Core i7/X58 Overclocking Thread

[Xello]

I haven't touched QPI PLL, i've still to find out what CPU / QPU PLL does.

All i did was set 1.44v QPI, remove other bottlenecks (low cpu multi, memory, 17x uncore 18x QPI). I went straight from stock to 1.44v aswell, so i'm not sure how much i actually need, it could be less.

Maybe chips will vary in max stable bclk as much as they do in max stable cpu clock?

P.s i've not seen any evidence that vcore affects QPI. I had 1.275v set when i was 1.44v QPI + 213mhz stable.

[Cooperdale]

Ummmm yeah i wouldn't run that high QPI

I don't want to go near 1.5v myself, though really some knowledgeable people are saying 1.65v is fine... But i mean if only 1.4-1.44v can get me 213 i think i'll just take it and be happy

Jumped from my 187x23 to 192x23 - Windows wouldn't even load So i dropped one at a time and i'm currently @ 189 bclk which is looking to be my max stable with 23x multi. Any higher cpu mhz and i will need to go over 1.5v which i just will not do. This gets me 4.35ghz~ though and 9.6s~ SP 1m which is not bad, and still loaded at 59c (HT off) I still think i'll end up dropping to 1.3-1.33v and sticking with the 4.1ghz for 24/7.

But isn't Intel's suggested limit for QPI/vtt 1.35V? Why should this limit not be taken into account, when the limit on vcore (1.55) is much higher than what most people are running on their rigs?

I can't cross BCLK 160 with 1.35 QPI.

[Splave]

because 1.55 comes with a massive amount of heat

its a suggestion as well, not sure if you saw the 2000mhz corsair cas7 preview but they require 1.65QPI/VTT at this point atleast

[zlojack]

dejanh, I'm having some trouble with the ratios theory here.

I've had two different settings pass 12h ond 8h Prime95 stable on two different bios versions and it definitely doesn't follow the "rule" you stated.

Setup number one does not follow the rule but it's stable. Setup number 2 does follow the rule and is also stable. The one constant is that the QPI stays below the 4GHz threshold. Could that be all that's required?

I've also tried on bios F4k and I've been having some stability issues with BOINC that I haven't worked out yet. Of course, all of this on the stock cooler doesn't help.

1) Bios F2 - Bclk 160 - RAM 10x - UC 20x - QPI 18x


2) Bios F4l - Bclk 200 - RAM 8x - UC 16x - QPI 18x

[Cooperdale]

because 1.55 comes with a massive amount of heat

its a suggestion as well, not sure if you saw the 2000mhz corsair cas7 preview but they require 1.65QPI/VTT at this point atleast

Are you sure it's not dram voltage? 1.65 QPI is damn high.

EDIT: I checked, you're right. I still am afraid, we're in testing stages with this new platform. I hope these QPI voltages are not deleterious for the system...

[ReverendMaynard]

Are you sure it's not dram voltage? 1.65 QPI is damn high.

It is 1.65v QPI in their profile. Kind of strange, and probably not an option for the aircooled crowd.

[Cooperdale]

I've got good air cooling, but I still can't bring myself to go over 1.35 QPI. I'm too afraid of making the system's life very short...

Does QPI involve the cpu temp, or the system temp?

[bmilos]

I've gone up to 1.65 VTT myself, although, since I learned (not too long ago) that Intel says 1.35 max, I've dropped down to 1.4 and under. Im not opposed to going over 1.35, as it helps stability, but I also not going to EVER go over 1.55 again...

[Splave]

Mine is at 1.4125 atm, probably could lower but will need to be raised as I keep pushing anyways

[Lukee]

I've gone as far as 1.83 vtt to try to boot into 220 Bclk,

I've settled for about 1.35-1.4 as my 24/7 setting.

[dejanh]

dejanh, I'm having some trouble with the ratios theory here.

I've had two different settings pass 12h ond 8h Prime95 stable on two different bios versions and it definitely doesn't follow the "rule" you stated.

Setup number one does not follow the rule but it's stable. Setup number 2 does follow the rule and is also stable. The one constant is that the QPI stays below the 4GHz threshold. Could that be all that's required?

I've also tried on bios F4k and I've been having some stability issues with BOINC that I haven't worked out yet. Of course, all of this on the stock cooler doesn't help.

1) Bios F2 - Bclk 160 - RAM 10x - UC 20x - QPI 18x


2) Bios F4l - Bclk 200 - RAM 8x - UC 16x - QPI 18x

Glad to have some feedback! The more discussion stirs up, the more we can dissect the intricate aspects of i7 overclocking

1. In terms of the QPI having to stay below 4GHz, yes, this is definitely correct. QPI higher than 4GHz will not run on 99.9% of the boards and chips. This is critical.
2. In terms of the Uncore:QPI ratio of 8:9...I think I edited the original post the next day to reflect some people saying "I'm running Uncore higher than my QPI". While this is very possible I found that this is (a) not stable when dealing with high clocks (3.9GHz+) at least in my test setup, and (b) not exactly an option on i7 920 and i7 940 when you get to high clocks at all as you are driving your DRAM, Uncore, and QPI all based on the same high BCLK. If you use a 200MHz BCLK your lowest option for QPI is 3600MHz which is already getting up there. I tried running DRAM at 1800MHz with this and the Uncore with 3600MHz and I was completly unstable. Lowering the memory to 1600MHz (more within the ratio I defined) made the system more stable. Same holds up if I go to 180MHz BCLK with a 23x multiplier (turbo on i7 940). I put the QPI below Uncore to the test by assigning a 18x multiplier (3240MHz) to QPI and running my DRAM at 1800MHz (3600MHz Uncore). The system was unstable. Then I restored the QPI ratio to 22x for 3960MHz and the system was once again stable.

Overall, the 8:9 Uncore:QPI is not a hard rule. I just found it to be a "safe" rule in my testing. Adding to it you have to consider that Intel does not randomly spec out their processors. If they felt that ratios between different components of the CPU could be different, they probably would have set them up different, so there must be some reasoning behind their factory set ratios. They had the option easily to make the Uncore and QPI run at the same speed from the factory, or even higher, but they chose not to. I believe I found why they did that through my testing, and I do not think that it is just becase there is *maybe* no performance increase in running Uncore more than 2x DRAM. Kind of like with AMD where on the X2 processors you had to keep your HT as close as possible to 1GHz to balance stability and performance. Much higher than 1GHz and poof, the system dies, much lower and you are sacrificing performance.

[nosboost300]

Please post the following:

CPU Mult
QPI Multi
Uncore Mult
Memory Mult

CPU V
Qpi/vtt V

I'm thinking that it is an issue with one of those multipliers that is causing you to not boot up at 200 Bclk.

cpu multi 20
qpi multi 26
uncore multi 16 (or even auto.. i've had it running 4000mhz just fine)
memory multi 8x (to get rid of that as a limiting factor)
cpu voltage 1.55 (also to eliminate that as a factor)
qpi/vtt v 1.55
cpu pll 1.9
ioh core (or whatever uncore voltage was) 1.45
tried with load line calibration on and off
clock skew's upped 50ps

WILL NOT RUN 200 BCLOCK... and i'm pissed... i was hoping to hit 26k 3dmark06 before shipping off these 260's i just sold.. but looks like that ain't gonna happen!

[Xello]

Got my max stable bclk and 23x cpu clock, time for some RAM!

How are you guys finding your 1.65v rated kits when it comes to overclocking? I have virtually the same Corsair kit as i had on my X48 only that was 2x2gb 1.8v and now it's 3x2gb 1.65v. I had no problems with the old kit going to 1.99v and being 1800mhz stable, i'm not sure what to expect from the new kit.

Is it still reasonable to pump an extra .2v into the 1.65v kits and get some performance gains, or should we be more conservative now? I don't want to just jump start it with 1.85v incase i fry something

[Splave]

1.85 sounds a bit much IMO maybe like 1.74-1.8?

[Xello]

1.85 sounds a bit much IMO maybe like 1.74-1.8?

Yeah just an example

For the previous kit i was able to judge it based on several reviews, but i can't find a single review for the new kit anywhere

[Lukee]

cpu multi 20
qpi multi 26
uncore multi 16 (or even auto.. i've had it running 4000mhz just fine)
memory multi 8x (to get rid of that as a limiting factor)
cpu voltage 1.55 (also to eliminate that as a factor)
qpi/vtt v 1.55
cpu pll 1.9
ioh core (or whatever uncore voltage was) 1.45
tried with load line calibration on and off
clock skew's upped 50ps

WILL NOT RUN 200 BCLOCK... and i'm pissed... i was hoping to hit 26k 3dmark06 before shipping off these 260's i just sold.. but looks like that ain't gonna happen!

That qpi mult, is that a typo? should it be 36x?

Ok, try this..

uncore multi - 15x
mem multi - 6x

I know.. you don't want to lower it.. just try it first and see if you can get that stable!!

You can probably lower Vcore to like 1.45 or so. and you can turn on turbo

I had the same problem you did and those adjustments allowed me to stabilize at 200 Bclk.

[nosboost300]

okay... well i'm running 197 b clock right now... i managed a 195 3dmark run.. but my freakin qpi is 1.6v now.. seems kinda scary is it not???

anywho.. i'm going to run 3dmark06... if i'm not back in about 7 minutes (i think thats how long 3dmark06 was...) then my computer screwed up.. haha

and yes that was a typo :p

197 qpi success!

[skypx]

I know, it's such a horrible board, the should just take and junk it right?
HMM, maybe not.. Mine has been at 100% load for the last 48 hours and is running like a champ and it ain't that bad considering it's on room air..
See any faster on air on this page? Yes, It's a smackover board I'm using!

http://en.ripping.org/database.php?cpuid=871

Cool!!!! What are your settings?

[dejanh]

Got my max stable bclk and 23x cpu clock, time for some RAM!

How are you guys finding your 1.65v rated kits when it comes to overclocking? I have virtually the same Corsair kit as i had on my X48 only that was 2x2gb 1.8v and now it's 3x2gb 1.65v. I had no problems with the old kit going to 1.99v and being 1800mhz stable, i'm not sure what to expect from the new kit.

Is it still reasonable to pump an extra .2v into the 1.65v kits and get some performance gains, or should we be more conservative now? I don't want to just jump start it with 1.85v incase i fry something

I have the Corsair Dominator XMS3 1.65V 3x2GB 1600MHz kit and I can easily get 1800MHz out of it without any voltage tweaks really. It runs 1800MHz just as it runs 1600MHz at 1.65V. Going to 1900MHz I have to up the voltage to about 1.68V - 1.7V, but that is it. Right now I am running them at 1.65V 1800MHz.

Cool!!!! What are your settings?

Wow that is a reference to an old post, hehe. Anyway, his settings won't matter to you. Every board is different from what I can tell so far. With Asus look only for Asus-specific settings. Just as an example, R2E overvolts every voltage you set, but the Gigabyte boards seem to undervolt.

[Lukee]

I hve the Corsair Dominator XMS3 1.65V 3x2GB 1600MHz kit and I can easily get 1800MHz out of it without any voltage tweaks really. It runs 1800MHz just as it runs 1600MHz at 1.65V. Going to 1900MHz I have to up the voltage to about 1.68V - 1.7V, but that is it. Right now I am running them at 1.65V 1800MHz.

What are your timings for both 1600 and 1800?

[zlojack]

Glad to have some feedback! The more discussion stirs up, the more we can dissect the intricate aspects of i7 overclocking

1. In terms of the QPI having to stay below 4GHz, yes, this is definitely correct. QPI higher than 4GHz will not run on 99.9% of the boards and chips. This is critical.
2. In terms of the Uncore:QPI ratio of 8:9...I think I edited the original post the next day to reflect some people saying "I'm running Uncore higher than my QPI". While this is very possible I found that this is (a) not stable when dealing with high clocks (3.9GHz+) at least in my test setup, and (b) not exactly an option on i7 920 and i7 940 when you get to high clocks at all as you are driving your DRAM, Uncore, and QPI all based on the same high BCLK. If you use a 200MHz BCLK your lowest option for QPI is 3600MHz which is already getting up there. I tried running DRAM at 1800MHz with this and the Uncore with 3600MHz and I was completly unstable. Lowering the memory to 1600MHz (more within the ratio I defined) made the system more stable. Same holds up if I go to 180MHz BCLK with a 23x multiplier (turbo on i7 940). I put the QPI below Uncore to the test by assigning a 18x multiplier (3240MHz) to QPI and running my DRAM at 1800MHz (3600MHz Uncore). The system was unstable. Then I restored the QPI ratio to 22x for 3960MHz and the system was once again stable.

Overall, the 8:9 Uncore:QPI is not a hard rule. I just found it to be a "safe" rule in my testing. Adding to it you have to consider that Intel does not randomly spec out their processors. If they felt that ratios between different components of the CPU could be different, they probably would have set them up different, so there must be some reasoning behind their factory set ratios. They had the option easily to make the Uncore and QPI run at the same speed from the factory, or even higher, but they chose not to. I believe I found why they did that through my testing, and I do not think that it is just becase there is *maybe* no performance increase in running Uncore more than 2x DRAM. Kind of like with AMD where on the X2 processors you had to keep your HT as close as possible to 1GHz to balance stability and performance. Much higher than 1GHz and poof, the system dies, much lower and you are sacrificing performance.

Very good response!

Thanks again for your input.

I'm going to continue to do more testing and once I get proper cooling I'll be able to start eliminating some factors from the stability equation (at least some heat!)

[nosboost300]

198 bclk... still can't get 200...

[Classic Satch]

Has anyone seen anthing about whether ASUS is going to be releasing any BIOS updates which will let us adjust the QPI and/or memory multipliers downwards? I think that it's holding my rig back significantly. I have 4.0Ghz rock stable at 1.36v with HT on but have to use a Bclk of 211 and a 19x multiplier to do it.

[Xello]

But isn't Intel's suggested limit for QPI/vtt 1.35V?

Yeah... I guess it's possible that everyone here running over that spec including myself will be shelling out for a new chip in a few months. I'm hoping not Also, there is Saaya's VTT graveyard thread which has some info on the first posts of rigs which have been running at certain settings, one of which is 1.45v. If i check that post one day and read that the cpu is bust, i'll reduce mine instantly and pray!

I have the Corsair Dominator XMS3 1.65V 3x2GB 1600MHz kit and I can easily get 1800MHz out of it without any voltage tweaks really. It runs 1800MHz just as it runs 1600MHz at 1.65V. Going to 1900MHz I have to up the voltage to about 1.68V - 1.7V, but that is it. Right now I am running them at 1.65V 1800MHz.

Cool, i've started with 1648mhz and 2 hour stable so i'll increase bclk 3 or 4 points and see what i can get before i increase V's. I assume they're still samsung IC's, they appear to overclock extremely well. If they can go to the 1800mhz~ mark that will make the final i7 OC alot better!

Is there a way to find out what your memory's stock timings are? The board has setting 1T for me this whole time and i didn't realize, i think it's meant to be 2T for stock, because under cpu-z SPD tab under the XMP-1600 profile it has 2T CR . All my sub-timings are on Auto in bios, i have 1.66v manually and 9-9-9-24 manually.

There's a 'performance' setting in bios where you can choose Standard, Turbo or Extreme, it's Turbo on default, is that something to do with memory sub-timings? Perhaps that's why i'm on 1T.

Anyway since it's setting 1T i was wondering if there might be any other timings it's tuning without my say-so, i really want them all @ stock while i find max clocks, i'll find timings later. CPU Tweaker is good for showing you all your timings in windows, but i don't know if the numbers i'm seeing are @ stock or not.

[dejanh]

What are your timings for both 1600 and 1800?

8-8-8-24-2T (default timings, need to tighten still so may require a bit more volts than 1.65)

Very good response!

Thanks again for your input.

I'm going to continue to do more testing and once I get proper cooling I'll be able to start eliminating some factors from the stability equation (at least some heat!)

I'm still baffled by one thing though...turbo and when it shuts off. If I Prime and the temp hits 85C turbo shuts off but the CPU does not crash, but if I just run some application (like Vantage CPU test) and the temp goes to 85C the system hardlocks. But this is only on 23x180MHz, not 23x179MHz

Has anyone seen anthing about whether ASUS is going to be releasing any BIOS updates which will let us adjust the QPI and/or memory multipliers downwards? I think that it's holding my rig back significantly. I have 4.0Ghz rock stable at 1.36v with HT on but have to use a Bclk of 211 and a 19x multiplier to do it.

I would not hold my breath on this. Nobody is doing it, so I do not think Asus will either...maybe post it as a Feedback and Suggestion on Asus forums?

Yeah... I guess it's possible that everyone here running over that spec including myself will be shelling out for a new chip in a few months. I'm hoping not Also, there is Saaya's VTT graveyard thread which has some info on the first posts of rigs which have been running at certain settings, one of which is 1.45v. If i check that post one day and read that the cpu is bust, i'll reduce mine instantly and pray!



Cool, i've started with 1648mhz and 2 hour stable so i'll increase bclk 3 or 4 points and see what i can get before i increase V's. I assume they're still samsung IC's, they appear to overclock extremely well. If they can go to the 1800mhz~ mark that will make the final i7 OC alot better!

Is there a way to find out what your memory's stock timings are? The board has setting 1T for me this whole time and i didn't realize, i think it's meant to be 2T for stock, because under cpu-z SPD tab under the XMP-1600 profile it has 2T CR . All my sub-timings are on Auto in bios, i have 1.66v manually and 9-9-9-24 manually.

There's a 'performance' setting in bios where you can choose Standard, Turbo or Extreme, it's Turbo on default, is that something to do with memory sub-timings? Perhaps that's why i'm on 1T.

Anyway since it's setting 1T i was wondering if there might be any other timings it's tuning without my say-so, i really want them all @ stock while i find max clocks, i'll find timings later. CPU Tweaker is good for showing you all your timings in windows, but i don't know if the numbers i'm seeing are @ stock or not.

Dominators are 8-8-8-24-2T stock.