Offset Voltage...

[Puppies04]

It is unfortunate that my programming capabilities are so extremely limited, and I'm not just trying to be humble there as I really live on a bubble when it comes to my coding capabilities, but it is clear that we enthusiasts could really use a software tool that (1) heated the processor up as LinX does, and simultaneously (2) executed computations on every single instruction in the ISA in various permutations (like memtest86+ does) while checking results to validate operation across the entire ISA at that temperature, voltage, and clockspeed.

Even better would be if this program would be able to tap into the BIOS to drive an auto-tune overclock program that mapped out the shmoo plot for any given unique combo of a consumer's CPU/mobo/HSF/ambients so we just launched the program and walked away.

Come back and you have a pretty graph that shows you your stable clockspeeds versus voltage, temperature, and CPU power usage to choose from. And then the BIOS would absorb those value and implement a truly customized turbo-clock profile that ramped the entire spectrum of multipliers.

If I could program I would endeavor to create such a utility. Alas, I suck in this dept

I fully agree that this would be an awesome program for somebody to write. They could charge good money for it as well and then I could torrent it for free.

 

[Zen Ninja]

Quote:
Originally Posted by Zen Ninja
Any ideas as to why my voltages are so high? Is there a setting I'm missing maybe?

Phase control?
No, but I would start with running at stock values and checking vcore under different loads (1,2,3 & 4 threads). Next overclock to say 4GHz on auto vcore to see what the mobo supplies in that case. I haven't found the phase control option to make much difference.

I did that--that's how I found my offset. I found Supplied load - my manual stable load vcore and used that as my negative offset. When I tested 4.0 on Auto it was providing 1.28v, however, I had it stable at 1.12v--I then found 4.2 stable at 1.14v (with auto providing 1.31v) etc...I'm not crazy so I wont use a -0.170 offset so I accepted -0.07 and so on.

 

[Zen Ninja]

I'm starting to think there's something wrong with my OC--I did as you said and reverted back to factory settings (just the OC) and tested the chip just to get an idea and my idle temps are 21-27C and 100% load are 48-57C (faactory is 1.6Ghz idle and 3.8Ghz load).

The odd thing is the voltage is HIGHER on default idle but the temps are lower...? I'm confused. I miss the old days of OC'ing where you just find a stable vcore for OC and EIST would find a stable vcore for idle.

I also noticed that on default BCLK frequency is 100.53 and not 100.00--is that bad? I had read on tomshardware.com forums that changing BCLK frequency is bad with IB chips--I used to do it with old chips (no choice in most cases).

 

[Ed1]

Yes, but I think the difference is too big. TS adds +0.005 and gets 1.30, I add +0.065 and get 1.18.

And indeed, the offset is what's added to or substracted from the auto value.



I agree with you. But the thing is, I overclocked a bunch of Core2 cpu's and if they could succesfullly survive a couple hours of Prime95 the system would be stable for all my purposes. For my i5 750 I mainly used LinX, and again, if it could pass 20 runs the system would be stable for all my purposes. Now with my i5 3570K I get weird errors/crashes no matter how long I test with Prime or LinX. So I don't bother with them anymore beyond a very quick 10 minute test to get a rough idea of stability.



No, but I would start with running at stock values and checking vcore under different loads (1,2,3 & 4 threads). Next overclock to say 4GHz on auto vcore to see what the mobo supplies in that case. I haven't found the phase control option to make much difference.

no, auto does not = VID , run core temp it gives VID .
For example I run auto and my Vcore at 100% load is 1.112v but my VID = 1.175v so its not same .

For mild OC (under 4.2 or so) set LLC to lowest and use as high -offset to get same stable low vcore as you did manually . with regard to 1.6ghz idle voltages i think your fine as long as above 0.900v but I guess that depends on chip too but worth testing .

 

[2is]

I'm starting to think there's something wrong with my OC--I did as you said and reverted back to factory settings (just the OC) and tested the chip just to get an idea and my idle temps are 21-27C and 100% load are 48-57C (faactory is 1.6Ghz idle and 3.8Ghz load).

The odd thing is the voltage is HIGHER on default idle but the temps are lower...? I'm confused. I miss the old days of OC'ing where you just find a stable vcore for OC and EIST would find a stable vcore for idle.

I also noticed that on default BCLK frequency is 100.53 and not 100.00--is that bad? I had read on tomshardware.com forums that changing BCLK frequency is bad with IB chips--I used to do it with old chips (no choice in most cases).

Your BCLK is fine, as long as you're below 102MHz you're ok. I think whatever you're using to read vcore is simply not accurate. With that said, what are you using?

 

[Zen Ninja]

Aida64 CPU ID, basic CPU-Z, CoreTemp and HWInfo64.

They all say the same thing.

 

[2is]

Odd, everything suggests the readings are inaccurate. Everything from the parity between your system and mine (as well as others who have posted) to your tempo/voltage results by going to stock.

btw, have you updated the bios?

 

[Zen Ninja]

Yes--I just built this system Wednesday and updated the BIOS then.

 

[Zen Ninja]

Could just be a bad chip in comparison to most.

I don't mind in all honesty--I've never kept a PC longer than 2 years.

Not because they break either--I just love doing massive upgrades every 2 years.

 

[Puppies04]

It is unfortunate that my programming capabilities are so extremely limited, and I'm not just trying to be humble there as I really live on a bubble when it comes to my coding capabilities, but it is clear that we enthusiasts could really use a software tool that (1) heated the processor up as LinX does, and simultaneously (2) executed computations on every single instruction in the ISA in various permutations (like memtest86+ does) while checking results to validate operation across the entire ISA at that temperature, voltage, and clockspeed.

Even better would be if this program would be able to tap into the BIOS to drive an auto-tune overclock program that mapped out the shmoo plot for any given unique combo of a consumer's CPU/mobo/HSF/ambients so we just launched the program and walked away.

Come back and you have a pretty graph that shows you your stable clockspeeds versus voltage, temperature, and CPU power usage to choose from. And then the BIOS would absorb those value and implement a truly customized turbo-clock profile that ramped the entire spectrum of multipliers.

If I could program I would endeavor to create such a utility. Alas, I suck in this dept

I have just been reading this months issue of PC Format and it has a very interesting article in it regarding running stress tests inside linux which purport to do what you propose in the highlighted text in your quote. Let me know if you are interested and I can post some more info, my linux experience is extremely limited but the article in question seems to be written in a very easy to understand way.

 

[coffeejunkee]

no, auto does not = VID , run core temp it gives VID .
For example I run auto and my Vcore at 100% load is 1.112v but my VID = 1.175v so its not same .

For mild OC (under 4.2 or so) set LLC to lowest and use as high -offset to get same stable low vcore as you did manually . with regard to 1.6ghz idle voltages i think your fine as long as above 0.900v but I guess that depends on chip too but worth testing .

Ok, I'm confused now, I always though that the auto value was the same as the VID but apparently that's not the case. Now I'm not sure what the VID actually does or how the auto value is established.

Some numbers:

stock freq
VID : 1.161-1.176
auto vcore : 1.056-1.080

4.0 GHz
VID : 1.231-1.236
auto vcore : 1.112-1.128
my vcore : 1.072-1.096 (using -0.045 offset, 1.112-0.045=1.067, close enough for me)

4.3 GHz
VID : 1.241-1.246
auto vcore : 1.240-1.272
my vcore : 1.176-1.192

For this last one i use a +0.065 offset, but apparently despite being a positive offset it should actually be subtracted from the VID, 1.241-0.065 is exactly 1.176. That's nice but it makes no sense for my 4 GHz oc.

Anyway, I don't bother with fixed vcore anyway and just increase or decrease offset value till I have a stable oc. I guess I advise anyone to do the same.

I'm starting to think there's something wrong with my OC--I did as you said and reverted back to factory settings (just the OC) and tested the chip just to get an idea and my idle temps are 21-27C and 100% load are 48-57C (faactory is 1.6Ghz idle and 3.8Ghz load).

The odd thing is the voltage is HIGHER on default idle but the temps are lower...? I'm confused. I miss the old days of OC'ing where you just find a stable vcore for OC and EIST would find a stable vcore for idle.

I also noticed that on default BCLK frequency is 100.53 and not 100.00--is that bad? I had read on tomshardware.com forums that changing BCLK frequency is bad with IB chips--I used to do it with old chips (no choice in most cases).

The vcore is continually fluctuating at idle so you might have catched a peak. Besides, c-states shut off parts of the cpu which makes for low temps even with fixed vcore. Lastly, the sensors are calibrated to give the most exact results when nearing tjmax throttle temp and Idle temps are not very accurate (sometimes I get lower than ambient temps which obviously can't be right).

But EIST never found a stable vcore for idle. If you use a fixed voltage there is no way EIST can lower it during idle. This is only possible when using offset vcore (or dynamic vcore as Gigabyte calls it). Before P35/P45 chipsets this option was not available at all.

Concerning bclk, what you see is the spread spectrum option at work, which is only really needed when you have a room full of computers. If you disable this the bclk will remain static. Tom's Hardware was also right, but they're talking about actual bclk overclocking, which is indeed very limited because this bclk is also used for pci-e slots, sata ports etc which have very small oc-ing margins.

 

[Ed1]

Ok, I'm confused now, I always though that the auto value was the same as the VID but apparently that's not the case. Now I'm not sure what the VID actually does or how the auto value is established.

Some numbers:

stock freq
VID : 1.161-1.176
auto vcore : 1.056-1.080

4.0 GHz
VID : 1.231-1.236
auto vcore : 1.112-1.128
my vcore : 1.072-1.096 (using -0.045 offset, 1.112-0.045=1.067, close enough for me)

4.3 GHz
VID : 1.241-1.246
auto vcore : 1.240-1.272
my vcore : 1.176-1.192

For this last one i use a +0.065 offset, but apparently despite being a positive offset it should actually be subtracted from the VID, 1.241-0.065 is exactly 1.176. That's nice but it makes no sense for my 4 GHz oc.

Anyway, I don't bother with fixed vcore anyway and just increase or decrease offset value till I have a stable oc. I guess I advise anyone to do the same.



The vcore is continually fluctuating at idle so you might have catched a peak. Besides, c-states shut off parts of the cpu which makes for low temps even with fixed vcore. Lastly, the sensors are calibrated to give the most exact results when nearing tjmax throttle temp and Idle temps are not very accurate (sometimes I get lower than ambient temps which obviously can't be right).

But EIST never found a stable vcore for idle. If you use a fixed voltage there is no way EIST can lower it during idle. This is only possible when using offset vcore (or dynamic vcore as Gigabyte calls it). Before P35/P45 chipsets this option was not available at all.

Concerning bclk, what you see is the spread spectrum option at work, which is only really needed when you have a room full of computers. If you disable this the bclk will remain static. Tom's Hardware was also right, but they're talking about actual bclk overclocking, which is indeed very limited because this bclk is also used for pci-e slots, sata ports etc which have very small oc-ing margins.

That is odd results , I have no explaination for that .
What MB are you using ?

My Vcore jumps around to at idle, anytime any kind of load from a process happens it can increase the vcore along with CPU frequency

 

[Idontcare]

I have just been reading this months issue of PC Format and it has a very interesting article in it regarding running stress tests inside linux which purport to do what you propose in the highlighted text in your quote. Let me know if you are interested and I can post some more info, my linux experience is extremely limited but the article in question seems to be written in a very easy to understand way.

Post it up, I'd like to read it just so I am aware of what is out there on people's minds. We can dream, right?

 

[Puppies04]

Post it up, I'd like to read it just so I am aware of what is out there on people's minds. We can dream, right?

I can't link to the article as it is only available in hard copy or digital subscription but the part of the article I thought you might be interested in is as follows (I will post a couple of paragraphs verbatim, if you need any more info then just ask)

Before I start a couple of pieces of info, the guide says any "flavour" of linux is fine but recommends the latest Ubuntu as it works with most pc hardware and also advises you install a linux package called 'lm-sensors' to monitor the temperatue of your hardware.

That said here we go....

'stress' isn't a comprehensive solution, but it is going to be powerful enough for most circumstances. Also, because it is designed specifically to put load on your system, it does a better job at using the entire instruction set of your CPU than 'Bash' commands, (Bash was a simpler stress test alluded to earlier in the article) as well as other components, such as memory and storage, if you let it. Once installed typing stress on the command line will output a simple set of arguements for its use. Adding --cpu 2, for example, will spawn two tasks constantly working out the square route of a number. You can add memory allocation/de-allocation with the --vm 1 arguement, followed with --vm-bytes 128M for the amount of memory you want to process to play with. Finally, if you want to add some filesystem input/output to the mix, add the --io 1 argument. This will force the filesystem to constantly flush it's buffers, as the linux community likes to say.

Putting this together you should use this command : stress --cpu 2 --vm 1 --vm-bytes 128M --io 1. After running this command, you should find that your system is operating under considerable load, hopefully to the point where it might become slightly unresponsive, although not to much ; the linux scheduler is excellent at maintaining interactive sessions while experiencing periods of heavy load. If you want to understand a little more about what the 'stress' command is doing, add the --verbose arguement. this will provide some background information on what the tool is actually doing to your system, as well as report on how the processes are running. But if you want to see more about how your system is coping with the load, you'll need to look at a couple of system utilities.

The article goes on to talk about how to check how much stress your system in under in real time as well over the last 1, 5, and 15 minutes and also talks very highly of a linux based stress test that "put our test system under the most amount of stress" called BSMBench ( http://bsmbench.org ). It also talks about a particular version of linux called "StressLinux" which is idea to run from a liveCD to burn in and test all aspects of a new rig before you install windows to make sure that all your hardware is bulletproof including CPU, GPU, storage, memory and networking.

 

[Puppies04]

Instead of using LLC, just bump up (instead of down) the offset. It will achieve the same results under load without compromising voltage during idle states.

I find this comment misleading, adding positive offset effects the voltage delivered to the CPU at load, idle and anywhere in between as far as I am aware so while I understand the perils of -offset at idle causing bluescreens + offset isn't the magic bullet you make it out to be.

 

[BonzaiDuck]

I'm just dropping in here for minute . . . noticed that more than one poster recommended using only the Offset -/+ and increment to find a stable voltage.

In my original SB over-clock, The Offset was configured to +0.010 or even +0.015 if the increments were 0.005. I read some more OC guides while further tuning my system to put the Offset between - 0.005 and +0.005 -- one notch away from "0."

I've also mentioned in other posts a third voltage (referenced by those OC Guides as well) called "Extra Turbo Voltage" or "Extra Voltage at Turbo," which only adjusts in positive increments. This voltage adjusts in positive increments of 0.004, so you can make 0.002 increments by adjusting the Offset between - and + the single 0.005 notch from zero, with one or two notches either way for this third voltage setting. In the original Z68 BIOS for my P8Z68-V Pro board, (or versions released through end of 2011), this third voltage adjustment was found under "Power Management."

I agree that you should be able to adjust either an SB or IB processor to 4.2 without LLC. In my case at 4.6, I pushed LLC to a "Medium" or "High" setting which left about 0.02V of droop at full load. So under light load, the maximum voltage would be somewhere between 1.35 and 1.36, while full load would give about 1.328 to 1.33+V. But those numbers are for Sandy Bridge.

Also note that I increased the voltage until LinX GFLOPS had minimum variation at maximum consistent readings -- running LinX with options and settings like IDontCare had explained in some of his own previous posts.

Wasn't there a consensus that a reasonable maximum for Ivy Bridge was 1.30V or 1.30+V?

I don't have any stability problems.

 

[2is]

I find this comment misleading, adding positive offset effects the voltage delivered to the CPU at load, idle and anywhere in between as far as I am aware so while I understand the perils of -offset at idle causing bluescreens + offset isn't the magic bullet you make it out to be.

It's not misleading. Yes positive offset does what you just said it does. Negstive offset does the same thing in the oposite direction. Lowers voltage across the frequency spectrum. LLC increases voltage under full load situations by minimizing vdroop. Negative offset in combination with LLC compromises voltage at low/idle loads, which CAN lead to instability. It's not magic, it's fact.

Most stress testing is done under load where LLC is providing a boost and everything appears stable. During idle or nearly idle states LLC isn't doing anything but your negative offset may have the voltage lower than it needs to be for stability.

 

[Puppies04]

Ok, let me try that again.

Instead of using LLC, just bump up (instead of down) the offset. It will achieve the same results under load without compromising voltage during idle states.

The reason I said it was misleading is because your statement is wrong and could give someone the impression that +offset only increases voltage delivered under loaded situations.

You specifically claim +offset will not compromise voltage during idle states when in actual fact it does, as I pointed out it raises voltage at idle as well as loaded situations and running a CPU with more voltage than it requires at any time is not ideal.

As I said I understand why it is the lesser of 2 evils compared to running with a -offset and LLC turned on but the fact remains that +offset will effect idle voltage compared to leaving it at 0 offset and LLC turned off which would be the idea situation until instability was evident.

P.s I'm not trying to split hairs here. I know you understand what you are talking about I was simply trying to point out that to someone unfamiliar with bios voltage adjustment that your comment could be read in a misleading way.

 

[2is]

Gotcha, by "compromise" I mean reduce to the point of potential instability.

Setting at full auto generally increases voltage more than +offset, at least it does on my system. Some people say offset adds voltage to the "auto" value, but that does not appear to be the case for me. I have my offset at +.020 (or maybe its .025, don't recall) and at that value, my vcore readings are lower than setting it to full auto at my current OC.

 

[Puppies04]

Gotcha, by "compromise" I mean reduce to the point of potential instability.

Setting at full auto generally increases voltage more than +offset, at least it does on my system. Some people say offset adds voltage to the "auto" value, but that does not appear to be the case for me. I have my offset at +.020 (or maybe its .025, don't recall) and at that value, my vcore readings are lower than setting it to full auto at my current OC.

I see, I take it that setting your voltage to "offset mode" and less than +.020 resulted in instability?

My 2500K is now dialed down from its original OC to 4.2ghz and running with a very slight -offset and no LLC and hasn't given me any indication of instability for months other than a bluescreen which was caused by the old firmware on my SSD.

 

[coffeejunkee]

That is odd results , I have no explaination for that .
What MB are you using ?

My Vcore jumps around to at idle, anytime any kind of load from a process happens it can increase the vcore along with CPU frequency

P8Z77-M Pro. But I think it's working like it's supposed to and for positive offset values your formula is totally correct. I think Asus way of offset voltage is just a bit unintuitive. Gigabyte uses a better system imho, where you can simply lock the cpu to its stock voltage and then either add or decrease volts. But in the end Asus solution works too and that's what matters.

P.s I'm not trying to split hairs here. I know you understand what you are talking about I was simply trying to point out that to someone unfamiliar with bios voltage adjustment that your comment could be read in a misleading way.

You are right, both + and -offset affects idle voltage. But indeed not worth splitting hairs, since there is no need at all to use llc when using a negative offset.

 

[Puppies04]

You are right, both + and -offset affects idle voltage. But indeed not worth splitting hairs, since there is no need at all to use llc when using a negative offset.

My experiences seem to differ from yours then, I specifically remember my first 2500K giving instability warnings in IBT at around 4.4ghz while running a -offset while being fine at idle. It ran for 6+ months with -offset and LLC set to low although I understand I could have increased the offset voltage and turned the LLC off.

It comes down to personal preference at the end of the day.

 

[2is]

I see, I take it that setting your voltage to "offset mode" and less than +.020 resulted in instability?

My 2500K is now dialed down from its original OC to 4.2ghz and running with a very slight -offset and no LLC and hasn't given me any indication of instability for months other than a bluescreen which was caused by the old firmware on my SSD.

I can probably go .010 lower and still pass all the usual stress tests but I like to have a stability cushion. At my current voltage I can pass all my stress tests at 4.3GHz, at 4.4, IBT will generate an error so I actually settled for 4.2. I have enough headroom for at least another couple hundred MHz with the proper tweaks but its just not necessary. Once it starts showing its age then ill bump it higher and maybe even delid for another couple hundred MHz on top of that.

I certainly can't run negative offset, that lowers my voltage to below stock voltages.

 

[Puppies04]

I suppose it comes down to luck of the draw tbh, I managed to get hold of a later 2500k which requires significantly less voltage than my original one. I now run a small -offset and LLC turned off with no problems at all. Intel build a decent "cushion" into their voltage tables to allow for issues like ageing and very high temps which can both increase the required voltage by a small but measurable amount so I'm somewhere within that buffer zone at the moment with no discernable instability.

That said I don't have to use that machine for professional use so I don't have to worry about silent data corruption even though I am yet to pick up any signs that would point towards it being a problem

 

[coffeejunkee]

My experiences seem to differ from yours then, I specifically remember my first 2500K giving instability warnings in IBT at around 4.4ghz while running a -offset while being fine at idle. It ran for 6+ months with -offset and LLC set to low although I understand I could have increased the offset voltage and turned the LLC off.

It comes down to personal preference at the end of the day.

Ehm, what I meant was if you're using a -offset you're obviously running a very modest overclock or even undervolting/underclocking in which case vdroop is never a problem because one doesn't use a high vcore in those circumstances.

But I didn't expect that a 4.4 Ghz oc would be (almost) possible with a -offset, although llc ups the vcore. I myself need +offset (without llc) for 4.2GHz and higher.