CPU degrading due to heat/voltage

[dug777]

I know it's talked about all the time, and I recall reading an AT article on it a while back, but as someone who's run a Athlon Xp at 1.8vcore on air (stock 1.6 iirc (xp2000+run at@2.1Ghz)), without AC, in Perth, Western Australia, for the past 5 years I would have thought I'd have seen it in action by now?

My cpu ran using the stock hsf for the first three years I had it (and routinely loaded in the 90'C range, I had to reset the motherboard overtemp cut-out to its maximim setting to stop that ).

Since then I've used an SLK-800, which brought temps down considerably.

Surely it would be quite easy to test? Pick up a few processors, set them up to run in various voltage/temp circumstances, and sit back to watch...

 

[jmmtn4aj]

I've never heard of a CPU 'wear out' before, to be honest.. What article?

 

[Nathelion]

CPU degradation is more noticeable at smaller process nodes, IIRC.
And yes, I've definitely seen it happen. I had an Opty 165 that initially OCd to 2.7, but eventually degraded to a still-respectable 2.5.

 

[CTho9305]

Originally posted by: jmmtn4aj
I've never heard of a CPU 'wear out' before, to be honest.. What article?

I don't know if there are any good articles about it, but it's fact. Maybe pm or someone else in industry can write an explanation of impact ionization, nbti, tddb, etc but I can't right now. You can read my related post here but it doesn't answer the exact question I think you're asking. It does address electromigration myths...

 

[aigomorla]

Originally posted by: jmmtn4aj
I've never heard of a CPU 'wear out' before, to be honest.. What article?

lol...

how do you think chips die?

Electron migration issues is another one.

 

[Mwing]

i guess you will be itchy to upgrade your cpu years before it happens since you are ATer with 20000+ posts

 

[dug777]

Originally posted by: Mwing
i guess you will be itchy to upgrade your cpu years before it happens since you are ATer with 20000+ posts

The old timer troll strikes!

 

[nerp]

I'm thinking that the CPU is so small and fragile that any "wearing out" = failure, for the most part. It's not like a set of tires where the tread wears down but can still run for a while before the tires pop. I think either you're good or you're busted.

Degrading OCs could be other things... Thermal paste wearing out or breaking down, a capacitor losing effectiveness, a motherboard flaking out, a memory stick gone bad, etc.

 

[Idontcare]

Originally posted by: nerp
It's not like a set of tires where the tread wears down but can still run for a while before the tires pop. I think either you're good or you're busted.

Actually there are a number of defect mechanisms analogies you can make between tires and CPU's.

A perfectly functioning tire with no manufacturing defects can still go "pop" were its owner to subject the tire to a brutal and harsh environment (driving on lava, broken glass, rumble strips, etc) just as an otherwise defect free CPU can go "pop" were it to be subjected to extreme temperatures and voltages.

Likewise when operated as intended, the tire travels on a paved and well maintained road (no potholes, the car's alignement is confirmed, tires are rotated, etc) then it still has a finite lifetime as it slowly loses tread.

So too do CPU's suffer from simple degredation mechanisms that ultimately limit its lifespan no matter how caring and gentle the owner may be. Electromigration is a process in which metal atoms slowly but surely over (enough) time migrate in the direction of an applied electric field.

Eventually, after enough time for a given temperature and operating voltage, enough atoms will leave a certain area so as to leave behind a void (in some views it is the voids that diffuse, coalescing into larger and larger voids) and the presence of that void requires the metal wire in that location to be narrower than it otherwise had been designed to be. Thus the narrower metal gets hotter (think how a candescent lightbulb works) and that promotes even more atoms to migrate making the void even bigger.

This kind of degradation will limit the clockspeeds, you have to clock your chip slower and slower, if the void is on a speedpath limiting circuit. Alternatively you can have degredation in the transistor, hydrogen diffusion into the channel for example, which likewise reduces the drive current and switching speed and thus lowers the maximum clockspeed of the chip over time.

(there are more examples, these are just some ways a chip slowly dies while operating under normal conditions)

 

[dakels]

Not a CPU but my GPU showed signs of "wearing out". My old Radeon 9800 Pro was OC'd to about 425mhz for the longest time from a stock 350 I think. After about 3 years of solid service it started getting flaky. I lowered the OC to about 400mhz to keep it stable. I could run it at 425 with additional cooling. A few months later started getting more artifacting and had to lower it to about stock 350mhz. A few months later it would have to be underclocked around 300mhz to not show artifacts. I don't know enough to tell you what was going on, but in lay terms, it sure was "wearing down". I replaced it with a X1950 pro.

 

[nerp]

Thanks for the information Idontcare.

 

[brettjrob]

I can't rule out the mobo as a factor, but when I first purchased my current rig in summer 2003, I was able to clock my P4 2.4 at 3.0 just like everyone else was doing at the time. Now it flakes out in Prime95 after a few minutes at that setting. It's not the RAM, as I set 5:4 when upping the FSB from 200 MHz to 250 MHz. Thermal paste degradation? Could be, though I have a Zalman 9700 on it, so I'd think the cooling should be more than adequate for the relatively modest overclock.

I'm about to upgrade to an E8400 or Q6600 system, and in all honesty, my plan after this experience is to stick with stock speeds until they become inadequate or limiting for my use, which shouldn't be for at least a year or two.

 

[GundamF91]

Originally posted by: brettjrob
I can't rule out the mobo as a factor, but when I first purchased my current rig in summer 2003, I was able to clock my P4 2.4 at 3.0 just like everyone else was doing at the time. Now it flakes out in Prime95 after a few minutes at that setting. It's not the RAM, as I set 5:4 when upping the FSB from 200 MHz to 250 MHz. Thermal paste degradation? Could be, though I have a Zalman 9700 on it, so I'd think the cooling should be more than adequate for the relatively modest overclock.

I'm about to upgrade to an E8400 or Q6600 system, and in all honesty, my plan after this experience is to stick with stock speeds until they become inadequate or limiting for my use, which shouldn't be for at least a year or two.

Did you overvolt on vCore?

 

[taltamir]

I would think that with a name like "CPU degrading due to heat/voltage" this thread WOULD be about overclocking.

Real and measurable? Has it actually happened to you, and have you definitely isolated the CPU as the culprit?

To answer your questions... Yes it is real! Sorta, it is measurable, but not in exact numbers! Yes, it has happened to me.

Don't let your CPU run into the 80c + temperature range (ideally, keep it below 60c, but that just might be paranoia).
Don't let your C2D / C2Q be over 1.4 volts.

If you follow the above two rules your CPU should be obsolete long before it fries.

 

[nyker96]

I think after about 1 year useage on my old A64 system I had to back down OC. But now I think about it may be the HSF is dusty and the fan isn't working as well as b4.

 

[MoMeanMugs]

As you increase the voltage, there is a larger probability for quantum tunneling at the gates of the FET's. Current there is a bad thing and creates a ton of heat. Like Nathelion said, this becomes a bigger issue as the nodes shrink because the insulating substrate thickness decreases as well. It'll be interesting to see how this is dealt with at a 32 nm process.

 

[VirtualLarry]

Originally posted by: taltamir
Don't let your CPU run into the 80c + temperature range (ideally, keep it below 60c, but that just might be paranoia).
Don't let your C2D / C2Q be over 1.4 volts.

If you follow the above two rules your CPU should be obsolete long before it fries.

I should be the poster child for CPU degradation then. My CPUs run at 83C, at 1.36v (CPU-Z), 1.425v (BIOS). We'll see.