Negative offset and "safe" overclocks - best way to do it?

[CakeMonster]

I'm building a SB pc for a friend and just to get him some extra performance I bumped the multiplier up to 40 and 4GHz. I don't care if the chip might be capable of 4.5GHz, I just want some extra performance over stock speed. It passes IntelBurnTest fine with 50+ runs. When I monitor it with HWInfo64 I see that the temperature maxes out at 88C (usually much lower) and that voltage goes to 1.371v. That voltage should not be necessary for only 4GHz.

I've looked into options to reduce it but I don't want to set manual voltage since it will run at that all the time. So I've looked at offset. I tried -0.05v but that gave zero difference at it still reports 1.371 at max when doing IBT while it theoretically should have been 1.321. I'm skeptical about lowering it further since from what I've read it will also lower idle voltage.

The motherboard is a P8P67 with latest BIOS and i5 2500K.

Any ideas what is the best way to go in order to get a cool "safe" overclock and avoid tons of unnecessary voltage?

(BTW I'm having the same problem with my i7 4770K, the mobo will give it way too much voltage even at stock load speed considering what it should need, so this seems to be a general problem).

 

[BSim500]

I'm building a SB pc for a friend and just to get him some extra performance I bumped the multiplier up to 40 and 4GHz. I don't care if the chip might be capable of 4.5GHz, I just want some extra performance over stock speed. It passes IntelBurnTest fine with 50+ runs. When I monitor it with HWInfo64 I see that the temperature maxes out at 88C (usually much lower) and that voltage goes to 1.371v. That voltage should not be necessary for only 4GHz.

88c and 1.371v is far too high for that speed. It sounds like you've got some auto-overvolt feature on the motherboard on. Try disabling stuff like LLC (Load Line Calibration) which shouldn't be needed for such a mild OC. A lot of this stuff varies from board to board, so read through your motherboard manual and see what OC related settings default to "auto" and whether they have any impact.

I have an i5-3570 (non K) which with a 4GHz mild OC still has enough headroom for a significant -0.1v undervolt (resulting in a crazy low 49w quad core, LOL). At 4GHz load it's about 0.996v and @ 1.6GHz idle it's down all the way to 0.680v. Temps never exceed 55c for Prime or about 50c for video encoding, all 4 cores loaded. Don't be afraid of negative offsets also reducing idle voltage as idle VID's still have headroom too. It's only a problem if it becomes unstable / you get WHEA errors.

 

[CakeMonster]

Thanks, I will definitely try that on my HW.

This is really strange on the SB though, I've tried setting LLC to "regular" since there's no disable option. I've also tried disabling Internal PLL overvoltage, as well as setting CPU Current Capability to 100% and no more. Yet with any offset and even with manual, I get the same voltages when measuring with HWInfo64! However something is different because it crashes with -0.1 or when setting it to 1.28 fixed (the latter really shouoldn't happen).

I'm starting to think this is not worth my effort. And I didn't even want a huge OC, I just wanted a really tiny OC that I know the CPU is easily capable of if the mobo just didn't insist on frying it because that's what the customers want apparently.........


Edit: OK, with the features above turned down/off I am now running a test at 3.9GHz and getting temperatures in the 90s... I give up.

Edit2: With BIOS defaults and auto speed which means 3.4GHz when running IBT on all cores, I get 81C max at load. I've reattached the heatsink now too.

 

[BonzaiDuck]

Thanks, I will definitely try that on my HW.

This is really strange on the SB though, I've tried setting LLC to "regular" since there's no disable option. I've also tried disabling Internal PLL overvoltage, as well as setting CPU Current Capability to 100% and no more. Yet with any offset and even with manual, I get the same voltages when measuring with HWInfo64! However something is different because it crashes with -0.1 or when setting it to 1.28 fixed (the latter really shouoldn't happen).

I'm starting to think this is not worth my effort. And I didn't even want a huge OC, I just wanted a really tiny OC that I know the CPU is easily capable of if the mobo just didn't insist on frying it because that's what the customers want apparently.........


Edit: OK, with the features above turned down/off I am now running a test at 3.9GHz and getting temperatures in the 90s... I give up.

Edit2: With BIOS defaults and auto speed which means 3.4GHz when running IBT on all cores, I get 81C max at load. I've reattached the heatsink now too.

Your overclock of 4Ghz is too low or your VCORE (as read by your software) is too high. For the 2500K, stock turbo-mode is only a couple hundred Mhz below your clock speed. P67 board was probably the earliest entry in the socket 1155 generation. But it shouldn't much matter.

Your load voltage (or momentary peak between load and idle) should not be much more than the stock requirement. You can, of course, run some tests with IBT or LinX designed to use all or most of the RAM, and watch the GFLOPS results for each iteration. Sufficient voltage should show consistent GFLOPS for all iterations. The difference between "sufficient" voltage for stability and the voltage for GFLOPS consistency should be probably around 10 mV.

I've had my 2600K Turbo-clocked to 4.60 Ghz for two years -- 24/7/365. My peak voltage at load-to-idle transitions is about 1.37V. With EIST and the other E-saving options enabled, the processor spends a lot more time at a voltage around 1.00+V I'm using a positive offset of 0.005V or "one notch" above 0. I adjust the "additional voltage in Turbo-Mode" beyond that.

I had a "glitch" which had haunted me for at least a year. The system is "multi-use" and defies the general wisdom of "dedicated HTPC" usage: I watch TV, play games and use business software simultaneously. Every couple weeks -- sometimes once a week -- the system would reset and cycle through the boot sequence -- without warning.

since the RAM (DDR3-1600 G.SKILL Ripjaws 1.5V totaling 16GB or 4x4GB) has been overclocked to DDR3-1866 and CMD=1, I had thought the VCCIO setting was sufficient. It looks as though bumping VCCIO up a couple notches to 1.13+V has solved the problem. I'll know for sure in another week or two.

 

[MadScientist]

Why don't you just use the auto overclocking feature in the bios. Set it to 4.0Ghz and let the motherboard do it's thing.
http://www.tomshardware.com/reviews/automatic-overclock-motherboard-cpu,3048.html

But if you want to do it it manually using offset voltage here's some tips:
http://rog.asus.com/forum/showthread.php?2162-Overclocking-Using-Offset-Mode-for-CPU-Core-Voltage

http://forums.anandtech.com/showthread.php?t=2194577&highlight=

Be aware though that if you set the voltage too far with the negative voltage scale, there will come a point where the idle voltage is no longer sufficient to sustain the idle processor clock frequency.

 

[CakeMonster]

Ok, after playing around for a good while it seems that offset does NOTHING regardless of what I put it on with this mobo/cpu.

I went back, set everything to auto, and enabled turbo for all cores (meaning 4 core load will now run at 3.7GHz instead of 3.4GHz). I enabled the Asus option of "EPU Power saving - MAX" which seems to drop the voltage under load from 1.371 to 1.361. I think this is all the voltage adjustment this combo will allow me. Max temp during a IBT run is now 90C.

The guy who will be using it won't do anything with it that always pushes it, since it passes the test, and the graphics card will make more noise than the CPU fan anyway, I figure he should be safe with the above settings. The heaviest thing he will use it for is Civ5.

After 10years+ in this game, I am getting really frustrated at the complexity and inability to configure things as I want anymore. So this experience, combined with the experience with my own current setup and voltage and temperatures off the charts even at stock, is definitely pushing me in the direction of not bothering with OC anymore in the future.

 

[BonzaiDuck]

Ok, after playing around for a good while it seems that offset does NOTHING regardless of what I put it on with this mobo/cpu.

I went back, set everything to auto, and enabled turbo for all cores (meaning 4 core load will now run at 3.7GHz instead of 3.4GHz). I enabled the Asus option of "EPU Power saving - MAX" which seems to drop the voltage under load from 1.371 to 1.361. I think this is all the voltage adjustment this combo will allow me. Max temp during a IBT run is now 90C.

The guy who will be using it won't do anything with it that always pushes it, since it passes the test, and the graphics card will make more noise than the CPU fan anyway, I figure he should be safe with the above settings. The heaviest thing he will use it for is Civ5.

After 10years+ in this game, I am getting really frustrated at the complexity and inability to configure things as I want anymore. So this experience, combined with the experience with my own current setup and voltage and temperatures off the charts even at stock, is definitely pushing me in the direction of not bothering with OC anymore in the future.

Quite frankly, I'm almost inclined to "not bothering . . anymore" myself. But in order to OC my own SB, I studied OC "guides" online for the P8P67 mobo, and the BIOS features for it weren't too different than what I found for my Z68 V-Pro board.

MadScientist is correct in his assessment about offset and idle-voltage. Frankly, I don't see much of a problem with a positive offset: I've overclocked my own system at the first notch of that option. It would only affect the peak voltage according to an analysis published here some years ago in the month of November for a Yorkfield core.

It isn't so much the load voltage -- which droops even with moderate LLC settings -- that would be as much concern as the transition voltage spike that occurs between load and idle. Unless there's something wrong with the processor, I can't understand why you're showing 1.37V (likely the unloaded turbo value) for such a minimal over-clock. Some others may offer insight, or simply express irritation at me for pursuing the issue.

The "auto" settings for turbo-overclock always gave excessive voltage to achieve stability. If they hadn't, I would not have bothered with manual tweaking.

Also, on these newer processors, some may recommend that the tedious and repetitive OC'ing should first be done with a fixed VCORE. I don't think this helps much. The OC'ing can be done with EIST, C1E etc. enabled -- I never had a problem, nor any delay, in doing so with my 2600K. These ASUS boards report actual voltages alongside the text-boxes for settings in the BIOS, and taking note of these things facilitates honing in on stable settings without too much trouble.

We've now passed through two generations after Sandy Bridge. For my personal uses, the SB processor takes anything I throw at it. I might have been more inclined to just run it with stock settings from the beginning, but @ 4.6 Ghz-turbo, it's probably the most stable system I ever built.

Also, since you're building this system for a friend, I would discourage you from using any but the most modest OC. And if that modest OC is such as you describe, I still think your SB is showing voltages higher than what it would otherwise need.

You really don't want your friend coming back to you for help when the system crashes, or in the event that voltage settings "drift" after many months of use.

 

[CakeMonster]

Yep I am puzzled by the voltage too, I just don't feel like pursuing it because I'm frustrated enough as it is. It is reported that way by both HWInfo64 and CoreTemp. Also, the temperatures points to the voltage reading being correct. It seems perfectly stable from long runs of IBT so I don't think that's going to be a problem, and I don't mind this particular guying calling me up IF it turns out to become unstable later on. Thanks a lot for the input.

 

[BonzaiDuck]

Yep I am puzzled by the voltage too, I just don't feel like pursuing it because I'm frustrated enough as it is. It is reported that way by both HWInfo64 and CoreTemp. Also, the temperatures points to the voltage reading being correct. It seems perfectly stable from long runs of IBT so I don't think that's going to be a problem, and I don't mind this particular guying calling me up IF it turns out to become unstable later on. Thanks a lot for the input.

With that board -- did you find some item labeled "Xtra Voltage for/in Turbo"?

I'm wondering if you might try resetting everything to stock. It is then your choice to use the TPU switch, or use the BIOS Turbo-Overclocking option -- to find a "stability" starting point at the high end. Then drop the voltage to what you'd expect it to be under speeds reported by other posters. There are some threads on the Sandy Bridge that have four or five gurus comparing their notes. Generally, 4.5 Ghz requires less than 4.6, and 4.6 evidences peak measurable voltage in the load-to-idle transition of 1.37V -- more or less, according to the experience of "various people." The actual load voltage with second-highest LLC setting still allows for a 20 mV droop.

What seems to be showing is either a problem with a voltage regulation feature or something with the sensor(s) itself. I dunno -- maybe the processor? Where did you get the 2500K? New or used?