Core i7/X58 Overclocking Thread

[Cooperdale]

Um, I think you got something mixed up here...

25x160MHz = 4.0GHz
25x167MHz = 4.175GHz

That's 175MHz difference between the two. Up your core voltage to about 1.40V and try from there...no need to increase QPI/DRAM voltage. Chances are however that unless you have a very good chip (i.e., one of the golden ones that run low volts and high clocks) you will most likely need about 1.46V - 1.47V at that clock though.

Oh, sorry, I simply mistyped. I meant 24x167Mhz=4008Ghz. That's why I think I should be able to keep the vcore I'm using for 25x160 (1.34375V). The cpu frequency is the same, but I have difficulties raising the BCLK from 160 to 167.

[dejanh]

Oh, sorry, I simply mistyped. I meant 24x167Mhz=4008Ghz. That's why I think I should be able to keep the vcore I'm using for 25x160 (1.34375V). The cpu frequency is the same, but I have difficulties raising the BCLK from 160 to 167.

Up your volts by just one notch (approx. 0.006V) and then try. Your clocks are not exactly the same and yes, before you say "but it's only 8MHz" it is 8MHz, and with these chips it can make a world of difference. Also, different multiplier combinations with different BCLK values require different voltages. On the i7 920 and i7 940 higher BCLK with lower multiplier though usually means slightly lower voltages. Maybe it is not always the case though.

Anyway, just try upping it by one, max two notches. Also, you got the XE chip and it overclocks the best on the R2E board, not the P6T-Deluxe. Why did you go for that board with an XE chip?

[unclewebb]

I'm always trying to think up new ideas and features for RealTemp so how about this one?

Intel designs the Core i7 to start to thermal throttle at about 98C which should keep the temperature below 100C as long as things aren't too crazy. This works fine at default clocks with default core voltage but when you're running a healthy overclock, 100C is just too damn hot. You'll usually lose stability and crash before this temperature.

How about a feature that let's a user choose the maximum temperature they'd like their CPU to run at. RealTemp could automatically adjust the thermal throttle for you and slow your CPU down when necessary to keep your temps at a maximum of 90C or whatever number you like and then automatically go back to full speed as soon as the temperature allows.

This way when you come up with an overclock that you are happy with, you won't have to adjust it at all for 24/7 use. If the sun comes out or some tiny thing changes, you'll still be OK. You'll be in complete control of the maximum temperature your CPU hits.

Open up the Settings window in RealTemp and you can play around with the thermal throttle by making adjustments in the Clock Modulation section while running Prime Small FFTs. It manually gives you complete control over your maximum temperature. I think if I modified this feature so it was automatic, it could be very useful especially for guys trying to run a Core i7 with air cooling. Any thoughts? Is this idea or

[dejanh]

I'm always trying to think up new ideas and features for RealTemp so how about this one?

Intel designs the Core i7 to start to thermal throttle at about 98C which should keep the temperature below 100C as long as things aren't too crazy. This works fine at default clocks with default core voltage but when you're running a healthy overclock, 100C is just too damn hot. You'll usually lose stability and crash before this temperature.

How about a feature that let's a user choose the maximum temperature they'd like their CPU to run at. RealTemp could automatically adjust the thermal throttle for you and slow your CPU down when necessary to keep your temps at a maximum of 90C or whatever number you like and then automatically go back to full speed as soon as the temperature allows.

This way when you come up with an overclock that you are happy with, you won't have to adjust it at all for 24/7 use. If the sun comes out or some tiny thing changes, you'll still be OK. You'll be in complete control of the maximum temperature your CPU hits.

Open up the Settings window in RealTemp and you can play around with the thermal throttle by making adjustments in the Clock Modulation section while running Prime Small FFTs. It manually gives you complete control over your maximum temperature. I think if I modified this feature so it was automatic, it could be very useful especially for guys trying to run a Core i7 with air cooling. Any thoughts? Is this idea or

Um, well...two things...

I am using RealTemp now (version 2.90, latest update) and my thermal throttling starts way before 98C. In fact, my throttling starts at about 85C - 87C, at least according to RealTemp as it is reporting that temperature when throttling starts to happen. How come? Where did you get this 98C figure?

In terms of RealTemp being able to cause the CPU to throttle, good idea, as long as the temps are correct

[Splave]

Im pretty sure its 98c where it throttles down, couple people in this thread have had it happen.

[unclewebb]

In fact, my throttling starts at about 85C - 87C, at least according to RealTemp as it is reporting that temperature when throttling starts to happen.

Does RealTemp show you are throttling at those temperatures? Can you post a screen shot of that?

If the Thermal Status area of RealTemp is reporting OK then your CPU hasn't been throttling.

The Core 2 CPUs throttled when Distance to TJMax was about 2 or about 98C. I've seen one or two Core i7 screen shots that seemed to show the same thing but I haven't personally run a Core i7 up to that temperature.

Edit: Here's a pic from a few pages ago of SteveRo's CPU hitting 97C without setting the thermal throttle off.
The Thermal Status section of RealTemp reads the bit within these CPUs that keeps track of any thermal throttling episodes.



Not to get too off topic here. If you're interested in testing out Automatic throttling then send me a PM and I'll get to work on it. Should be very easy to code.

[Cooperdale]

Up your volts by just one notch (approx. 0.006V) and then try. Your clocks are not exactly the same and yes, before you say "but it's only 8MHz" it is 8MHz, and with these chips it can make a world of difference. Also, different multiplier combinations with different BCLK values require different voltages. On the i7 920 and i7 940 higher BCLK with lower multiplier though usually means slightly lower voltages. Maybe it is not always the case though.

Anyway, just try upping it by one, max two notches. Also, you got the XE chip and it overclocks the best on the R2E board, not the P6T-Deluxe. Why did you go for that board with an XE chip?

Well, I just aim for 4Ghz, I don't want to do extreme oc, since I'm on air. So I thought the P6T would be fine. Moreover, I have 8 Sata drives and the R2E only has 7 connectors, didn't want to bother with addon cards.

But I didn't know there would be big differences. What's the advantage of R2E in particular? Lower voltages?

Oh and anyway, what are the suggested QPI voltages? Is there an absolute upper limit, or would say 1.4V be fine?

[fornowagain]

snip

You're kidding, big time

[dejanh]

Does RealTemp show you are throttling at those temperatures? Can you post a screen shot of that?

If the Thermal Status area of RealTemp is reporting OK then your CPU hasn't been throttling.

The Core 2 CPUs throttled when Distance to TJMax was about 2 or about 98C. I've seen one or two Core i7 screen shots that seemed to show the same thing but I haven't personally run a Core i7 up to that temperature.

Edit: Here's a pic from a few pages ago of SteveRo's CPU hitting 97C without setting the thermal throttle off.
The Thermal Status section of RealTemp reads the bit within these CPUs that keeps track of any thermal throttling episodes.



Not to get too off topic here. If you're interested in testing out Automatic throttling then send me a PM and I'll get to work on it. Should be very easy to code.

RealTemp is reporting OK, but the CPU is definitly throttling. I found out this morning while doing some more testing on my 4.14GHz OC. As soon as temps hit around 85C-ish the multiplier started to drop (reported by CPU-Z). I am pretty sure I can get a before and after screenshot for you so you can see it.

Give me a few minutes, I'll try to get you the screens...

[unclewebb]

dejanh: Show a sceen shot of RT and CPU-Z if you can. Does the multi drop because the Turbo turns off or does it drop right down to 12? rge noticed this depending on what bios version he was running and how many threads of Prime he had running.

[dejanh]

dejanh: Show a sceen shot of RT and CPU-Z if you can. Does the multi drop because the Turbo turns off or does it drop right down to 12? rge noticed this depending on what bios version he was running and how many threads of Prime he had running.

No it does not drop right down to 12. It throttles from 23 (turbo) down to 22. But throttling is throttling. It is happening because the thermals are getting too high so turbo shuts off...always at the same point too, when the temps are around 85C-87C. The multiplier will drop to 22.5x, then 22x, then if the temps drop a bit (say below 86C) it will go back up to 23X. The screen shot below shows it throttling at 89C. I don't know what the business of BIOS is here, but every R2E BIOS behaves the same as this. There have been no changes to anything that has to do with thermal management.

Not saying that it's a bug so much as that it may just be a feature oversight (if there is two-stage throttling, it should be reflected somehow). Anyway, here it is...and 8 thread test...

[unclewebb]

Thanks for posting that. If the multiplier is dropping from 23 to 22 that's not technically thermal throttling, it's just the turbo kicking out.

I'm not 100% convinced that CPU-Z is correct though. rge provided me with the Intel documentation for calculating the multiplier for Core i7 and I worked with him to hopefully get this right. Neither of us have come across any documentation from Intel showing that half (0.5) multipliers exist for Core i7. They exist for 45nm Core 2 CPUs but not for Core i7. I'm not sure what CPU-Z is reporting when it reports this.

Here's a post from rge over on the XS RealTemp forum:
http://www.xtremesystems.org/forums/...postcount=2731

The Intel approved / recommended method for calculating the multiplier when the Core i7 Turbo is enabled uses timers located in the processor that CPU-Z does not seem to be using. RealTemp is showing the multi steady at 23 in both of your screen shots. We'll have to wait for a few more updates from CPU-Z to see what the truth is.

[dejanh]

Thanks for posting that. If the multiplier is dropping from 23 to 22 that's not technically thermal throttling, it's just the turbo kicking out.

I'm not 100% convinced that CPU-Z is correct though. rge provided me with the Intel documentation for calculating the multiplier for Core i7 and I worked with him to hopefully get this right. Neither of us have come across any documentation from Intel showing that half (0.5) multipliers exist for Core i7. They exist for 45nm Core 2 CPUs but not for Core i7. I'm not sure what CPU-Z is reporting when it reports this.

Here's a post from rge over on the XS RealTemp forum:
http://www.xtremesystems.org/forums/...postcount=2731

The Intel approved / recommended method for calculating the multiplier when the Core i7 Turbo is enabled uses timers located in the processor that CPU-Z does not seem to be using.

Well isn't that interesting...I'm curious if it is really kicking turbo off or not. RealTemp seems to be showing it as using turbo still, even on my screenshot...

Edit: I decided to do another run, and yes, the multiplier does drop. CPU-Z is not wrong on that one (not sure about the 1/2 multipliers). RealTemp also started showing 180x22 instead of 180x23 (not visible in my screenshot as this happens a few minutes after CPU-Z already reports a lower multipler). So what is kicking in here? Why is it dropping the multiplier and always around 87C+ if this is not some sort of throttling kicking in? I just quickly reviewed the Intel specs and they mention two types of thermal throttling that work in conjunction, but I do not have time to review the details...

[unclewebb]

I told rge to drop by here sometime so he can shed some light on this. I know he has experimented with different bios versions that have handled kicking the turbo off differently. He ran some Super PI benches originally to show me that the early versions of RealTemp and Everest were wrong. I'm not sure if a Super Pi bench or an XS Bench while running Prime 95 to maintain your core temperature would be able to prove this or not. RealTemp doesn't update the MHz or multiplier while an XS Bench is in progress so keep that in mind if you use that. It's definitely accurate and repeatable enough to show whether your computer is using a 22 or 23 multi but running it while Prime is running will take CPU cycles away from Prime which might cause your temperatures to drop and your CPU-Z reported multi to go back to 23. Maybe 7 threads of Prime with one left over to run an XS Bench might keep the temp high enough to prove this. If you can think of any way to prove this, give it a shot.

I followed the Intel Turbo white paper that rge sent me exactly which was published in November 2008. I don't think I screwed up following their directions and I'm curious too which is really correct. I don't understand why CPU-Z is not using this method.

http://download.intel.com/design/pro...=tech_tb+paper

[Splave]

dejanh, how about if you disable turbo? Does it drop a multi then when you hit those temps?

[rge]

Is that with 1.38v vcore and HT enabled, RGE?

Yep, you can tell that though with 8 threads on prime running.


Regarding turbo....turbo disengaging, multi decreasing to stock, or whatever you want to call it, can be set via bios manufacturer complying or not complying with TDP, current, and thermal limits. Sounds like Asus has thermal limited turbo at ~85C, or could be TDP limited and just that you hit that temp whenever TDP limit is reached, or could be both. Also, both on CPUZ and Realtemp, if you are near the turbo TDP/current/temp limit, when the turbo kicks out, temp/tdp decreases, turbo kicks back in, etc the multi bouncing back and forth between 23 and 22 is typical. Though if you run prime and linpack at same time, you might find turbo stays disengaged but dont know for sure as dont know how ASUS is limiting their turbo. The intel white paper lists the three possible constraints, which can be turned off or turned on in some intel bioses, apparently a hidden setting in ASUS and GB bioses where they have implemented for us in some manner:

From what I have seen GB bios F3 is TDP limited, havent tested for thermal though, and F4betas like I am using I believe has TDP etc limits turned off.

[Xello]

I believe RGE is on an F4 GB bios right now, like me, unfortunately (actually it's pretty neat in a way) turbo is permanently on, so there's no way to test the whole thermal control system And in fact i don't recall seeing anything in the GB bios to do with turning that thermal control on or off. There is something there but i think it's to do with the actual thermal threshhold of the cpu when it will shut down / throttle, i don't think there's anything specifically to do with the turbo mode like there is on the Asus board.

Yep, you can tell that though with 8 threads on prime running.

Heh yeah, i'm sitting between 45 and 50c right now with 1.35v prime blend loaded, i dread to think what will happen to temps if i turned HT on.

[Cooperdale]

Has this something to do with Vista System panel reporting wrong cpu speeds for high multipliers? In my case only some softwares (cpu-z, Real-Temp, TurboV, some Everest pages) report the right cpu frequency when I'm at 160x25. Most softwares report 3.84Ghz, as though the multi were 24. Turbo is off though, but I think the problem is with Vista reading the default cpu multi (24 for the 965EE) and multiplying it for the Bclk. This happens with any software I've tried except for the ones I've mentioned.

Cpu-z though reports 4Ghz as if it were a constant frequency, whereas Real-Temp reports many variations, with the multi covering all the range from 20 to 25.

[dejanh]

Yep, you can tell that though with 8 threads on prime running.


Regarding turbo....turbo disengaging, multi decreasing to stock, or whatever you want to call it, can be set via bios manufacturer complying or not complying with TDP, current, and thermal limits. Sounds like Asus has thermal limited turbo at ~85C, or could be TDP limited and just that you hit that temp whenever TDP limit is reached, or could be both. The intel white paper lists the three constraints, which can be turned off or turned on in some intel bioses, apparently a hidden setting in ASUS and GB bioses:

From what I have seen GB bios F3 is TDP limited, havent tested for thermal though, and F4betas like I am using I believe has TDP etc limits turned off.

Ahhh, this is turning into a real headache Asus does not even have the ability do disengage TDP limit on any of their boards, so there is no way for me to test this. I can technically disable the whole CPU TM function which supposedly disables the TDP limit as well except that the fallout of that can be a dead CPU in case the temps get too high. Not exactly something I am willing to do on a $800 CAD processor.

In any case, this seems to be another nuisance on top of many already when trying to OC this chip. Now it looks like that unless I can keep the temps below 85C all the time I better not be expecting turbo to give me a performance boost. I wish the board manufacturers would standardize on something...

Maybe it's time to go to a 965 XE to avoid this mess in overclocking...anyone interested in buying my Core i7 940? I'll give a good price

[rge]

Cpu-z though reports 4Ghz as if it were a constant frequency, whereas Real-Temp reports many variations, with the multi covering all the range from 20 to 25.

With speedstep enabled, Realtemp will report varying multis depending on background activity of computer ie cores active vs not. CPUZ reports constant multi with speedstep enabled, ie does not seem to work correctly with speedstep enabled. If you disable C1E and EIST, then both Realtemp and CPUZ should report a stable multi.

[Hornet331]

With speedstep enabled, Realtemp will report varying multis depending on background activity of computer ie cores active vs not. CPUZ reports constant multi with speedstep enabled, ie does not seem to work correctly with speedstep enabled. If you disable C1E and EIST, then both Realtemp and CPUZ should report a stable multi.

for me cpu-z also reports varying multis, same as realtemp.

[rge]

for me cpu-z also reports varying multis, same as realtemp.

LOL...interesting, and you also have GB extreme? Crazy how everybody is seeing different things with different bioses, setups, OS, etc. Are you on vista? I just enabled speedstep and checked again, Realtemp fluctuates and reads lower multis, cpuz stays rock solid at bios setting 21 multi, but I am on XP...apparently something different between my setup and yours.

[Splave]

any tests done 1600 cas9 vs 1200 cas6 etc?

Because if what was said is true about the x16/x16/x8 and droping memory down to x6 and uncore to x14 or x15 @ 200x21 you are at ddr3-1200. Why buy a 1600mhz kit then right?

[zlojack]

any tests done 1600 cas9 vs 1200 cas6 etc?

Because if what was said is true about the x16/x16/x8 and droping memory down to x6 and uncore to x14 or x15 @ 200x21 you are at ddr3-1200. Why buy a 1600mhz kit then right?

Well, hopefully tighter timings.

Anyway, I'm on x18/x16/x8 right now with 200 bclk and Priming. I had it at x18/x18/x8 and it failed after 5 hours.

I'm at 3200 CPU/3600 QPI/3200 Uncore/1600MHz RAM

Are people using SmallFFTs or Blend or both?

[Silaz]

Here is my i940. I forgot to take a picture of it after 8 hours of prime95. Highest temp was 88 during that run.
vcpu 1.325
pll 1.81
qpi 1.39