Is there a speed vs. temp rule of thumb?

[dscline]

I did some searching, but really couldn't find anything. I know the answer will vary greatly depending on the chip and other factors, but does anyone know the rough correlation between cpu temp and max attainable speed? We all know lower temps can give you greater stability when overclocked, and can even help you reach higher speeds. But by how much?

I have a Celeron-T 1.2, which is stable up to 1656Mhz. This is at 1.7v, using pretty decent air cooling which keeps it down to 35°C at full load. I'm considering trying chilled water cooling. Lets say I can get that 35°C full load chip temp down to 10°C. Any guesstimates as to how much higher I could take the chip with a 25° cooler core? 1%? 5%? 10%?

 

[vonkaar]

That's too broad of a question to answer... precisely due to the problems you specified. CPUs differ far too much to create a standardized rule for temp vs overclock. I may have a Barton 2500 that can hit 2.6ghz at 35° but someone else may have one that hits 35° at stock speeds. Also, since the temperature is governed by the cooling power of your system, what's 'standard' for me may not be standard for you.

I would do some archive searching and look for the highest overclock that your chip has done. If you can find your exact stepping on a decent overclock, you can get a good idea as to the potential of your chip... given enough cooling.

 

[Foxie3a]

School in 5 minutes, so I didn't read anything but the subject, but wanted to help quickly.

Once you reach the processors "ceiling" (like in aviation) it can not go any farther, so decreasing the temps, or increasing the voltage will have no effect.

Dont think about it too much, its easier when you just quickly think about it, thinking too hard you won't understand.

 

[dscline]

Once you reach the processors "ceiling" (like in aviation) it can not go any farther, so decreasing the temps, or increasing the voltage will have no effect.

I don't agree with that at all. Lets say I I reached a "ceiling" of 1300Mhz with the stock HS/Fan and no thermal grease, and it was running at 50°C. Better cooling would likely get it higher. That's precisely why sub-zero systems can run so fast. And why cooling is so important to overclocking. If that weren't the case, there wouldn't be much of a cooler market... why would we go out & spend extra money on heatsinks if we didn't get better results than with the free intel HS/fan?

There IS a correlation between speed and temperature. I realize it's a hard one to quantify, because all chips are slightly different, and there are a lot of variables. I'm just trying to determine if switching to chilled water would be worth the effort & money. If 25°C only gets me 1%, then I'd say it's not worth the effort. But if it's 5-10%... maybe so... at least the majority of the expense could be reused over & over as I upgrade in the future.

 

[johan851]

Every processor has a ceiling...but most processors won't be realizing that ceiling at the 50C you specified...of course it's going to run higher at a lower temp.

It's just that even with phase change or a chiller of some kind, there comes a point where the CPU just won't go any higher, no matter what you do.

I doubt dscline's processor has reached its ceiling at this point.

 

[vonkaar]

He was referring to an actual 'ceiling' regardless of cooling. The people with top-notch cooling systems (hi) understand that there is a limit that any given processor can reach, regardless of the cooling provided. If I reach that ceiling with my system, yet feel that it *should* go higher, my main option is to find a better chip. Some processors simply overclock better than others. Overclocking is *very* often a game of luck. The 'ceiling' is a very real one... a trash cpu that either runs hot or doesn't take overclocking very well will have a low ceiling for max overclock, regardless of the cooling.

 

[dscline]

Oh, yes, I realize there will be a curve... For example, maybe going from 50° to 35° raises my ceiling 200Mhz, and another 15° lower to 20° only gets another 100Mhz. I'm not suggesting there's an infinite speed increase available with lower and lower temps (though I'd expect some interesting things could happen at absolute zero).

But sub-zero systems are becoming more popular. Surely there's SOME kind of expectation of what they will give you. I probably don't want to dive THAT low, as I don't want the hassles that come with it. But I am considering running water at JUST above freezing.

 

[vonkaar]

Like I said... it's too broad of a scope to narrow down into a specific 'catch-all' rule. I guess you could figure up an average number based on actual results people are getting with various CPUs, cooling and steppings... but that's an awful lot of work. If you stuck to a (for example) Barton 2500 and went around looking at 'champion' steppings like the AQUCA 0307, you might find enough examples to create a general 'rule of thumb' for speed vs temp. It simply becomes too broad to incorporate all chips and all cooling methods into a single rule.

 

[Foxie3a]

I was saying...

Lets say you get the absolute best cooling in the world, you do everything right on your Athlon..and you get like..2.7ghz

And your at 0*. a high Vcore, and you can not get anymore power.

That would the its ceiling. a CPU's ceiling can be stock speed, or 1ghz higher...all depends on the chip.

 

[dscline]

Yes, I agree there will be a point where more voltage and less temperature will not help any more. But I don't have all the cooling in the world, and I'm not at 0°. I expect that I'm still at a point where some extra coolling could gain a little extra speed, and also some safety margin for running a higher vcore, which could also help. I'm just trying to quantify what a resonable expectation would be in going from 35° to 10°C.

 

[Foxie3a]

how would you get to 10C?

and I do not think that u will get much more from that change.

 

[dscline]

Chilling water in a chest freezer.

 

[Foxie3a]

now come on, is that worth it?

what do you WANT to gain? 100mhz? for all that work?

 

[eg1441]

Oh, I see what he's asking.

The rule is roughly that max o/c is inversely proportional to degrees above absolute zero (-273c). Probably a pretty dodgy rule but you asked what the difference between 35c and 10c is so....


maxoc10/maxoc35 =
(273 + 35)/(273 + 10) =
1.088


So the answer to your question is you *may* gain about 8.8% higher clock.

Sounds like a dodgy rule to me... here's the link: http://www.icecoldcomputing.com/index.php

 

[dscline]

Thank you eg1441! That is exactly the type of information I was looking for. That at least tells me what a reasonable expectation is. This isn't something I wanted to pursue if anything more than a couple Mhz was unlikely... although I know there are no guarantees, as long as there is a decent chance of a noticeable increase, then it may be worthwhile. I was hoping to be able to get 1.75-1.8Ghz out of this chip, so this may be something worth checking out.

Foxie3a: All that work? have you seen watercooling setups? This would be no more work than that. This is an overclocking site... that's what we do: lots of work to squeeze as much out of the chips as we can. May not always be logical, but it has a certain amount of satisfaction to it. Sure, I could go spent $400 on a whole new PC from Dell that would be a lot faster than mine. But what fun would that be? Or, I can spend $250, put a little bit of effort in to it, and squeeze a little more out of what I have. But the beauty is I won't have to spend that cost again... I'll have the equipment to push all my future hardware a little farther, and at most, only have to buy new mounting hardware, or MAYBE a new waterblock. And as a bonus, I'll have nice cold water to chill my video card with as well, allowing it to go farther.

If 9% is a reasonable expectation, then it's something I can do without feeling like I'm totally wasting my time. Maybe I'll only get 1%, but that's luck of the draw. To me, there's a difference between putting effort into something that has little or no chance of success, and something that does have a chance of success but no guarantees. THAT is what I was trying to discern.

 

[vonkaar]

Sure, I could go spent $400 on a whole new PC from Dell that would be a lot faster than mine. But what fun would that be?

Beautiful attitude! That's the spirit!
clapping ^_^

 

[Foxie3a]

First of all, no dell is faster than a custom.. ;-)

And yes, I know that this is an overclocking site..I kinda figured that out a while ago. But in my opinion you will not get a higher OC from a few degrees. And it wouldn't be worth it. But its not me, its you..

tell us what results you get.

 

[dscline]

Just as a quick update, I did a quick test to see if lower temps really would give me higher clocks. It's cold out this morning, so I opened the window above the PC and opened the case. This gave me load cpu temps of about 73° F, as opposed to a more typical 95°. I was able to get up to 1704Mhz. The next notch up (1716) would crash instantly as soon as I changed the bus frequency. At 1704, I only ran Prime95 briefly (a couple of tests), probably not stable there, just really wanted to get confirmation that there'd be measurable increase with a drop in temperature. Since this brief test resulted in temps that probably aren't even half way between my current air temps and estimated chilled water temps, I'd say it's good enough for me to pursue. My goal is to get to 1.75Ghz out of this chip, any higher will be icing.

This'll probably set me back $250-300, which of course is kinda silly for that much of an increase. But it looks like heat is going to be an issue for future chips, and everything I do will be re-usable for future upgrades. May as well get a little bit of benefit now, and enjoy a silent PC in the mean time. It'll probably take me a while, as I've got other things going on too, but I'll report back when it's done.

 

[gusgizmo]

hmm. . . i dont feel that anything short of lowering the cpu's temperature another 30 degrees or so will increase its overclock at a given vcore. but i do feel that better heat removal will allow you to increase your vcore and allow for higher overclocks with out becoming unstable due to heat issues. i would wager that every degree lower celsius will give you a gain of 3 MHz. also, the maximum switching speed of a transitor is a linear function all the way to jsut above 0K, so there really is no ceiling. you can always do something to make it go faster. whether it is worth it is another question.

 

[Nirbush]

When i had normal air cooling i could only get my athlon 2000xp @ 2.200 that was at 52c, but when i put water cooler on it i can get it to 2.400 but i have noticed if i turn off my pump for a sec and temp raises to 50c it crashes instantly.