Just wondering. there's so much range it what people think is a maximum cpu operating temp, the temps those of us on SPCR accept would certainly be considered nuts by many overclocking/supercooling sites so what do you all feel are ok temps. I know AMD and Intel chips are rated to run safetly at some pretty high peaks but at what temps do you all start getting nervous?

With an XP1700+ cooled by a Volcano7 and a 6v Vantec Stealth I currently max out around 60c which I'm sort of ok with, how bout the rest of you? I know better airflow would help as my computer is built into my desk and the temps drop but 5-9c when I leave the front of the drawer open however I don't really want to add fans if I can help it so am telling myself that I'm cool with 60c.

Cosine

I feel nervous when I am over 80°C internal diode temp on my Palomino.

50 deg C used to be my old scare temp. now that my barton runs 52 on fulll load it's raised to to 55.

I don't like to guess on temps, so I did try to get as much facts as I could.
Currently the facts about cooling is in Swedish only, so I guess posting a link here won't do much good.
Instead I guess it is better if I do a quick translation of parts about of the chapter about CPU temp.
All temps in Celcius.

The basic idea is that any chip will fail sooner or later. The higher temp it has the sooner it will fail. All numbers are based on manufacturers specs and calculated on average time the chip will live at a certain temp (assuming the chip will be constantly on the same temp).
Most chips very rarely reach there top temps, so you should have a top temp that keeps the system stable and has a little extra margin. On op of that you should check your normal CPU temp and decide if the time to live for the CPU is enought for you.

First calculate how many hours per year your computer is on.
Eg. "home computer" most days on 3 hours. ~1000 hours/year
Eg. "server" always on 8760 hours/year.

Temp (Intel/AMD) Time to live (home/server) in years
65/88 2.5/0.28
55/78 5.0/0.57
50/73 6.5/0.74
45/68 8.0/0.91
40/63 11.5/1.31
35/58 15.0/1.71
30/53 20.0/2.28
25/48 30.0/3.42

According to me your top load should never raise above 65/88 (Intel/AMD) since that will drastcly shorten the lifespan of the chip.

Example of how to read the chart:
I would be satistied with an AMD for home use with normal workload temps of 78 C (I guess this will shock many readers), but that will give me an average lifespan of 5 years on the CPU. After less than 5 years I will probably replace the computer anyway.
On the other hand a server at home might live for 5 years, so even 48C is probably unacceptable.

I think your underestimating the lifespan a bit. But going above 88 definately will cut down your lifespan more than a bit

I start to worry at about 70 for a stock chip, lower if its overclocked since it'll crash.

My Duron 1.3GHz maxes out at 55C/111F.

I'm figuring this was caused by my crappy thermal pad application, but I'm not gonna bother fixing it.

Unless you're running a distributed computing app or are an insane hard-core gamer your temps are hardly ever going to be much over "idle" so those numbers are absolutely worst case.

Over the last 20+ months that I've been running P4 systems I've been completely happy with 50-55°C load temps and none of the 40+ systems I've built has had any problems. About 8 of them are run 24/7 but the average temp probably runs in the 40°C range except for very rare occaisions of intense usage. Going by the numbers for 40-45°C on that chart I shoud be starting to see CPU failures by now and it's not happening (knock wood).

It's very difficult to say, imho.

It depends so much on the chip, diode and mobo.

On Athlon XP (thoroughbred/barton) I get nervous at 55 degrees under load, although most systems I've used can handle up to 65 degrees centigrade with no problems.

The trouble comes from the fact that not all thermal diodes are very accurate.

Some abit sensors on NF7 series claim to measure the temp from the thermal diode inside the CPU, but actually just produce some average using some calculation mechanism.

So, even so called "on-die-diode" temps can be several degrees off compared to the real temps.

I have no experience with P4 series, so no comment on those.

Thanks for the formula Silver, so I guess I can really figure using my pc 3 hrs a night even now at about 44c it should last me at least 3 to 4 yrs by then I will have a new one.

I am still worried how I am running 25 case and now my cpu is at about 42 to 44 at idle but I keep adding case lining to every crack n hole too.
Course gotta add a Slk to it with artic silver down the road. :}

I don't know where those numbers come from, but they're not even close.

The standard formula for such temperature degradation of component life is:

Lifespan=MTBF/[((273+High Temp)/(273+Base Temp))^M]

Where in:

Lifespan= estimated average life of component
MTBF= Mean Time Before Failure. Calculated by failure testing at a base temperature. For a P4 Intel gives an MTBF of 30000 hours (3.42 years, if you notice this does correspond with the table's 25°C rating)
High Temp= The max temperature at which the component will be run. For this equation the assumption is made that it is run at this temp its entire life. (obviously not true, but is a worst case scenario)
Base Temp= The temp used in the determination of the MTBF. Intel reports this as 25°C
M= Is determined by the manufacturer's temp tests and varies according to the specific component. Intel's M number are generally around 10.

Using that formula running your machine at 65°C will give your CPU an average lifespan of 8513 hours. (Or 0.97 years) Thats about 4 times what that table reports.

But even that calculation is nowhere close to real-world use. It ignores the fact that thermal cycling has a bigger impact on component life the max operating temp does.

Basically it boils down to this:

You're better off to run your system at a constant temp, with that constant temp being the lowest you can get, but realise that you will have replaced that CPU due to its functional lifespan having passed long before its thermal lifespan is reached.

In my note I did wrie that you basicly just go with the normal temp on the chip, and just make sure you have a reasonably low top temp.

I guess the values I had in my table originates from something close to the formula Rusty075 wrote, except that mine are slightly simplified and recalculated to years instead of hours. It seems that they do have some magin put into my formula, and I guess that is because of "normal" thermal fluctuations and that most people do turn their computers on and of.
So while Rusty's formula is more scientifically correct and perhaps more true for 24*7 systems I would still go for the simple table for most non-server computers since it is probably a quite good estimate of real life usage.

Frosty, from your sig it looks like you use an AMD. In that case I would not worry a bit about the 42 to 44 C at idle.

By the way, thanks Rusty for the formula. I'll add that to my swedish document as well.

I begin to worry when the CPU reaches around 80-85 C. That is, if it is during a test. If it is during the operation (like, every day) I would say around 65-70.

There is absolutely no practical evidence to support any of those figures. Does anyone reading this know ANYONE who has had a CPU chip die on them -- before it was so totally obsolete that nobody cared any more?

To answer your original question: as long as my machine is stable (at all ambient temps) I dont worry. My hobby/hacking basically expects an 18 month life from any CPU and so far I have only had 1 CPU failure - a Cyrix P166+ and that probably had more to do with the basic design than overheating.

Yup - I used to get neurotic about the numbers in MBM, but have realised (at least for Athlons) they are indicative only. A recent mobo upgrade raised my CPU temp more than 12C: well, actually raised the MBM reported number. So in practice on the old mobo I had to worry when MBM climbed above 40, on this new mobo things seem OK with MBM running well into the 50s.

Nope, not me. Ever.

And I don't know anyone else that ever has either.

RAM, PSU's, mobos, yes, but CPU's, no.

I haven't had a CPU die on me yet. Which is pretty suprising, the chip I'm using right now has small chips in the corner. I don't know how they got there.... And another CPU is still working fine after a water accident killed its mobo. And my palomino is nearing two years and according to my mobo it hasn't dropped below 50C in months....

Water accident? How?

Well figured are from manufacturers specs. Chip manufacturers do make a lot more tests and calculations than normal people will ever do.
Try seeing things from the opposite side (opposite to Wumpus question) would be: How many people has at normal load been running their AMD processor at no less than 78 C for more than 5 years without burning it out?
Anyone in this situation please reply to this note.

I guess that is hardly anyone at this forum that runs there chips on that high temps. We have already discussed constant server temps and it seems that the numbers seem to have a factor 4 security buffer in them.
If you have a computer that is more than 5 years old and it does die on you, do you even bother trying to figure out if it is the motherboard or the CPU that died?
Did you constantly monitor temps on anything? (most people never do).

I talked some to a computershop (that puts their own computers together) many years ago, and they did get a few customers with broken CPU's every year. But on the other hand they could have been broken by lost of reasons. E.g. physical abuse, massive overheating, electical shorted out and they didn't bother about checking anyway. Usually there was no guarantee left, so the reason the chip was broken was rather irrelewant.
I know this is not even close to a scientific evidence.

50 is the max limit for me for not getting the shakes, luckily my barton never goes over 45