Why is my VCORE almost 0.1v to high in offset mode?

[DougFrippon]

Well put this simply here is my setup :

CPU : 3770k
Mobo: Asus z77 sabertooth
Cooler : noctua nh-d14
Case : CM Haf-X

I did a simple overclock in offset mode using a 45multiplier.
Result: The CPU runs at around 74-79degrees celcius under prime95 load. CPU-Z reports a vCore of 1.36-1.37v. Which I found quite high.

Using the Asus software, TurboV EVO, I went in and manually reduced the VCORE until I hit the sweetspot of stability. I found I could use a 1.27v vcore and have a PERFECTLY stable machine under load. Obviously I get much better temps averaging 10 less degrees (64-69degrees celcius)!. I'm very happy with those temp as I'm planning on delidding it soon and should get even better temps, and I'd hope to get a stable 4.8-4.9ghz perhaps.

I'm a little bit of a beginner here I guess because I really don't understand this. I though offset mode would automatically choose a reasonable voltage to ensure stability? Why is it overvolting so much as it clearly does not need 1.37v

I know there is a +/- offset you can set but using a negative offset, my idle vcore gets to low and the system just crash booting or shortly in windows. Plus there is no way to reduce the voltage by 0.1v using offset, it's way to much (the numbers already appear redat 0.02v or something, and I think the MAX value is somewhere along 0.060v).

Anyways, all I'm asking is to be able to set a certain load vcore without affecting the idle vcore(which is 1.024 - 1.032v) and I have NO IDEA how to set this up in my bios. How is it so simple to do in the asus sofware, and so complicated in the BIOS? I don't want to use TurboV EVO because I have to reaaply the new voltage each time I reboot, plus I hear sofware O/C'ing isn't recommanded.

I've done so much reading and search and i just can't find an answer to this question which appear to be quite simple, why is that?

Also I often read people recommand using Load-Line calibration to overclock.. and it just doesn't work at all for me. Currently I have it set to Regular(0%) and if I increase it my load vcore gets higher and higher. I got 1.46v+ vcores using a 75% Loadline calibration. I think LL Calibration is only useful if you run a fixed voltage, is this right?

Damn I hope I'm being clear, sorry if it's not. Thanks for your help.

 

[tracerbullet]

Keeping in mind I'm not real experienced with this, here's what I found as a template for overclocking:

1) Run manual voltage and multipliers to determine the minimum voltage required to be stable. 2) Do a dance with offset, turbo, and LLC to do get you there via the automated functions. 3) Screw with C3, C6, etc. as required to help any issues off-load or at boot.

What seemed to work for me, but may not be best of course or work for you, For that 2nd thing was: Compare several LLC values (i.e. 1, 2, 3, 4) and compare VID to VCore (read by the Core Temp and CPU-Z programs), and choose the LLC setting that made them close to equal when using my 45 multiplier, which for me that happened to be LLC=2. Once I had that determined, I decide on an offfset, -.005V for example, using only the stock multiplier and made sure I was stable at idle. Next up I dialed up the turbo offset to be a value (i.e. +.055) that would get me back to the proper voltage when under load, while using the 45 multiplier again.

Offset, turbo, and LLC seem to be the big 3 (well multiplier as well obviously but assuming for now it's 45). Maybe post up what you are using for each of those? VID and Vcore results might help too.

Hope that helps...

 

[Grooveriding]

Is there an LLC setting lower than 'regular' ? You can try reducing LLC to a lower setting and then raising your offset setting to see if you find something more stable.

Generally you can get a better overclock not using offset, but then you lose all the advantages of using offset overclocking, which are not worth giving up at all imo.

 

[DougFrippon]

Is there an LLC setting lower than 'regular' ? You can try reducing LLC to a lower setting and then raising your offset setting to see if you find something more stable.

Generally you can get a better overclock not using offset, but then you lose all the advantages of using offset overclocking, which are not worth giving up at all imo.

no, regular = 0%, and by default it's set to auto, which selects the regular.

And yeah I could probably get the O/C I want using the oldschool manual voltage, but my computer uses around +110watts when just running a prime95 run. It's powered on 24/7 so using offset mode would save me a good amount

 

[DougFrippon]

Keeping in mind I'm not real experienced with this, here's what I found as a template for overclocking:

1) Run manual voltage and multipliers to determine the minimum voltage required to be stable. 2) Do a dance with offset, turbo, and LLC to do get you there via the automated functions. 3) Screw with C3, C6, etc. as required to help any issues off-load or at boot.

What seemed to work for me, but may not be best of course or work for you, For that 2nd thing was: Compare several LLC values (i.e. 1, 2, 3, 4) and compare VID to VCore (read by the Core Temp and CPU-Z programs), and choose the LLC setting that made them close to equal when using my 45 multiplier, which for me that happened to be LLC=2. Once I had that determined, I decide on an offfset, -.005V for example, using only the stock multiplier and made sure I was stable at idle. Next up I dialed up the turbo offset to be a value (i.e. +.055) that would get me back to the proper voltage when under load, while using the 45 multiplier again.

Offset, turbo, and LLC seem to be the big 3 (well multiplier as well obviously but assuming for now it's 45). Maybe post up what you are using for each of those? VID and Vcore results might help too.

Hope that helps...

Step1) I already know the multiplier / voltage I want, I found out in the Asus Suite
Step2) This is exactly what I'm asking in this thread. How can I make automated functions to get me the voltage I want..?
Step3) Damn dude.. I have no idea what C3 and C6 are..?

 

[Vectronic]

2. Like he said, it's a dance... start with full//max LLC, set your offset... compare voltages during load.

A. voltage higher than desired voltage?... drop LLC by 20-25%.
B. voltage lower?... drop offset.
C. crashes?... raise offset.

LLC has a fairly small effect, like 0.05v... so if the difference is more than that, change offset instead because it's not LLC.

If you find the VID voltage when using fixed VCore (or whatever you did using AI Suite)... do some quick math, VID - VCore = Offset.

For instance if VID report 1.325v, but it's stable at 1.280v using fixed voltage, then:
1.325 - 1.280 = 0.045

Using max LLC, offset should be -0.045.

If that ends up making your VCore 1.300 or something (ie: higher than expected), then try lowering LLC... if that causes crashes, raise LLC again, and lower offset.

etc, etc, etc.

3. C1E/C3/C6/States will be found in an "Advanced" area of the BIOS.

The C states are basically how far the CPU shuts down when it's not doing anything. Typically you want to disable C3 and C6 right off the bat when overclocking... C1E can be left enabled for longer/higher clocks, but helps to disable that as well.

 

[Idontcare]

Typically you want to disable C3 and C6 right off the bat when overclocking... C1E can be left enabled for longer/higher clocks, but helps to disable that as well.

This type of advice seems to come and go in terms of being viewed as correct or no longer applicable.

Personally I leave all the power-saving features enabled in my BIOSes, overclocking or not. And for all that I've done over the years with my various OC'ing exploits it has not been a problem for me that I leave them all enabled.

I usually don't waste my time, or the time of others, posting this when I see the usual mantra "disable all your power options" rolled out, but in this case I am compelled to post something simply because the OP has made it quite clear that power consumption is a priority to him.

There is nothing wrong with savvy OC'ing. OC for performance but still keep an eye on power usage. Keeping your C-states enabled is imperative in those cases.

 

[Gryz]

I have the exact same question as the OP.
I have an i5-3570K on an ASRock Z77 Extreme4.

If you find the VID voltage when using fixed VCore (or whatever you did using AI Suite)... do some quick math, VID - VCore = Offset.

For instance if VID report 1.325v, but it's stable at 1.280v using fixed voltage, then:
1.325 - 1.280 = 0.045

Using max LLC, offset should be -0.045.

That is what I did. And what the OP did.

The problem then becomes the voltage at idle.
The offset will be applied at the idle voltage too.
Making the idle voltage too low. So low, the system will crash at idle. Right after booting. Or during booting.

The best answer would be to have two offsets.
One offset that would influence idle voltage (at base (16) multiplier).
And another voltage that would influence voltage at load (at load (34-50) multiplier).

Example:
With auto settings, your voltage would be: 0.95V at idle, and 1.3V at load.
Now suppose I know my cpu is stable at load at 1.2V.
I would like to apply an offset of -0.1V. That would case my cpu to run at 1.2V at load.
The side-effect is: with -0.1V offset, my idle voltage will drop to 0.85V.
And that's too low. My cpu will crash at idle.

Suppose I could set my offset: -0.05V at idle and -0.1V at load ?
That would be perfect !
My board has an "additional turbo voltage" setting.
Not 100% sure what it does. But it seems like you can get extra voltage at turbo. Not sure if the turbo value replaces the normal offset, or adds to it.
So what I would need is: regular offset: -0.05V and turbo offset: another -0.05V. (Or, if it replaces the other value, I obviously need -0.1V turbo offset).

The problem ?
The "additional turbo voltage" setting does not allow negative values.
I don't see an easy solution.

(And to be clear: just like the OP, I also care about power usage at idle, power savings, etc).

 

[john3850]

On my ASRock Z77 Extreme4 the best I found was to leave (extra voltage at turbo on auto) because the turbo is always changing at idle.

I disabled turbo by turning c1e off and set the board up for the best oc then I turn turbo back on.

Above x46 I need llc-1 or the highest and PLL Overvolt enabled.
Below x46 I use llc-2 Overvolt disabled.
idle 0.900-0.920
load 1.240-1.234 at 46ghz
1.295V -1.310 for 4.7ghz

 

[tracerbullet]

Couple thoughts - My ASRock Z77 board has plenty of - (negative) turbo offset options as well as positive. I'd have guessed all boards had this, perhaps not?

Gryz - If you don't boot with a particular - offset, don't necessarily assume that the offset value itself is too low. If I set too high of a multiplier, say 45, with too low turbo/offset/LLC combination, I wouldn't boot. But if I dialed only the multiplier back down to stock, with *no* other changes, I'd boot just fine. You wouldn't think a high multiplier would be a problem booting or until you ran say Prime95, but for me, it kept me from booting. I guess I'm saying - when you are dancing the dance, make sure you are keeping your multiplier at stock to truly find out if your offset is too low.

Also, what you're talking about is kinda / sorta how offset and turbo work. Well, for most of our board / CPU combinations anyhow... You can dial back offset until you get some vcore value that can't sustain an idle, and then of course bump it back up until you do. Now that your low end is set, you just do an additional turbo + voltage to make up for what is needed at load. (LLC is in there too, if you adjust it the other values change as well, IMO find a value and stick with it).

Idontcare - I mentioned before, disabling C3 and C6 made me stable at idle, and so purely from an O/C standpoint, I was glad I did that. I still have C1E, speedstep, etc. enabled, and my rig goes back to an idle state at around .9xx-ish Volts. I understand the need to keep the power draw low, owning an old house with window A/C units... and C3/C6 disabled still allows me to go to idle and have the correspondingly low power draws. I fundamentally agree that it's best to not disable any of that, but - as for these two (C3/C6), it doesn't seem to make a noticeable difference.

 

[Idontcare]

Idontcare - I mentioned before, disabling C3 and C6 made me stable at idle, and so purely from an O/C standpoint, I was glad I did that. I still have C1E, speedstep, etc. enabled, and my rig goes back to an idle state at around .9xx-ish Volts. I understand the need to keep the power draw low, owning an old house with window A/C units... and C3/C6 disabled still allows me to go to idle and have the correspondingly low power draws. I fundamentally agree that it's best to not disable any of that, but - as for these two (C3/C6), it doesn't seem to make a noticeable difference.

I'm not arguing against doing it (disabling power features) when you know for certain what doing it is gaining you in return.

You knew exactly what you stood to gain by disabling the features (clockspeed and Vcc wise), and what you stood to lose (power savings wise)...making a data-driven decision is always best.

What I do argue against is the wholesale approach where the OC'er defaults to disabling them simply because they are told to do so without ever assessing or characterizing the value added (or lost) by taking that course of action.

 

[tracerbullet]

Cool, I can agree w/ that 100%.

 

[coffeejunkee]

I'm not arguing against doing it (disabling power features) when you know for certain what doing it is gaining you in return.

You knew exactly what you stood to gain by disabling the features (clockspeed and Vcc wise), and what you stood to lose (power savings wise)...making a data-driven decision is always best.

What I do argue against is the wholesale approach where the OC'er defaults to disabling them simply because they are told to do so without ever assessing or characterizing the value added (or lost) by taking that course of action.

I never agreed with the overclockers turn off speedstep/c1e mantra. I also care as much about low power as I do about high freq. But there is no way around it that keeping C3/C6 states enabled beyond a certain frequency causes instability, at least in my case. Stresstesters don't quite catch it and I also have no problems during idle, but some games just keep crashing or causing whea errors, no matter how much vcore i apply.

That said, the problem with idle crashes imho is using negative (or very low) offset in combination with high llc settings. But there is no reason to do this, vdroop is much less with offset voltage compared to manual voltage. For 4.5GHz I use +0.135V offset and regular llc: idle 1.024, load 1.240. For manual I need 1.265V with high llc which gives the same 1.240 under load.

 

[Grooveriding]

I leave C1 enabled and disable 3 & 6. 3 & 6 caused me instability at my overclock. You need C1/speedstep enabled to allow downvolting/clocking, at least on my motherboard that is how it behaves.

 

[Vectronic]

Granted I don't have that much experience with new boards/CPU's... but of the ones I have used, C1E =/= SpeedStep/EIST... So I often get confused when that is brought up.

If I disable C1E it (typically) reduces those crashes caused at idle speeds (or probably specifically those very brief idle > middle > idle clock jumps caused by an application/driver)... but the CPU still downclocks (and thus reduces voltage if using offset).

On my most used board (ASRock Z77 Ext4)... C3/6 have a huge chaotic effect that makes it much harder to get stable. Likely because of the quality of the board, rather than intrinsic behaviors of those states/functions.

From my perhaps false or simplistic understanding, C1E is more like park vs in gear, ie: full speed, or off where it's only really enabled/used in "Sleep"... whereas EIST/SpeedStep is more like a throttle, but always in gear. If the computer is always on, then C1E should never really kick in, if you use Sleep, then C1E might be useful for reducing power consumption.

C1E only comes into effect, or causes problems when I'm at about 130% overclock, otherwise it doesn't seem to cause any problems.

 

[Idontcare]

On my most used board (ASRock Z77 Ext4)... C3/6 have a huge chaotic effect that makes it much harder to get stable. Likely because of the quality of the board, rather than intrinsic behaviors of those states/functions.

I wonder if C3 and C6 get borked (make your system unstable) for the same reasons we have to "enable internal PLL overvoltage" our chips above certain clockspeeds if we want the chip to be stable at those clockspeeds?

I have no idea what "enable internal PLL overvoltage" does, I know that when you do it you basically resign yourself to having the sleep functions borked in windows. But why that happens and how it messes with the other C-states is a question I have pondered.

Would we really care that C3 and C6 were disabled if it was bundled and included under the BIOS option "enable internal PLL overvoltage"? If when enabling that overvoltage the BIOS also disabled C3 and C6, for the same stability-driven purposes?

 

[coffeejunkee]

I don't think internal pll overvoltage has anything to do with it. This is what I regard as experienced overclockers say about it:

The purpose of PLL overvoltage is that it allows higher mulitpliers to boot. Generally it should be disabled for multi's less than 48x.

So just for booting Windows. If c3/c6 states don't cause instability you have a) a very good cpu, b) a very good mobo or c) just not experienced it with your workload.

 

[Gryz]

Couple thoughts - My ASRock Z77 board has plenty of - (negative) turbo offset options as well as positive. I'd have guessed all boards had this, perhaps not?

I checked again. And the Turbo Offset does not have the ability to configure negative values.

I got better cooling this weekend. And did a little more testing.

Gryz - If you don't boot with a particular - offset, don't necessarily assume that the offset value itself is too low. If I set too high of a multiplier, say 45, with too low turbo/offset/LLC combination, I wouldn't boot. But if I dialed only the multiplier back down to stock, with *no* other changes, I'd boot just fine.

I think I made the wrong conclusion when my system crashed right after boot. I assumed that the systems would be idle, and that the multiplier would be 16x on the desktop. Because what would my cpu be doing ? That seems a wrong assumption.

When I boot, the first thing I do is start CPU-Z. And it seems my system is always running at max-multiplier after boot. Even when it seems there is only the desktop, and nothing else running. The weird thing is, multiplier is max, but the voltage is the voltage you would normally see when idle ! Then after 30-60 seconds (90 sometimes even), the multiplier would drop to 16x. Next time there is real load, the multiplier and voltage would go up as expected. And drop again when the load is over.

But the first 30-90 seconds are weird. No load, but high multiplier and low voltage. I think this is what crashes my system after boot if I set the offset too low.

I tested 43x. I had IBT and Prime95 running for an hour. But if I reboot, and start Firefox during that first 30-90 second window, my Firefox would crash. Repeatedly. I think it is because the system is running 4300MHz with 0.944V and some load. Weird. I don't think this is proper behaviour. Right after boot the system should either:
1) run 16x and 0.9xx volt at idle. Or
2) run 43x and 1.2V. Even at idle. And then drop down later.
But running at 43x with only 0.9xx volt is asking for problems.
(I'm running with the latest BIOS even, 2.80).

I decided to stay at 42x, -0.040V offset, resulting in 1.2V at load and 0.928V idle, temps stay below 60C. Good enough for me.

 

[tracerbullet]

When I was doing this recently I saw the same thing, multipliers all over during the first 30sec - minute or so after boot, but low voltages. I never really made sense of it, maybe it was only a single core going up, or some sort of PC equivalent of revving a car engine at a stoplight? Smarter people here may know. But for what it's worth you're not crazy, I saw the same thing. Or we are both crazy.

When I was testing my 3570k, I got a lot of help here from AT. Here's the thread, you might be able to compare some of your numbers: http://forums.anandtech.com/showthread.php?t=2306046&highlight=

When it comes down to it it seems that some CPU's just aren't destined to be overclocked very high. Luck of the draw. It's fun to play with and you might figure something out to get you higher, but - 42 all on it's own is pretty fast, and 1.2V is nice and easy on everything.

 

[john3850]

I checked again. And the Turbo Offset does not have the ability to configure negative values.

I got better cooling this weekend. And did a little more testing.

I think I made the wrong conclusion when my system crashed right after boot. I assumed that the systems would be idle, and that the multiplier would be 16x on the desktop. Because what would my cpu be doing ? That seems a wrong assumption.

When I boot, the first thing I do is start CPU-Z. And it seems my system is always running at max-multiplier after boot. Even when it seems there is only the desktop, and nothing else running. The weird thing is, multiplier is max, but the voltage is the voltage you would normally see when idle ! Then after 30-60 seconds (90 sometimes even), the multiplier would drop to 16x. Next time there is real load, the multiplier and voltage would go up as expected. And drop again when the load is over.

But the first 30-90 seconds are weird. No load, but high multiplier and low voltage. I think this is what crashes my system after boot if I set the offset too low.

I tested 43x. I had IBT and Prime95 running for an hour. But if I reboot, and start Firefox during that first 30-90 second window, my Firefox would crash. Repeatedly. I think it is because the system is running 4300MHz with 0.944V and some load. Weird. I don't think this is proper behaviour. Right after boot the system should either:
1) run 16x and 0.9xx volt at idle. Or
2) run 43x and 1.2V. Even at idle. And then drop down later.
But running at 43x with only 0.9xx volt is asking for problems.
(I'm running with the latest BIOS even, 2.80).

I decided to stay at 42x, -0.040V offset, resulting in 1.2V at load and 0.928V idle, temps stay below 60C. Good enough for me.

If you use ibt or real temp the data is updated more often then cpuz and you will see that you cpu multiplier is always moving and not staying at x16 at idle.

 

[Gryz]

If you use ibt or real temp the data is updated more often then cpuz and you will see that you cpu multiplier is always moving and not staying at x16 at idle.

You were right. RealTemp shows constantly different values, between 1600 and 4200 MHz. Where CPU-Z shows a constant 4200 MHz. RealTemp shows an (estimated) multiplier too. But I don't trust that, the value isn't an integer.

I wonder why. One reason I can think of is:
After boot, one core is running constantly at 4.2GHz. While other cores change their frequency.
Then RealTemp shows the average of the frequencies/multipliers of all 4 cores.
While CPU-Z only shows the frequency/multiplier of the highest running core.

Of course I know of no way how to check this for sure. As Windows has terrible debugging and informational utilities.

The weird thing is still: during the first 30-90 seconds after boot, the frequency of all cores, or at least one core, still peaks higher than 1.6GHz. But the voltage stays at 0.9xx volt. Too low, it seems.

 

[coffeejunkee]

Try HWinfo64, it shows multiplier per core, vid and also vcore as read by mobo's io-chip (same value cpu-z reads). I noticed the same behavior on boot with the cpu running full speed while it's loading drivers and other background stuff. But I'm fairly sure the vcore just isn't updated fast/often enough because the vid value does in fact change to regular load levels.

So I'm not sure this is actually the problem, and it definitely isn't for me. But I don't use llc when using offset since I don't see the need.

 

[Vectronic]

>1.6GHz but ~0.900 volts is do-able.

At 1.6GHz my 3570K will run at the minimum my board allows (0.6v), and that's under stress/load at 1.6GHz.

Just because the GHz is high, doesn't necessarily mean the CPU is being *used*, that's why there are temperature differences between P95 and LinPack yet both are "maxing" out the CPU.

I assume it's because it might be fully loaded as far as instructions/commands/pipeline go, but the specific instructions it's running are simple.

Mine uses 0.928v at 3.4GHz (locked/max)... so if the CPU is around that area... 1.6 to 3.4... ~0.900 might be ok... 3.6GHz...maybe not.

 

[john3850]

I trust the graph on IBT at least it tells you how many active cores ,there speed etc.
When I enable C3 the multiplier and vcore jump around even more or maybe update more often.
The graph runs as long as a hour or as short as a minute.
On my x58 or the 1rst generation turbo the multiplier and vcore move the lease.
The i7 ib or 3rd generation turbo jumps the most.

 

[tracerbullet]

How do you get IBT to run? Anything I've downloaded yells about it not being able to test the 3570k and then quits. That's right off the Intel site.