LOW voltage .86V on 2600k

[Hogan773]

So I've been testing an "ultrasafe 24/7 OC" that I can leave running with no guilt in terms of voltage degradation. Right now I've got 42X on -.05V offset, which is giving me a Vcore of about 1.21 on load, but at idle it can drop as low as .86V!!! (actually HWMonitor shows a minimum of 0.80V at some point this afternoon.....)

So my question is, is there any problem with the LOW end voltage at idle, or will I just know it if I see it (ie BSODs at random times while using the PC)

I ran 3 hours of Prime, plus Cinebench. May try Intel BT too.

Since my stock Vcore is around 1.18V, I'm VERY comfortable if I can keep max around 1.21 or so, and then I'm still getting a nice little everyday pop to 42X from 35X at stock, and my CPU shouldn't be getting any voltage aging at all.....

Any thoughts?

 

[Castiel]

That's perfectly fine as far as I know. The lower the better

 

[OVerLoRDI]

Software monitors are known to give you unreliable voltages.

That being said, if it works, low voltage won't hurt things. The worst that will happen is the system crashes, but it shouldn't damage the chip.

 

[Hogan773]

I have CPU-Z 1.56 which I think is accurate....?

But yeah I guess I was just surprised by something so low and still working...but at idle I guess it doesn't need much juice.....as soon as it kicks in gear then it shoots up to 1.20 area

 

[MrTransistorm]

I have CPU-Z 1.56 which I think is accurate....?

But yeah I guess I was just surprised by something so low and still working...but at idle I guess it doesn't need much juice.....as soon as it kicks in gear then it shoots up to 1.20 area

You need at least 1.56.3 or .4 to be accurate. Previous versions showed strange voltages on Sandy Bridge. Also, the current HWMonitor is way out of whack as far as voltage reporting goes. If you want something similar, try HWiNFO32. Latest betas of monitoring/benchmarking software are available here.

If you're not having any problems when it switches to low-voltage idle, then keep at it. My 2600K runs at 1.404V at 4.7GHZ under load, and it drops to 0.972V at 1.6GHz when idle. No problems when switching back and forth. I have all power-saving features enabled.

 

[Blastman]

I had my i3-530 running at 0.86v at idle (with a -0.05 offset) and I had no stability problems. The i3-530 is a 0.32 process technology chip like the new SB's. Although, from reading a bunch of overclocking forums on the i3's I think mid to low 0.80's is probably getting close to the lower limit at idle. As long as it's stable, no harm, you're good to go.

 

[Idontcare]

Chips can run fully loaded at those voltages provided the max clockspeed is lowered.

Heck my rather old 65nm QX6700 would run Prime95 small FFT stable with 0.86V at 1.6GHz.

No harm comes from lowering the operating voltage, just understand your max stable clockspeed at any given operating temperature will be lowered because of it is all.

 

[Accord99]

One thing to test also is to use only 1 thread with Prime95/LinX. For certain voltage configurations and 1155 MBs, the CPU voltage when 4 cores are loaded is higher than when only 1 core is loaded. This causes the unexpected problem where the CPU is stable when all 4 cores are loaded but BSODs when only 1 or 2 cores is loaded.

 

[mb103051]

my x58 dfi/920 idles at 0.91v and goes up to 1.23v at 3.8mhz with turboboost ...im amazed at the low voltage and it never crashes and its been running like this for at least the last 6 months....I really like this x58 rig as its been bullet proof from the first day i built it 14 months ago....

 

[Hogan773]

You need at least 1.56.3 or .4 to be accurate. Previous versions showed strange voltages on Sandy Bridge. Also, the current HWMonitor is way out of whack as far as voltage reporting goes. If you want something similar, try HWiNFO32. Latest betas of monitoring/benchmarking software are available here.

If you're not having any problems when it switches to low-voltage idle, then keep at it. My 2600K runs at 1.404V at 4.7GHZ under load, and it drops to 0.972V at 1.6GHz when idle. No problems when switching back and forth. I have all power-saving features enabled.

Wait so now I'm confused. The ACTUAL CPUID website just shows 1.56 (Oct 2010) and then there is a different link for the Sandy Bridge Reviewers' Toolkit which includes 1.56.1. I believe that I am using 1.56.1

So why then are there 1.56.4 and .3 available on Tweaktown and other places but not even available from the actual CPU-Z site? Are these just mods to the CPU-Z program that aren't sanctioned by CPUID?

My CPU-Z and HW Monitor generally agree with each other PLUS agree with the ASRock Extreme Tuning Utility in terms of voltage. But now you've got me scared that I've been pumping 1.4V through this while thinking its only 1.25 or something.........

Pls explain re: version numbers and download sites - thanks

EDIT - I recall something that this new version of CPU-Z was only in response to making sure the voltages read correctly from GIGABYTE boards....is that right? (I have ASRock). Plus the link you sent shows a 1.56.4 (32 bit) - I have Windows 64 bit - does that matter?

 

[Acanthus]

Chips can run fully loaded at those voltages provided the max clockspeed is lowered.

Heck my rather old 65nm QX6700 would run Prime95 small FFT stable with 0.86V at 1.6GHz.

No harm comes from lowering the operating voltage, just understand your max stable clockspeed at any given operating temperature will be lowered because of it is all.

This, i did it for fun with the rig in sig years ago and got it down to 0.85v @ 1.2ghz

 

[Hogan773]

Chips can run fully loaded at those voltages provided the max clockspeed is lowered.

Heck my rather old 65nm QX6700 would run Prime95 small FFT stable with 0.86V at 1.6GHz.

No harm comes from lowering the operating voltage, just understand your max stable clockspeed at any given operating temperature will be lowered because of it is all.

Yeah the way I look at it, I'm trying to get the MAX CPU speed at something close to the stock voltage. This way I can rationalize that I'm not doing any more "voltage damage" than I would have otherwise at stock settings, but I'm getting faster performance (42x vs 35x on stock).

 

[MrTransistorm]

So why then are there 1.56.4 and .3 available on Tweaktown and other places but not even available from the actual CPU-Z site? Are these just mods to the CPU-Z program that aren't sanctioned by CPUID?

They are beta versions. Tweaktown just happens to be a good source of betas (including BIOS). I think some of the developers are frequent posters there (or maybe that was HWBot. I don't remember for sure).

EDIT - I recall something that this new version of CPU-Z was only in response to making sure the voltages read correctly from GIGABYTE boards....is that right? (I have ASRock).

You're probably right about the updates being related to Gigabyte boards. Maybe the previous versions work fine for ASRock boards. I just like to use the latest versions anyway.

Just look how messed up the voltage and fan readings are on HWMonitor compared to HWiNFO:



CPU Core should be Vtt
CPU should be Vcore
Vbat should be Vdimm
and lol 12 RPM on my CPU fan :awe:

The temps are mostly correct, though (the differences in min temps are due to starting the monitors at different times). HWiNFO has been updated several times recently, but it appears that there is no new version of HWMonitor yet.

Plus the link you sent shows a 1.56.4 (32 bit) - I have Windows 64 bit - does that matter?

The 32bit version works just fine in Windows x64. It even validates correctly on the Canard PC site:

 

[Idontcare]

Yeah the way I look at it, I'm trying to get the MAX CPU speed at something close to the stock voltage. This way I can rationalize that I'm not doing any more "voltage damage" than I would have otherwise at stock settings, but I'm getting faster performance (42x vs 35x on stock).

I hate to be the bearer of bad news but if you are thinking that lifetime is independent of clockspeed then I have some, uhm, bad news for you.

The best analogy I can think of is an incandescent light bulb.

An incandescent light bulb "burns out" because atoms migrate under the bias provided by the applied electric field (you turned on the switch). Even though the field is AC in this case it is just enough of a persistence bias that atoms will migrate along something that is called "grain boundaries" in the metal filament in the light bulb. After enough operating time the filament develops a weakspot and it literally falls apart from a mechanism called joule heating.

Your transistors wear out from a form of joule heating degradation, atoms actually move around (infrequently, but eventually) within the channel and gate of the transistor where the electric field is highest, as well as critical regions in the metal wiring for the same reasons that it happens in a light bulb.

This damage is cumulative and is dependent on both voltage as well as total exposure time (amps, as it is the flow of charge carriers that actually enable the specific redox chemistry at play in most degradation mechanisms).

Operating your chip at 4GHz versus 2GHz and at the same voltage will most certainly degrade the xtors and metal wiring in the 4GHz at a rate that is faster than that of the 2GHz case.

Going back to the light-bulb example, think of your xtor's switching as being akin to you standing at the power-switch to the lights in your room and you rapidly flip them on and off. Eventually you'll kill the light bulb doing this, flip the switch faster (clock your CPU higher) and the inevitable just comes sooner.

The question most people want to answer though is will the chip live out its useful lifespan (say 5-6 yrs). And the answer there is yes, most definitely. CPU's are designed to have lifespans that measure in decades, the stuff you are doing to your CPU is merely shortening the lifespan from 20yrs to say 15 yrs.

 

[VirtualLarry]

The question most people want to answer though is will the chip live out its useful lifespan (say 5-6 yrs). And the answer there is yes, most definitely. CPU's are designed to have lifespans that measure in decades, the stuff you are doing to your CPU is merely shortening the lifespan from 20yrs to say 15 yrs.

Are you suggesting, that it would be unwise to overclock my Q9300, from 2.5Ghz to 3.0Ghz, if I expect to keep this CPU for 10-15 years? Even at default (stock) voltage?

 

[Idontcare]

Yes.

 

[Hogan773]

I hate to be the bearer of bad news but if you are thinking that lifetime is independent of clockspeed then I have some, uhm, bad news for you.

The best analogy I can think of is an incandescent light bulb.

An incandescent light bulb "burns out" because atoms migrate under the bias provided by the applied electric field (you turned on the switch). Even though the field is AC in this case it is just enough of a persistence bias that atoms will migrate along something that is called "grain boundaries" in the metal filament in the light bulb. After enough operating time the filament develops a weakspot and it literally falls apart from a mechanism called joule heating.

Your transistors wear out from a form of joule heating degradation, atoms actually move around (infrequently, but eventually) within the channel and gate of the transistor where the electric field is highest, as well as critical regions in the metal wiring for the same reasons that it happens in a light bulb.

This damage is cumulative and is dependent on both voltage as well as total exposure time (amps, as it is the flow of charge carriers that actually enable the specific redox chemistry at play in most degradation mechanisms).

Operating your chip at 4GHz versus 2GHz and at the same voltage will most certainly degrade the xtors and metal wiring in the 4GHz at a rate that is faster than that of the 2GHz case.

Going back to the light-bulb example, think of your xtor's switching as being akin to you standing at the power-switch to the lights in your room and you rapidly flip them on and off. Eventually you'll kill the light bulb doing this, flip the switch faster (clock your CPU higher) and the inevitable just comes sooner.

The question most people want to answer though is will the chip live out its useful lifespan (say 5-6 yrs). And the answer there is yes, most definitely. CPU's are designed to have lifespans that measure in decades, the stuff you are doing to your CPU is merely shortening the lifespan from 20yrs to say 15 yrs.

Ok I didn't know that.....! Now you make me feel guilty for running it faster though. But try this on for size and see if it fits.....

Assuming I'm doing the SAME TASKS.....at the same voltage......and I do them at 42x vs 35x.....then my 42x is wearing the CPU at a higher rate BUT it is done faster, so all the quicker to head back to idling at 16x and .86V.....

so maybe I'm not really wearing it all that much differently. I agree with your logic if its a matter of 4 hours at 42x vs 4 hours at 35x. But if its 4 TASKS that are done more quickly, maybe that's different.

Not trying to trump you, but genuinely interested and I'm not an EE by any stretch....

what do you think?

 

[Idontcare]

Ok I didn't know that.....! Now you make me feel guilty for running it faster though. But try this on for size and see if it fits.....

Assuming I'm doing the SAME TASKS.....at the same voltage......and I do them at 42x vs 35x.....then my 42x is wearing the CPU at a higher rate BUT it is done faster, so all the quicker to head back to idling at 16x and .86V.....

so maybe I'm not really wearing it all that much differently. I agree with your logic if its a matter of 4 hours at 42x vs 4 hours at 35x. But if its 4 TASKS that are done more quickly, maybe that's different.

Not trying to trump you, but genuinely interested and I'm not an EE by any stretch....

what do you think?

Yep, now you got something there.

The key here being thermodynamics and the work done on the system while the system is doing work.

If you do not task the system with performing any more work, in closed loop form, at the elevated clockspeed than you aimed to perform at the lower clockspeed then the net work done by the system is the same and the net work done on the system (which we view to be degradation) is the same.

And you hit on the other key enabler of this happy outcome, idle clockspeeds. You are saying "hey I am only going to flicker my lights on and off just 50 times per day, and whether I do that at a pace that gets me done in 15 minutes or 30 minutes is up to me, but the light bulb isn't going to experience any more than 50 on-off cycles regardless".

So you nailed it, thoroughly I might add. (so refreshing to get the opportunity to interact with folks who put two and two together so quickly and come back at you with calculus questions the next day! good on ya!)

Where this special condition fails is if the user takes advantage of the newly found idle time and puts the computer to work doing even more stuff. The net work the system can do before it reaches critical degradation thresholds is pretty much fixed if you are holding all other things constant (voltage, temps, apps, etc).

 

[Hogan773]

Yep, now you got something there.

The key here being thermodynamics and the work done on the system while the system is doing work.

If you do not task the system with performing any more work, in closed loop form, at the elevated clockspeed than you aimed to perform at the lower clockspeed then the net work done by the system is the same and the net work done on the system (which we view to be degradation) is the same.

And you hit on the other key enabler of this happy outcome, idle clockspeeds. You are saying "hey I am only going to flicker my lights on and off just 50 times per day, and whether I do that at a pace that gets me done in 15 minutes or 30 minutes is up to me, but the light bulb isn't going to experience any more than 50 on-off cycles regardless".

So you nailed it, thoroughly I might add. (so refreshing to get the opportunity to interact with folks who put two and two together so quickly and come back at you with calculus questions the next day! good on ya!)

Where this special condition fails is if the user takes advantage of the newly found idle time and puts the computer to work doing even more stuff. The net work the system can do before it reaches critical degradation thresholds is pretty much fixed if you are holding all other things constant (voltage, temps, apps, etc).

Thanks for the kudos, but thanks more for now letting me take my guilt away so I can go back to my mild OC settings :awe:

I guess to split hairs even more, the mild OC might do a wee bit more "damage" to the CPU because one would assume that temps will rise higher when running at 42x vs 35x, although perhaps you can enlighten me there too.....is voltage the primary driver of temps I assume?

And so this all is a big vote for leaving the various Speedstep stuff turned on, PLUS using an AUTO or OFFSET voltage setting in the BIOS rather than a FIXED, when setting up one's overclock. No sense in making the engine run at 44x and 1.30V 24/7 if it can spend much of its time kicking back with a Corona on the beach and lolling along at 16x and 0.90V. I know there seems to be a sense from some OCers that you get your "best" OC by turning off those options and locking everything down as fixed, but I'd have to think that for the vast majority of users, there is no way they are taxing their CPUs at close to 100% for close to 100% of the time the PC is turned on, unless they are doing Folding or something. Maybe the people who are posting such stuff are veteran OCers and that is what you had to do "back in tha' day"

Thanks for educating me :thumbsup:

 

[Hogan773]

One more thing I'm interested in knowing........

Does the 2600K allow independent speeds on the 4 cores, or is it all or nothing? By that I mean, if I'm idling at 16x and then the CPU is called to task, I can see in CPU-Z that the multi then jumps up to [42] or whatever we've got set in BIOS. But does that mean that the whole chip is now running at 42x even if a program is only really using 1 or 2 cores while the others are not being used? Or can it throttle just a couple cores while letting the others stay at 16x idle? I can't really tell by looking at CPU-Z since it just shows one CPU multiplier.

I'd assume the former, as it would seem to be quite complicated to throttle up and down individual cores all the time.

Just interested to know a little more about my system

Thanks

 

[Idontcare]

Thanks for the kudos, but thanks more for now letting me take my guilt away so I can go back to my mild OC settings :awe:

I guess to split hairs even more, the mild OC might do a wee bit more "damage" to the CPU because one would assume that temps will rise higher when running at 42x vs 35x, although perhaps you can enlighten me there too.....is voltage the primary driver of temps I assume?

And so this all is a big vote for leaving the various Speedstep stuff turned on, PLUS using an AUTO or OFFSET voltage setting in the BIOS rather than a FIXED, when setting up one's overclock. No sense in making the engine run at 44x and 1.30V 24/7 if it can spend much of its time kicking back with a Corona on the beach and lolling along at 16x and 0.90V. I know there seems to be a sense from some OCers that you get your "best" OC by turning off those options and locking everything down as fixed, but I'd have to think that for the vast majority of users, there is no way they are taxing their CPUs at close to 100% for close to 100% of the time the PC is turned on, unless they are doing Folding or something. Maybe the people who are posting such stuff are veteran OCers and that is what you had to do "back in tha' day"

Thanks for educating me :thumbsup:

Yes if the temps are higher then the degradation will occur at a faster rate.

Thermally activated degradation mechanisms tend to double their rates of degradation for every 10C higher the temp goes.

So if you are at 50C for one clockspeed then your degradation rate is "X" if all things are constant (voltage, clockspeed, usage, etc).

If you keep all else the same but raise the operating temps to 60C (a 10C increase) then the rate of degradation had now doubled, roughly, to "2X" and your chip will now, statistically speaking, die in 1/2 the time it was going to die had you kept the operating temps at 50C.

But don't worry, your chip was designed to be capable of operating at TJmax for a decade, that is what determines the specific value of TJMax (lifetime, warranty costs, etc), so if you have the temps well below TJmax then it really isn't an issue whether they are at 50C or 60C or 70C, you've got decades of life to come from the chip.

 

[Hogan773]

Yes if the temps are higher then the degradation will occur at a faster rate.

Thermally activated degradation mechanisms tend to double their rates of degradation for every 10C higher the temp goes.

So if you are at 50C for one clockspeed then your degradation rate is "X" if all things are constant (voltage, clockspeed, usage, etc).

If you keep all else the same but raise the operating temps to 60C (a 10C increase) then the rate of degradation had now doubled, roughly, to "2X" and your chip will now, statistically speaking, die in 1/2 the time it was going to die had you kept the operating temps at 50C.

But don't worry, your chip was designed to be capable of operating at TJmax for a decade, that is what determines the specific value of TJMax (lifetime, warranty costs, etc), so if you have the temps well below TJmax then it really isn't an issue whether they are at 50C or 60C or 70C, you've got decades of life to come from the chip.

Ok perfect.....but technically then going back to our scenario of a fixed piece of work, at fixed voltage, then running that work at 44x and hurrying back to idle more quickly than running it at 35x would still result in a bit more decay if one assumes that the max temp will be higher at 44x than 35x while said work is running.

I realize that this is all just intellectual curiousity by the way, because in practice its probably a few degrees, and as you point out, the chips are designed to last a long time vis a vis temperatures. If temps were a very important determinant of life, then I'd think everyone would be wise to spend $20-30 on a decent CPU cooler like my Hyper 212+ so they could lower their everyday temps vs stock cooler. But it seems that voltage is the real potential killer if we're ranking things......

 

[Idontcare]

Ok perfect.....but technically then going back to our scenario of a fixed piece of work, at fixed voltage, then running that work at 44x and hurrying back to idle more quickly than running it at 35x would still result in a bit more decay if one assumes that the max temp will be higher at 44x than 35x while said work is running.

Correct.

I realize that this is all just intellectual curiousity by the way, because in practice its probably a few degrees, and as you point out, the chips are designed to last a long time vis a vis temperatures. If temps were a very important determinant of life, then I'd think everyone would be wise to spend $20-30 on a decent CPU cooler like my Hyper 212+ so they could lower their everyday temps vs stock cooler. But it seems that voltage is the real potential killer if we're ranking things......

Actually on that topic your HSF is an investment that has NPV/NFV considerations because, if you do it right, your HSF selection will allow you to operate your CPU at a lower Vcc for any given clockspeed because the temps will be lower at that given clockspeed.

Lower Vcc means lower power-consumption which means your performance/watt improves and you spend less money on powering your rig.

Intel actually came out and publicly stated that the primary motivation for their changing TJmax on the 65nm steppings of some quadcores was so that they could save ~$2 by using a cheaper stock HSF. What they were doing, in essence, was transferring the cost of ownership from an up-front expenditure to a TCO line item that the customer paid for by having elevated power bills for the lifetime of the CPU.

For modern IC's the power consumption scales as the cube of the Vcc under load.

Depending on how much time your rig spends fully loaded it might not take much time at all before your $30 HSF purchase actually starts saving you money if you elected to leverage it to reduce your Vcc instead of increasing clockspeeds.

 

[bankster55]

as long as your BCLK never shows lower than 100 at idle
then your readings are correct
If BCLK also shows lower then theres another issue thats involved

 

[Hogan773]

My CPU-Z BCLK always shows 99.8, even though I know its set at 100.0 thru the mobo. It never changes though and never really concerned me since I just look at the multiplier anyway. I just figured it was a slight error on CPU-Z's part.