CPUs dying slowly?

[Agentbolt]

My AMD CPU is being annoying these days. I OCed pretty much right out of the box to 3200+ speeds (it was a Barton 2500+) because it seemed that was pretty much what everyone else was doing. It would boot at 1.65 volts, but to be stable it had to put at 1.75. That worked well for about 8 months, but now I have to up the voltage to 1.825 or even 1.85 before it'll stay stable without locking up.

1.85 volts is gonna cook this thing in a matter of months, most likely. So what I'm wondering is, can a CPU start to die and need more voltage to work correctly? Or do they just instantly go from working to crapped out? I don't know a whole lot about CPU architecture, but I don't see how the CPU could be dying slowly and needing more voltage as it does to stay stable?

If it's a "symptom" of the CPU getting sicker and dying, then I'll just start socking away some money to grab a mobile 2500+ XP for when it croaks (I'm not replacing the mobo as well, don't have the $, so I'll be sticking with Bartons) But if I'm right and CPUs are either "working 100%" or "dead" then what the heck else could be neccessitating the extra voltage? Like I said, at 1.85 volts I think this poor guy is going to heaven soon, so if there's something I can do to make it stable at 1.75 again, I'd sure like to know what it is.

 

[Rhin0]

I suppose it could be a PSU problem

 

[Agentbolt]

Do you mean rails, or that it's underpowered? I use a 420 W Thermaltake PurePower, so it's highly unlikely that MY system is drawing too much juice for that.

 

[Amaroque]

I have an AXP 2000+ (1.67MHz) It used to clock to ~1.8MHz. But now it won't even run at stock speed, even with a strong PSU. And yes, I tried it in different boards, different PSU's.

I had it as a crunching drone on Prime95 24/7 for about 2 years. That is the first CPU that ever did that to me. :roll:

 

[funggorgor]

never OC a cpu , its not worth to do so

 

[Agentbolt]

Why not? I can see how I've been idiotic by OCing on stock cooling, but OCing in general isn't neccessarily a bad thing to do. I definitely notice the difference between 2500 and 3200 speeds.

I'd still really love to know why the hell this CPU is doing this, it's driving me nuts. Perhaps it's been cursed or something. I don't see how a CPUs performance can degrade, it either works or it doesn't. Right? Maybe? I don't know anything.

 

[otispunkmeyer]

when you think about it its not really that big of an overclock anyway 1.85Ghz ----> 2.2Ghz is less than 400mhz, doesnt seem that big to me, not when people are hitting up to 2.6Ghz, and people with intels gettin nearly another ghz out of their chips

heh remember people used to say AMD chips were crap because they were fast when you bought them then they degraded and slowed down? everyone knows its bull.....but yours seems to be doing jus that lol

just keep an eye on it, and keep an eye on temperatures, can see alot you can do, if its gonna croak its gonna croak, jus be there with a replacement when it does

 

[dug777]

Originally posted by: funggorgor
never OC a cpu , its not worth to do so

stupidy stupid wrong forum 2 say something like that

Xactly the same prob i've got with my xp2000+ (@2600+), over the last eight months i've slowly had 2 pump the voltage 2keep it prime stable, and with the stock hsf i dont really want 2go over 1.7v so i ve just had 2 say f&*k it, and haven't had any probs in the last ~2months, even tho it is impressively prime unstable

I suspect it's a PSU thing, as its a cheap no name piece of cwap, and currently i m running w/out my burner & FDD just 2 keep it running- whereas i had no probs with all that 6months ago...

 

[PingSpike]

Originally posted by: funggorgor
never OC a cpu , its not worth to do so

Why not? Or are you one of those people who believe running 10mhz over stock causes spontanous combustion of the CPU die?

 

[Amaroque]

Originally posted by: funggorgor
never OC a cpu , its not worth to do so

The point I was trying to make is that I've used and overclocked dozens of CPU's under a 24/7 heavy load, and that only ever happened to one CPU.

Yes, I abuse my crunchers, but that's after I'm done using the CPU in my main rig. If you take some proper care, there isn't any worries.

 

[pm]

Yes, this is expected behavior. All CPU's have this problem - that they gradually slow down with time. Increasing the voltage increases the rate at which it slows down.

The specific cause of the problem is a little esoteric, but since my statement above is likely to be questioned... here's the specific mechanism.

The problem is specifically caused by a mix of hot-electron gate impact ionization increasing the threshold voltage of the nFET's over time as energetic electrons tunnel from the conduction channel into the gate oxide material. The gradual build-up over time of electrons in the gate creates effectively a negatively charged source within the gate that needs higher voltage to overcome in order to turn on- or more time for the gate potential to rise to a higher value. The problem is also seen on pFET's to a lesser degree caused by a lesser understood phenomena known as "PMOS BTI" (p-type Metal Oxide Semiconductor Bias Temperature Instability" that is thought to be caused by holes attaching to lattice charge traps at the oxide/channel interface resulting in a similar problem of the Vt shifting.

There's not really a laypersons explanation of the problem, but googling for "hot-electron gate impact ionization" or "bias temperature instability" leads to some interesting esoteric reading on the subject.

 

[AwesomeJay]

Originally posted by: pm
Yes, this is expected behavior. All CPU's have this problem - that they gradually slow down with time. Increasing the voltage increases the rate at which it slows down.

The specific cause of the problem is a little esoteric, but since my statement above is likely to be questioned... here's the specific mechanism.

The problem is specifically caused by a mix of hot-electron gate impact ionization increasing the threshold voltage of the nFET's over time as energetic electrons tunnel from the conduction channel into the gate oxide material. The gradual build-up over time of electrons in the gate creates effectively a negatively charged source within the gate that needs higher voltage to overcome in order to turn on- or more time for the gate potential to rise to a higher value. The problem is also seen on pFET's to a lesser degree caused by a lesser understood phenomena known as "PMOS BTI" (p-type Metal Oxide Semiconductor Bias Temperature Instability" that is thought to be caused by holes attaching to lattice charge traps at the oxide/channel interface resulting in a similar problem of the Vt shifting.

There's not really a laypersons explanation of the problem, but googling for "hot-electron gate impact ionization" or "bias temperature instability" leads to some interesting esoteric reading on the subject.

NERD ALERT

 

[yuchai]

I suppose it could also be because over time the HS/F starts getting clogged with dust, degrading cooling performance and thus stability?

 

[Amaroque]

Thanks for the informational reply pm. I'm assuming that what you're referring to the same as, or similar to "electro migration"

 

[ScrapSilicon]

Originally posted by: Amaroque
Thanks for the informational reply pm. I'm assuming that what you're referring to the same as, or similar to "electro migration"

entropy

The tendency for all matter and energy in the universe to evolve toward a state of inert uniformity.

 

[Googer]

CPU's Are said to have an 8-12 year lifespan acoriding to Intel (correct me here if necessary). Some can go longer, but they are design for about a decade of use.
'

If you overclocked it, you can kiss any warrenty good bye. Any time you run anything out of its design specification you shorten its life. It is true for cars, televisions, kitchen blenders, lawn mowers, elevators, paint, computers, and just about everything.

 

[m3rcury]

Originally posted by: pm
Yes, this is expected behavior. All CPU's have this problem - that they gradually slow down with time. Increasing the voltage increases the rate at which it slows down.

The specific cause of the problem is a little esoteric, but since my statement above is likely to be questioned... here's the specific mechanism.

The problem is specifically caused by a mix of hot-electron gate impact ionization increasing the threshold voltage of the nFET's over time as energetic electrons tunnel from the conduction channel into the gate oxide material. The gradual build-up over time of electrons in the gate creates effectively a negatively charged source within the gate that needs higher voltage to overcome in order to turn on- or more time for the gate potential to rise to a higher value. The problem is also seen on pFET's to a lesser degree caused by a lesser understood phenomena known as "PMOS BTI" (p-type Metal Oxide Semiconductor Bias Temperature Instability" that is thought to be caused by holes attaching to lattice charge traps at the oxide/channel interface resulting in a similar problem of the Vt shifting.

There's not really a laypersons explanation of the problem, but googling for "hot-electron gate impact ionization" or "bias temperature instability" leads to some interesting esoteric reading on the subject.

Oh, thats all? Pffft, no wonder. Here I was, thinking it was something complicated. =)

 

[InlineFive]

If they are being fed too much voltage then I would say absolutely. I.E., bad PSU or overclocking.

 

[pm]

Originally posted by: yuchai
I suppose it could also be because over time the HS/F starts getting clogged with dust, degrading cooling performance and thus stability?

Yeah, this is actually a more common explanation that's a likely possibility. Forgot about that. <laughs at himself>

Originally posted by: Amaroque
Thanks for the informational reply pm. I'm assuming that what you're referring to the same as, or similar to "electro migration".

Electromigration is a different mechanism. Electromigration (EM) is caused by metal atoms in the wires 'migrating' down the wire under the steady bombardment of electrons causing either an 'open' (where the wire essentially breaks), or the build-up of metal atoms in a pile causes a 'short' to another nearby wire. EM used to be a bigger problem back in the 'old' days when people used aluminum wires which are prone to grain boundary formation and are lighter atoms. Now with the heavier copper atoms making up the bulk of the wire, the problem is far less prevalent. In fact, EM has ceased to be a real problem in current process technologies barring a designer really messing up. As wiring continues to get thinner in future generations it will no doubt come back though.

EM usually exhibits itself as a 'hard' logic failure ('hard' referring to the fact that no combination of temperature, voltage or frequency will cause the part to generate the correct result). Hot-E/BTI show up as a gradual slow-down which is effectively a 'soft' failure in that you can get the part to work by changing the operating parameters.

 

[Googer]

Originally posted by: PorBleemo
If they are being fed too much voltage then I would say absolutely. I.E., bad PSU or overclocking.

Send me a letter, I'd like to hear from you: porbleemo@gmail.com

You must get S*it loads of Spam at porbleemo@gmail.com, Posting it on a website like this is how spammers get them in the firstplace. They have computers that crawl the internet looking for email addresses just like the one you have posted.

YOU ARE SPAM BAIT!