Is there a downside to undervolting my CPU?

[wgoldfarb]

I am trying to get my CPU as cool as possible, so I can try to silence my system as much as I can. To that end I decided to try to lower Vcore as low as I could go and remain stable.

I started at 1.27V (this was the actual voltage supplied by my MoBo under full load when Vcore was set to Auto, as read by Speedfan). I set Vcore to manual, C1E and EIST disabled, and have been trying lower and lower values, testing my system for stability each time. I am now at 1.1875V in the BIOS (actual Vcore as read by Asus Probe and Speedfan is 1.14V under full load) and everything is still stable -- I ran Orthos Super Prime for 14 hours with no problems. I can probably keep going down!

Is there any danger (other than a blue screen if the system becomes unstable) of UNDERvolting a CPU? I know increasing the voltage can shorten the lifespan of a CPU, but what about decreasing it? If the system is stable, is there any reason to avoid supplying low voltage to the CPU?

TIA!

 

[Brunnis]

No, you're actually doing your CPU a favor by lowering the voltage. It will probably last considerably longer when run at such low voltage. Not that it would go bust in any near future anyway, but undervolting is atleast good for the CPU and not the other way around.

BTW, I have done the same test on my E6600 and a Gigabyte GA-965P-DS3. With this combo, I managed to run the system stable at 1.0V on stock 2.4GHz. The same CPU needs 1.17V to run stable at 3.0GHz.

 

[wgoldfarb]

Brunnis,

I will keep going then. If you CPU is any indication, I still have some room to go lower.

Thanks! :beer:

 

[xenolith]

AFAIR, significantly undervolting only becomes a concern if your using an AMD K8 A64 CPU that has an on-die memory controller and you're using high vdimm. A large delta between vcore and vdimm can cause excessive stress on the CPU's memory controller and cause premature death.

Intel CPUs have no on-die memory controller, so it's not a concern for you. Even so, I've never understood the reason for undervolting and introducing the possibility of instability with no performance increase. Other than maybe "because I can." Yes, you will very likely extend CPU life expectancy a great deal, when maintaining proper cooling, but why? I don't know too many computer enthusiasts still actively using an old Intel 286.

 

[jazzboy]

Originally posted by: xenolith
Other than maybe "because I can." Yes, you will very likely extend CPU life expectancy a great deal, when maintaining proper cooling, but why? I don't know too many computer enthusiasts still actively using an old Intel 286.

It allows you to then use quieter or even passive cooling.

 

[xenolith]

Originally posted by: jazzboy

Originally posted by: xenolith
Other than maybe "because I can." Yes, you will very likely extend CPU life expectancy a great deal, when maintaining proper cooling, but why? I don't know too many computer enthusiasts still actively using an old Intel 286.

It allows you to then use quieter or even passive cooling.

Yes, but stock cooling fans are already so quite that I personally can't even hear them at all in a closed case. But I guess some people could get a defective fan, or can detect even the slightest fan noise and be annoyed by it... in that case get a thermalright Ultra-120 passive cooler, you still don't need to undervolt.

 

[CTho9305]

Just keep in mind that you're giving yourselves all the stability / silent data corruption risks that come with overclocking.

Originally posted by: xenolith
AFAIR, significantly undervolting only becomes a concern if your using an AMD K8 A64 CPU that has an on-die memory controller and you're using high vdimm. A large delta between vcore and vdimm can cause excessive stress on the CPU's memory controller and cause premature death.

I'm skeptical for a few reasons (not just that the various built-in under-clocking schemes such as Cool'n'Quiet or speedstep in CPUs lower the core voltage). I'll have to ask some people more knowledgeable than myself if it's possible that could cause damage / increase wear.

 

[wgoldfarb]

Ctho9305, what is "silent data corruption"?

 

[Zap]

Originally posted by: CTho9305

Originally posted by: xenolith
AMD K8 A64 CPU... A large delta between vcore and vdimm can cause excessive stress on the CPU's memory controller and cause premature death.

I'm skeptical for a few reasons (not just that the various built-in under-clocking schemes such as Cool'n'Quiet or speedstep in CPUs lower the core voltage).

I believe that there was a lot of anecdotal evidence at the defunct DFI Street forums. Because the DFI boards allowed such high possible vDIMM numbers, some people were experiencing CPU failures at questionable settings (as in while testing memory overclock and CPU was at or near stock). Someone postulated that it was the large DIFFERENCE (delta) between CPU and memory voltage which caused it, and challenged any brave souls who weren't afraid to (or could afford to) kill their CPU to test out the theory by overvolting vDIMM quite a bit and undervolting the CPU (even leaving all clocks stock). A number of dead CPUs ensued.

Of course I could be imagining the whole thing...

 

[CTho9305]

Basically, when your CPU makes a mistake, there is some chance you notice - for example, due to a program crashing, or visual artifacts in a game - but there is also a chance that you don't notice. There are different errors a CPU can make that have different results.

One error that would not cause any problems is a mistake in a branch prediction. The branch predictor doesn't affect correctness, so worst case there would be a tiny slowdown. Another error that wouldn't matter is an error that corrupts information that isn't reused. Every time the CPU adds two numbers, it tracks some "flags" that note if the result is positive, negative, or zero. If a flag is miscalculated but it isn't checked, nothing bad happens.

An error that would cause a crash might be a mistake in a TLB while the OS kernel is doing something, or a mistake reading / decoding an instruction (which could result in the CPU trying to do something illegal).

An error that could silently corrupt data would be a bit flipping in a floating point calculation. If, say, you're doing your taxes and your CPU is overclocked, it's possible that one of the bits of a floating point calculation doesn't get back into the register file before the clock ticks, so the wrong answer is stored. You might not notice so long as the error doesn't cause an obvious discrepancy... that's silent data corruption.

 

[ineedaname]

Honestly if you want your computer to be silent just buy any good cooler and don't put a fan on it. They'll run cooler than your stock heat sink fan and be ABSOLUTELY silent.

Some examples of decent coolers are Ultra 120, Scythe Infinity, Tuniq Tower 120. Even with no fan these coolers will run cooler than that POS intel gives you.

 

[wgoldfarb]

Ctho,

Thanks for the explanation! I might as well not push the undervolting. I will follow the advice from several people in this thread and just get a silent cooler. Point very well taken

So, I am now at 1.1625 in BIOS, which translates into 1.128 "Actual" Vcore. What is considered the minimum "normal" Vcore for a E6600? I have seen many places that mention the Vcore for a Core 2 Duo, but they always quote a wide range, which is not very helpful.

I went to Intel's website and found 2 different sets of specs for the E6600. My E6600 is the SL9S8, which indicates a Vcore between 0.85V and 1.3525V. This is a VERY wide range. Does it mean that if I stay above 0.85V I am safe, or is the 0.85 only under certain circumstances (like when C1E or EIST are enabled and active). The "other" E6600 indicates a Vcore range of 1.187V - 1.325V, a much narrower range (on both ends of the range!).

Does anyone know why there are 2 versions of the E6600, with such different voltage specs? What is so different about them? According to Intel both are desktop processors, and seem identical in all other respects.

 

[myocardia]

Originally posted by: wgoldfarb
My E6600 is the SL9S8, which indicates a Vcore between 0.85V and 1.3525V. This is a VERY wide range. Does it mean that if I stay above 0.85V I am safe, or is the 0.85 only under certain circumstances (like when C1E or EIST are enabled and active).

That's correct. The lower number is the vcore that processor needs when it's speed is reduced by Speed Step. The higher number is kind of in doubt; some people swear that it's the recommended max, most people think that it's the stock vcore.

 

[StopSign]

I've tried as low as 1.200v (in CPU-Z) but that was unstable because my CPU usage would occasionally shoot up to 100% for no reason. Right now I'm sitting at around 1.24v.

But 1.128...wow that's really impressive!

 

[F1shF4t]

My laptop cpu is undervolted to 0.95vcore, which reduced max temps by about 18C, and prolly increased battery life... as long is its stable its good, lower temps and chip will last longer, lasting longer is not really an issue cause i doubt u will keep it for 10+ years anyways

 

[Hyperlite]

I am doing this also...when i built a system for my little brother over christmas i sortof skimped on ram so i could get much past 2.6ghz. I undervolted to 1.15 (IIRC) @ 2.6ghz. 36C idle, 42-44C load on stock cooler.

 

[wgoldfarb]

I guess the downside of using a relatively new CPU is that there are still many unknowns. There does not even seem to be a consensus on what the "normal" operating Vcore is for the E66600. Is there anyone from Intel on these boards that can perhaps give a definite answer as to what a "safe" Vcore (on the low side of the range) should be at stock speeds? Now that I decided not to undervolt, I don't even know if my current voltage is within specs or not!

 

[CTho9305]

Originally posted by: Zap

Originally posted by: CTho9305

Originally posted by: xenolith
AMD K8 A64 CPU... A large delta between vcore and vdimm can cause excessive stress on the CPU's memory controller and cause premature death.

I'm skeptical for a few reasons (not just that the various built-in under-clocking schemes such as Cool'n'Quiet or speedstep in CPUs lower the core voltage).

I believe that there was a lot of anecdotal evidence at the defunct DFI Street forums. Because the DFI boards allowed such high possible vDIMM numbers, some people were experiencing CPU failures at questionable settings (as in while testing memory overclock and CPU was at or near stock). Someone postulated that it was the large DIFFERENCE (delta) between CPU and memory voltage which caused it, and challenged any brave souls who weren't afraid to (or could afford to) kill their CPU to test out the theory by overvolting vDIMM quite a bit and undervolting the CPU (even leaving all clocks stock). A number of dead CPUs ensued.

Of course I could be imagining the whole thing...

I asked a couple of people (people who would know) what they thought about this, and they thought it was unlikely that a low vcore would ever harm anything. Keep in mind that that's not a guarantee that it's safe, and it's not a recommendation to try it. See my sig.