Degradation and Voltage

[Smartazz]

I'm sure many of you believe in degradation and I've seen it happen before. Let's assume we had two different processors, say 2500Ks in two different motherboards. Motherboard A has less vdroop so that under load it hits 1.3v. Motherboard B has more vdroop so that it hits 1.2v under load. Is the processor experiencing less degradation in the motherboard that suffers from more vdroop?

 

[blastingcap]

Do we really need yet another thread about this?

http://forums.anandtech.com/showthread.php?t=2243736

http://forums.anandtech.com/showthread.php?t=2275762

 

[Smartazz]

This question is specifically about the effect of vdroop on degradation? Is it significant enough to make a difference. CPUz reports -.1v whatever I set in the bios.

 

[Idontcare]

Vdroop is good for your processor in terms of increasing longevity. The lower the voltage under load the better.

The reason people avoid Vdroop is because it can make your system unstable and crash or produce corrupted/wrong results because the voltage can droop to a value that is too low for the CPU for the given operating temperature and clockspeed.

An unstable CPU that lasts 10yrs is worth less than a stable CPU that only lasts 1 year.

But yes, provided that both the Vdrooped case and the no-droop case are providing sufficient voltage for stability to be maintained then the Vdroop situation is better for the CPU.

 

[Smartazz]

Vdroop is good for your processor in terms of increasing longevity. The lower the voltage under load the better.

The reason people avoid Vdroop is because it can make your system unstable and crash or produce corrupted/wrong results because the voltage can droop to a value that is too low for the CPU for the given operating temperature and clockspeed.

An unstable CPU that lasts 10yrs is worth less than a stable CPU that only lasts 1 year.

But yes, provided that both the Vdrooped case and the no-droop case are providing sufficient voltage for stability to be maintained then the Vdroop situation is better for the CPU.

Thanks IDC! Just the answer I was looking for. Is CPUz more or less accurate though?

 

[Yuriman]

My experience has been that when I have LLC enabled (to combat vdroop), I tend to crash when changing load or frequencies rather than when at full load. This may just be a result of a poor implementation of LLC though. Because of that I need to run a slightly higher loaded voltage with LLC enabled to get the same clocks.

I suppose what the question becomes then is, which is better for CPU longevity of these two scenarios? :

LLC enabled:
Part load - 1.285v
Full load - 1.285v

LLC disabled:
Part load - 1.325v
Full load - 1.275v

^
Does most of the degradation occur under full load when temperatures are highest, or is running a higher voltage through at partial loads more damaging?

 

[Smartazz]

My experience has been that when I have LLC enabled (to combat vdroop), I tend to crash when changing load or frequencies rather than when at full load. This may just be a result of a poor implementation of LLC though. Because of that I need to run a slightly higher loaded voltage with LLC enabled to get the same clocks.

I suppose what the question becomes then is, which is better for CPU longevity of these two scenarios? :

LLC enabled:
Part load - 1.285v
Full load - 1.285v

LLC disabled:
Part load - 1.325v
Full load - 1.275v

^
Does most of the degradation occur under full load when temperatures are highest, or is running a higher voltage through at partial loads more damaging?

Are you running it 24/7 and what are your temps?

 

[Ajay]

Wish someone would invent a good *reasonably priced* LC chiller system so I could run my cpu @ 0 deg F (under load) and worry a bit less about electromigration (in reality, it takes much lower temps than that to really net zero on a high overclock, IIRC). Someday...

 

[Idontcare]

Thanks IDC! Just the answer I was looking for. Is CPUz more or less accurate though?

CPUz is not very accurate for one specific reason - the program quantizes the voltage it reports in 0.008V increments.

So even if it was obtaining accurate voltage readings it still wouldn't be reporting the voltage correctly 7/8's of the time.

In Excel the formula is as follows:

=ROUNDDOWN(([COLOR=darkgreen]A1[/COLOR]/0.001)/8,0)*8/1000

^ Where "A1" is the cell containing the actual Vcc per the BIOS (or multimeter if you have one setup).

Not only does CPUz quantize the values, it rounds them down when it reports them.

CPUz is good enough for most things though, and you can't beat the price

I suppose what the question becomes then is, which is better for CPU longevity of these two scenarios? :

LLC enabled:
Part load - 1.285v
Full load - 1.285v

LLC disabled:
Part load - 1.325v
Full load - 1.275v

^
Does most of the degradation occur under full load when temperatures are highest, or is running a higher voltage through at partial loads more damaging?

For those specific numbers I would have to say the LLC enabled is better for reliability than the disabled case.

The difference at full load is minimal, but the difference for part load is substantial and not in a good way.

That said, in both cases the end result is going to be only of academic interest. You are looking at basically deciding between 10yrs lifetime and 11 yrs lifetime, in either case you are probably going to discard the CPU in less than 6 or 7 years tops.

Wish someone would invent a good LC chiller system so I could run my cpu @ 0 deg F (under load) and worry a bit less about electromigration (in reality, it takes much lower temps than that to really net zero on a high overclock, IIRC). Someday...

Gotta worry about that condensation when you start pushing the temperatures below ambient. Relative humidity and all that. Don't want dew collecting on your mobo

 

[Ajay]

Gotta worry about that condensation when you start pushing the temperatures below ambient. Relative humidity and all that. Don't want dew collecting on your mobo

There's plenty of stuff out there to solve that problem. Silicon RTV, plastic foams, neoprene and a drip line wouldn't be a problem (if needed).

 

[Idontcare]

My $1k vapophase LS system sits in my basement unused for a reason, just sayen.

 

[Ajay]

Well, vaporsphase is allot harder than sub ambient LC, IMO. But that's just from what I've read about each. I've only done custom and non-custom H2O.

 

[Smartazz]

Is phase change cooling too loud for you IDC?

 

[Idontcare]

Is phase change cooling too loud for you IDC?

It is, but that wasn't why I ditched it. I ditched it because of the power-consumption.

Basically for every watt of heat you want the vapoLS to remove it takes another watt of electricity to get it done.

My QX6700 at 4GHz and 1.55V was using ~270W of power. The vaporphase LS was also using ~270W of power. (it has its own power cord, so tracking power for just the vaporphase unit was easy)

That is a huge price premium to pay on an annual basis for what amounted to around an extra 10% OC versus what I could get with top-end air cooling that only consumed 4-5W for the fans.

Mind you I do not regret it at all. Having a quadcore in 2006 was awesome, the pinnacle of nerdvana. Having 4GHz CPU in 2006 was the pinnacle of nerdvana. Having a 4GHz quadcore kentsfield in 2006 was ridiculously fun.

And I ran it for around 2yrs without concern for the cost or the noise.

But now I can get so much more performance out of an air-cooled 5GHz 2600k without all the cooling expense. Not interested in going back to phase cooling anymore. Doesn't hold the excitement for me like it did in 2006 when hardware was changing so quickly.

 

[Smartazz]

It is, but that wasn't why I ditched it. I ditched it because of the power-consumption.

Basically for every watt of heat you want the vapoLS to remove it takes another watt of electricity to get it done.

My QX6700 at 4GHz and 1.55V was using ~270W of power. The vaporphase LS was also using ~270W of power. (it has its own power cord, so tracking power for just the vaporphase unit was easy)

That is a huge price premium to pay on an annual basis for what amounted to around an extra 10% OC versus what I could get with top-end air cooling that only consumed 4-5W for the fans.

Mind you I do not regret it at all. Having a quadcore in 2006 was awesome, the pinnacle of nerdvana. Having 4GHz CPU in 2006 was the pinnacle of nerdvana. Having a 4GHz quadcore kentsfield in 2006 was ridiculously fun.

And I ran it for around 2yrs without concern for the cost or the noise.

But now I can get so much more performance out of an air-cooled 5GHz 2600k without all the cooling expense. Not interested in going back to phase cooling anymore. Doesn't hold the excitement for me like it did in 2006 when hardware was changing so quickly.

You have a delidded Ivy Bridge right? Have you considered using the phase change cooler on it?

 

[Idontcare]

You have a delidded Ivy Bridge right? Have you considered using the phase change cooler on it?

Yeah I've considered it but realistically all that will do is enable me to push it another 10% at most. So I hit 5.5GHz instead of just 5GHz.

The last time I was actually in need of an extra 10% performance was in 2006-2007 when I was using MetaTrader 3 for developing autonomous foreign currency trading algorithms/programs. Time was money.

But then I migrated to MetaTrader 4, which I am still using some years later, and expanded the programming horizontally across cores rather than vertically in clockspeed. Replacing the vaporphase rig with 5 Q6600 rigs that did roughly three times the calculations per day.

For the stuff I do now I appreciate a fast machine so I can rapidly iterate through prototype designs before pushing them off on the farm to do production runs. But it is a convenience now, not a matter of money, so it is harder to justify the noise and expense of firing up the vaporphase cooler just for an extra 10% clockspeed.

I might do it just for the fun of checking out what I can do with vaporphase cooling on a bare-die mount with IB. That does intrigue me a bit, I must admit

 

[BonzaiDuck]

My experience has been that when I have LLC enabled (to combat vdroop), I tend to crash when changing load or frequencies rather than when at full load.

Just the passing observation: that symptom was noted in that old, frequently-cited Anandtech article about over-clocking the QX9650 (late 2007? or thereabouts . . . ). What trade-offs one has to make to insure the transitional stability, I either don't remember or can't be sure.

I can only say that I bumped up my LLC with reluctance and to the point where I was satisfied with both the peak (idle-turbo) voltage and the loaded voltage, and I left myself a remainder of maybe 20 mV of droop. So under that regime, I'd either adjust the voltage or the speed to avoid any other instability.