability at the same vcore, only using 2 ram modules. Once it is stable, put the other 2 ram modules in again and re-test it.

[spoof]

It may be the board but I doubt it. I'm actually leaning towards it being your cpu. Try decreasing the cpu speed until you find stability at the same vcore, only using 2 ram modules. Once it is stable, put the other 2 ram modules in again and re-test it.

Actually, i did this in the beginning. With 1.375 (Or 1.4v, doesn't make that difference under water)vCPU, 1.47 vNB, 1.52CPU PLL, 1.34 FSBt with 65x GTLREF it's rock stable. No lockups, Prime95 for hours. This is with 2 and 4 sticks. Also did try 472 with 8.5 and 8 ratio with 2 sticks - it won't load windows at all. So it's FSB issue, but i don't know which is better - CPU or MB. I have here GA-P35-DS3R i'll see if it gonna boot at 472. At this time, any ideas would be greate (:

[mikeyakame]

You'll find above 475MHz FSB it's very hit and miss. You have to use low multiplier to get it stable in my experience. I use 7x, but on a 45nm chip I'd use 7.5 since it would give best results. High CPU freq on high fsb on the X48 boards that arent the Rampage Extreme is very difficult to stabilize. Voltage circuitry can't provide stable enough voltage on all the FSB / CPU phases.

I personally found 7x 466MHz on my RF with the Q6600 is a sweet spot. Only need 1.35v Vcc for 3.26GHz, and 1.49V Vnb for PL7 on 5:6 divider.

Edit: here we go seems ok, i'll visit the bios and photo the screens and add them here for current settings.

[mikeyakame]

bios settings 482mhz fsb q6600 g0 7x482.

[CryptiK]

Actually, i did this in the beginning. With 1.375 (Or 1.4v, doesn't make that difference under water)vCPU, 1.47 vNB, 1.52CPU PLL, 1.34 FSBt with 65x GTLREF it's rock stable. No lockups, Prime95 for hours. This is with 2 and 4 sticks. Also did try 472 with 8.5 and 8 ratio with 2 sticks - it won't load windows at all. So it's FSB issue, but i don't know which is better - CPU or MB. I have here GA-P35-DS3R i'll see if it gonna boot at 472. At this time, any ideas would be greate (:

472 FSB is getting high for an older X48 board, there are a couple that will now do it but perhaps not the P5E - look in your mobo's thread for other people running quads for any recommendations.

As you said try it in the P35, although I'm not sure if it's capable of the FSB/clocks you're after. the P5K-Deluxe certainly would be but the Gigabyte I've not had experience with.

If only running 2GB of ram did not haelp at all, I doubt it is the NB, so that leaves the cpu itself or the FSB capability of the mobo. You can verify the cpu can run that FSB & overall clock stable on another board, really raise the vcore and test if its the cpu (some cpu's just cannot go over a certain FSB and it is not the board holding it back) or get very aggressive with the vNB for a few tests.

[mikeyakame]

i concur with CryptiK, some chips are better for high multiplier/freq & low fsb, some mid fsb, high frequency, mid multi, others high fsb, lower multi, mid frequency, and then u got chips which just completely suck!~

[spoof]

Well, i am very, very thankful that you are trying to help! I appreciate it. Actually, i'm familiar with so called FSB Wall, but on my other CPU (6750) i did actually push it a bit with new BIOS and got more than 20Mhz more with CPU PLL voltage settings. I'm gonna test it here, because it works on 470. If it's ok - than it's Q9550 faults. If not - i'll change the mobo.

Also, i wonna thank for the bios pic, may be i'll test some settings(the "small" RAM timings).

[mikeyakame]

few minor adjustments, incr fsb by 1 to 483mhz, mem to 5-5-5-16 tRFC 70, pull in phase PHA1, ai clock twister moderate, dram clock skew 100ps advance channel b, Vnb 1.53v, Vdimm 1.98v, Vcc 1.3875v, CPU GTLREF 0.65x.

nb: note read delay adjust, dimm fine delay values, i set these manually by the combination of settings for ai clock twister, ai trans booster pull ins and dram clock skews.
read delay phase adjust +1 is from phase pull in on A1, B1 caused errors so instead I advanced the clock skew on chan b by 100ps which should be equiv to -2T on Dimm 3 fine delay. ai clock twister stronger -> moderate reduces dimm fine delay value, stronger was 5 , 3 , 5 , 3. Moderate changed to 3, 2 , 3 , 2 on dimm 1 - 4 fine delay. Pull in phase A1 on AI trans booster for PL, incremented Dimm 1 fine delay by 1 to 4T, and 100ps advance on Dram Skew CHB while not changing Dimm 3 value directly corrected the diff in skews and fixed memory errors!

latency is improved. this is as stable as it'll ever be, been playing red alert 3, thrashing my gpu on Vantage, linpack, etc and no crashes or errors.

[The Water Dog]

is this right?:
cpu vcore 1,39
Cpu gtl refrence 0\2:0.670
Cpu gtl refrence 3\4:0.670

I use Formula 2, and struggle to understand this:S
Thanks for the help=)

[ci2kla]

May I join in
I was playing with these values recently but
unfortunately...my GTL gettings has become +-mv which allow fine tuning but difficult to control. Spec as sig.
The problem is I was never be able to find out what the current values of CPU GTL Ref and NB GTL Ref are (especially on "AUTO") so I wonder where should I start with...
I've heard alot of overclockers said CPU GTL Ref. don't have much effect with Dual Core chips. Is that true...if so..seems I'm wasting my time messing up these settings

Adjusting the CPU GTL Ref. will case wierd screen display, frequent crashes in programs:
CPU GTL 0: auto
CPU GTL 1: -5mv
CPU GTL 2: auto
CPU GTL 3: -5mv
NB GTL: auto

VTT was about 1.28v, can someone please tell me why this would happen? Any help would be much appreciated.

[rob2k]

Hi, does anyone know if clock skews have an effect on GTL settings ?

I have just updated my bios and had to change from -'ve to +'ve GTL settings after setting :-
CPU Clock Skew: Delay 300ps
NB Clock Skew: Delay 200ps

Maybe Asus just programmed the bios wrong ???.

[mikeyakame]

Hi, does anyone know if clock skews have an effect on GTL settings ?

I have just updated my bios and had to change from -'ve to +'ve GTL settings after setting :-
CPU Clock Skew: Delay 300ps
NB Clock Skew: Delay 200ps

Maybe Asus just programmed the bios wrong ???.

hmm its possible I suppose, since gtl reference determines clock period, skewing it probably changes the clock assert/deassert timing in relation to the base clock and gtl reference voltage is used to determine a valid low or high clock with respect to the clocking source (bclk). It may be skewing the clock assert/deassert for either the cpu or nb could create a situation where either assert/deassert for nb/cpu clock happen too early or too late and occur inside the crossing threshold or basically not within the valid area of a clock period, which exists where the high clock crosses the crossing threshold on the rising edge and crosses it again on the falling edge (determined by slew rate, rising of voltage / distance unit), and on the low clock it happens the other way around.


nb clock skew and cpu clock skew work a little different than the name suggests. nb clock skew actually refers to mch side of nb with relation to the fsb base clock, and cpu clock skew is the cpu clock in relation to the fsb bclk. dram clock skew is actually the skewing of the address strobe clock between the dram controller side of mch and actual memory modules.

[truehighroller]

Yeah. English man.

[mikeyakame]

Yeah. English man.

i'm tired, translating a foreign electronics dialect into english is a lot of work! I'll translate it tomorrow after I get some sleep, right now i'm too worn out from work to explain it for the people

[truehighroller]

i'm tired, translating a foreign electronics dialect into english is a lot of work! I'll translate it tomorrow after I get some sleep, right now i'm too worn out from work to explain it for the people

[Pyr0]

The problem is I was never be able to find out what the current values of CPU GTL Ref and NB GTL Ref are (especially on "AUTO") so I wonder where should I start with...

i was wondering what multiplier/percentage asus are using for the nb gtl ref on the rampage extreme too (0.67?). anyone know? or know how to find out please?

[rob2k]

hmm its possible I suppose, since gtl reference determines clock period, skewing it probably ............(edited)

Hi Mikeyakame, and many thanks for taking the time for the concise reply although I don't understand it 100% it's good to know that it is in theory possible and that I haven't got things horribly wrong