OC & Voltage help for my setup.

[TylerS]

I am trying to squeeze a bit more power out of the below system in order to stave off a CPU/mobo/RAM upgrade for another year or so.

System: C2D e7500 @ 3.93 / GA-P35-DS4 (v.2.0) / 6GB PC 800 Patriot @ 895 / Asus 7870 (stock) / Corsair HX620 psu / Thermaltake Water 2.0 Performer (cooler)

BIOS Settings:

• 
  3.93 gHZ (358X11)
  RAM @ 895; 5,5,5,15 (Motherboard doesn't allow 1:1 as far as I can tell so the System Memory Multiplier is set at 2.50)
  DDR2 Overvolt: +0.40 (2.2v)
  FSB: +0.15v
  gMCH: +0.125v
  Loadline Enabled
  CPU vCore: 1.36875v
  LinX: 20 passes successfully (set to Max 75C – not hit)
  Prime95: 4(ish) hours w/o errors

Temps seem okay; under stress test they run mid to high 60s on CPU and System temp. Not sure how hot my ram in running but I have been running 2.2v for a while now. The other settings and the cooler are new.

I have a few questions for the community about this setup:
1) How are my voltage settings? Safe, too high, can they be pushed a bit more?

2) I cannot seem to trim the timings on the RAM (I’d like to see 4,4,4,12), but should I bother? If there is noticeable performance to be gained, where would I start to try to get a bit more out?

3) The is a board specific question: the ability to go higher on the CPU overclock seems to be limited by my RAM, as it is forced to higher speed by the “System Memory Multiplier” in the Gigabyte bios. Am I missing something, is there a way to set to a 1:1 multiplier?

Thanks

 

[Idontcare]

I am trying to squeeze a bit more power out of the below system in order to stave off a CPU/mobo/RAM upgrade for another year or so.

CPU vCore: 1.36875v

1) How are my voltage settings? Safe, too high, can they be pushed a bit more?

Since you are only looking for this CPU to last you another year or so, you can burn through a good deal of what will basically become unused lifespan (when you toss it) by juicing up that Vcore (and GHz) right now.

1.37V is fine, it'll last you 10yrs at that voltage. The max spec voltage for those 65nm Conroe's is either 1.5V or 1.55V (I don't remember exactly, I just remember Intel revised it upwards a couple years ago and it is now >1.5V).

So long as your temps are in check, just to avoid thermal throttling at TJmax, you may as well crank up that Vcore and clockspeed for an extra bump in spead for your remaining year of using the chip.

Sure it will degrade the chip faster than running it at 1.37V but why care about that provided it still lasts a year?

 

[Saylick]

Since you are only looking for this CPU to last you another year or so, you can burn through a good deal of what will basically become unused lifespan (when you toss it) by juicing up that Vcore (and GHz) right now.

1.37V is fine, it'll last you 10yrs at that voltage. The max spec voltage for those 65nm Conroe's is either 1.5V or 1.55V (I don't remember exactly, I just remember Intel revised it upwards a couple years ago and it is now >1.5V).

So long as your temps are in check, just to avoid thermal throttling at TJmax, you may as well crank up that Vcore and clockspeed for an extra bump in spead for your remaining year of using the chip.

Sure it will degrade the chip faster than running it at 1.37V but why care about that provided it still lasts a year?

IDC has a good point. If you're gonna be keeping the chip for another year at most, give the chip the beans and get some extra performance out of it while its practical lifespan is coming to an end.

NOTE: Your chip is a Wolfdale, NOT a Conroe. It's based on Intel's 45 nm lithography. A quick look up on Intel's data base indicates that the voltage spec is from 0.85V to 1.3625V. You are currently running it slightly beyond Intel's specifications. This doesn't mean your chip is going to explode but it means your chip is going to degrade quicker than Intel's target lifespan, which I am guessing is on the order of a few years (5 perhaps? Someone will have to confirm this value). Since you only have a year left with this chip, I doubt the chip will fail before you decide to retire it.

 

[Idontcare]

NOTE: Your chip is a Wolfdale, NOT a Conroe. It's based on Intel's 45 nm lithography. A quick look up on Intel's data base indicates that the voltage spec is from 0.85V to 1.3625V. You are currently running it slightly beyond Intel's specifications. This doesn't mean your chip is going to explode but it means your chip is going to degrade quicker than Intel's target lifespan, which I am guessing is on the order of a few years (5 perhaps? Someone will have to confirm this value). Since you only have a year left with this chip, I doubt the chip will fail before you decide to retire it.

Thanks Saylick for catching my mistake there, yeah that definitely makes a difference in terms of what is safe and what is not safe (65nm vs 45nm).

The upper voltage value will be based on a target expected lifetime for 90% of the chips if operated at TJmax. These are industry standard practices for all semiconductor manufacturers. Some will cut corners in QRA and loosen their internal specs when they run into trouble of the financial kind, but a fiscally robust company like Intel won't be doing that.

So basically at max Vspec and TJmax, 90% of the chips are expected to last 10yrs. They determine this by way of accelerated lifetime testing, over-volting and over-heating to induce failures in weeks and months in a way that conforms to an established model that then analytically speaks to lifetime reliability as you reduce voltage and temperature.

It is from this testing that they determine what TJmax and Vmax will be set to. It is both process node and chip design dependent. Two different chips on the same node will have different TJmax and Vmax as needed to hit the same internal target of 10yrs/90%.

Overvolting starts to chew through that 10yr lifetime, and of course you also run the gamble of having one of those chips that was going to die before 10yrs anyways (the 10%) even if you ran it at stock.

The rate of voltage induced degradation depends on the degradation mechanism, some are linear with voltage, others are exponential. The exponentially dependent mechanisms are what kills your chip at high volts, the linearly dependent mechanisms are what kills your chip at stock volts.

As enthusiasts we like our rules of thumb, they are easy to remember and are close enough to the truth to be good enough for what we do. Rules of thumb like "for every 10°C that you lower the peak operating temperature you will double the lifespan of the chip". (comes from the Arrhenius equation that describes the kinetics of thermally activated degradation mechanisms)

Exponentials are tough to deal with conceptually but can be expanded into Taylor Series and we then develop rules of thumb based on the behavior of just the first few dominant terms.

The exponential function (in blue), and the sum of the first n+1 terms of its Taylor series at 0 (in red).

This is where we get our rules of thumb where we approximate the actual device physics (which are exponentially dependent) by saying things like "it scales as the square of the voltage" or "it inversely scales as the cube of the temperature".

So with that in mind, the rate of degradation as induced by voltage scales as the fourth power of the voltage.

But that is assuming the temperature is held constant. Raising voltage usually results in the temperature increasing as well, so there is an additional effect in terms of reducing the IC's lifetime which comes from both the voltage and the temperature combined.

 

[TylerS]

These are great answers and give me some piece of mind for the CPU oc. The motherboard seems to be limiting my ability to push further because I can't set the RAM to run any slower. The modifier can't (as far as I can tell) be set lower so any further increase in FSB will up the RAM as well.

Does anyone know this board well enough to confirm?

Any tips on squeezing a bit more stable speed from the RAM?

Thanks

 

[Kenmitch]

Any tips on squeezing a bit more stable speed from the RAM?

If you've already tried juicing it or loosening up the timings about all you could do is replace it with faster ram....Most likely not worth the cost these days tho.