Does an overclock go bad?

[loganex]

I built a nice gaming system in Feb 09 around a Q9300 that I got on sale from microcenter. I oc'd it to 3.5ghz with voltage 1.35 (showing up as 1.28-1.32 depending on speedstep in windows 7 x64). See my earlier thread here for my settings: http://forums.anandtech.com/showthread.php?t=270927. (System has changed slightly, see below.)

At the time I ran all kinds of prime 95 torture tests, even looping 3DMark06 over the top of it to try to kill the system. I did intel burn test and had everything passing with flying colors before I left it alone. It ran prime 95 for 36 hours with no errors. I haven't messed with it since then.

But just yesterday I installed a new Thermaltake Frio. I've been folding a lot and my 92mm CPU cooler wasn't keeping up with the heat so well. So after the Frio was on, I pulled out prime 95 again to stress the cpu and see what temps I was getting. I had an error on core 0 within 10 minutes. Shut it down, started again, and same story - error within 10.

So do overclock setting go bad? Do I need to go back in and readjust? Update bios? Any idea what could have gone wrong since I set it up and tested it? At the time it was stable under windows vista, and now we're in windows 7, would that be the difference?

Intel Q9300 @ 3.5 GHz
Thermaltake Frio with both fans push/pull (with AS 5)
Gigabyte GA-EP45-UD3R
Patriot Viper 4 GB (2 X 2GB) PC8500 @934
3 X WD Caviar Black 640GB
EVGA GTX 460 1GB EE
Corsair 650TX
Windows Vista Business 64-bit
All in a Cooler Master CM 690

 

[Daedalus685]

A CPU can certainly degrade if exposed to high voltage. 1.35 seems a touch on the large side for these CPUs.

Other factors that can come into play other than the CPU simply not ageing gracefully are simple ones such as time of year (ambient temeprature), cleanliness of the heat sink, and condition the heatsink CPU bond. If you travelled at all with the computer (and maybe even if you didn't) re application of thermal compound may not be a bad idea.

At any rate, it is very normal for a CPU to require more voltage to get to the same clocks as it degrades. You have a pretty large OC there and should expect a noticeable shortening of its life time. Might simply have to pull back the OC.

 

[Idontcare]

A CPU can certainly degrade if exposed to high voltage. 1.35 seems a touch on the large side for these CPUs.

loganex, just to elaborate on what my colleague Daedalus685 has already stated above, a CPU degrades with any voltage...its just that more voltage makes it degrade faster than less voltage does.

And the response function is not linear. A 5% increase in voltage does not mean 5% lower lifespan for the chip, it could actually mean a 20% lifespan reduction for the chip (or more).

That sounds needlessly scary, rest assured most cpu's (even OC'ed ones) will have a lifespan that exceeds 10yrs of constant usage...far longer than most people will bother to keep using the CPU.

But yes, if you over-volt the chip then at some point you can expect it to have degraded enough that it will no longer be stable at a given clockspeed and voltage...in those cases you will find yourself slowly rolling back the overclock to stay ahead of the pace of degradation.

Some find their chips won't even work at stock clocks any longer. Upgrade time!

 

[OCGuy]

To sum up what IDC said: Electromigration is a mother******!

 

[Martimus]

As current flows through the doped area, the characteristics of that area will change. (Which makes logical sense, when you think of what is happening on the electron level) The more current you use, the more it will change. (Again, logical) Higher voltages causes more current to flow though the doped region (V=IR), which makes the region change quicker. So higher current, or longer exposure to current will make a transistor (or any doped silicon) change its electrical characteristics faster than lower current, or shorter exposure to current.

 

[loganex]

Upgrade time!

All right, just what I wanted to hear! Now to convince the wife...

Thanks guys, I guess I'll find some time to go back in and ease down my OC. It's been 18 months, so I have to do all my "how to OC" reading again.

 

[darckhart]

it's not that your overclock has "gone bad," but that your components have changed. when you first set everything up, for example, they were all comfortably sitting on a shelf being unused. now you're trying to set up things after it's been operating full tilt for X time. contacts within the cpu and pins are bound to be in a different shape (Y position with Z conduction). if you're curious, take out your cpu and look at the contacts underneath and the socket on the motherboard. see if everything is clean and shiny as it was when new. most likely not. not only that, but the motherboard (nb/sb chipsets, traces) and its subsystems are also in different shape. btw, changing a hsf is a big change because you're asking it to redistribute a bunch of forces during contact such as maybe you had clips and now push pins or screws. it's difficult to get the exact same contact as before. i noticed this myself when switching from a scythe to a noctua.

now are you asking if over volting has somehow destroyed your cpu sooner? possibly, but the lifespan of these chips, even at your 1.35v with adequate cooling should definitely be longer than 2 yrs on average.

but hey, now you have excuse to upgrade! hehe

 

[GuitarDaddy]

I agree with what others have said, but also would point out that you made two significant changes. You changed the OS which can most definately change your O/C and you installed a new CPU cooler which also may be affecting things. I would first double check the fit of the CPU cooler and the application of the thermal compound. How are your temps? It could possibly improve as your thermal compound sets.

 

[loganex]

How are your temps?

Temps are actually improved with the new cooler. The Frio is much bigger than the 92mm cooler I had on there before. We're down about 8 degrees C from before, reading 64C in RealTemp under full load.

I haven't experienced any trouble with the CPU or the stability of my system, it's just that prime95 is finding errors now where it wasn't before. So since it has been 18 months, it seems like I just need to get back in there and redo the overclock settings to match my current setup and account for any degradation that might have happened.

 

[Rifter]

I have ran overclocked systems going back to the K6/Pentium days and i have never had a CPU die on me, lots of motherboards/RAM but never CPU's. I have had chips degrade where they would not take the same overclock after 3-4 years and they would get unstable and i would have to back off the overclock or raise the voltage though.

 

[Idontcare]

It is rare to have a chip die. Sadly my QX6700 died after about 2yrs. I didn't kill it doing suicide runs or nothing, just some serious 24x7 100% load for 2yrs 4GHz/1.55V at -10C (vaporcooling).

I loved that chip, without a doubt I'd buy it again were it Oct of 2006 again. But yeah, cpu kill is rare.

 

[Acanthus]

It is rare to have a chip die. Sadly my QX6700 died after about 2yrs. I didn't kill it doing suicide runs or nothing, just some serious 24x7 100% load for 2yrs 4GHz/1.55V at -10C (vaporcooling).

I loved that chip, without a doubt I'd buy it again were it Oct of 2006 again. But yeah, cpu kill is rare.

Yeah, 1.55v is pretty hardcore, im pushing it at my voltage.

 

[taltamir]

Any electronic device degrades when a voltage runs through... not high voltage, ANY voltage.
When you run your CPU it degrades, even if it is under volted it still degrades. The higher the voltage you run through it, the FASTER it degrades. Also, the faster its clock the faster it degrades. A non overclocked CPU has a lot of headroom, an overclocked one doesn't... it is completely normal to have to reduce your OC every now and then.

How does intel bin a CPU? according to intel they set VID and clockspeed so that after 3 years of (constant) operation, 10% of their CPUs will no longer be stable at their default values (they would still be functional at lower values though). Intel spends a whole lot of efforts into measuring and quantizing the exact rates of degradation under various scenarios.
Anyways, since a home user isn't constantly pushing their CPU (unless they do distributed computing), then it should last much longer at default clockspeed and voltage...but if you push it to its max stable speed then it will certainty require reducing the OC over time.

 

[CTho9305]

To sum up what IDC said: Electromigration is a mother******!

Not really. Electromigration is straightforward to design around, and there are a few effects that reduce its severity as manufacturing processes allow narrower wires (for example, if the width of wires is smaller than the size of the metal grains, mass transport is significantly reduced because you can no longer transport atoms along grain boundaries and have to transport across them instead). Nowadays the primary failure mechanisms would probably be NBTI, HCI, and TDDB. The first two slow down transistors as they age, which usually results in a reduced maximum frequency (though in certain cases it can cause a hard failure, for example by delaying a clock and causing a hold time failure). I think dielectric breakdown is always a hard failure.

Wikipedia says: "In modern consumer electronic devices, ICs rarely fail due to electromigration effects. This is because proper semiconductor design practices incorporate the effects of electromigration into the IC's layout. "

edit: taltamir, do you have a source for that Intel lifetime claim?

 

[taltamir]

Wikipedia says: "In modern consumer electronic devices, ICs rarely fail due to electromigration effects. This is because proper semiconductor design practices incorporate the effects of electromigration into the IC's layout. "

If you are overclocking the heck out of something it isn't running as a "proper design" now is it?

edit: taltamir, do you have a source for that Intel lifetime claim?

Nope. Saw a video tour of intel along with a tech article in some review site where they explained it, but until I get a computer attached to my brain (I can't wait) I am unable to properly hyperlink my knowledge. It is my most greatest of wishes that I could. I do, however, try to "tag" memories with a memorized metadata tag, such as "dream", "personally verified fact", "unknown source", "known to be unreliable source" etc. For example, I have a vivid recollection of breathing underwater with a memorized metadata tag that says it is a dream I had from when I was 11 (this is my earliest memory to be tagged), and living in bet arye. The really annoying thing is that I know for a fact I have some corrupted memory metadata tags, it pisses me off to no end at how imperfect human memory is!

 

[SZLiao214]

Since the k6 days i've never had a cpu go bad on me. I have had several motherboards slowly develop additional quirks and annoyances till i replace them out of anger instead of actual need.

The cpu is so mething that is over engineered. As long as you don't try anything too crazy you should be fine

 

[taltamir]

I have had several motherboards slowly develop additional quirks and annoyances till i replace them out of anger instead of actual need.

This usually happens to me too...

The cpu is so mething that is over engineered. As long as you don't try anything too crazy you should be fine

something crazy like OC and overvolt it heavily or/and run distributed computing apps that keep it at 100% utilization 24/7?

I have to say though, CPUs tend to be very solid... if you don't do something to them, then they usually don't go bad,

 

[Martimus]

Not really. Electromigration is straightforward to design around, and there are a few effects that reduce its severity as manufacturing processes allow narrower wires (for example, if the width of wires is smaller than the size of the metal grains, mass transport is significantly reduced because you can no longer transport atoms along grain boundaries and have to transport across them instead). Nowadays the primary failure mechanisms would probably be NBTI, HCI, and TDDB. The first two slow down transistors as they age, which usually results in a reduced maximum frequency (though in certain cases it can cause a hard failure, for example by delaying a clock and causing a hold time failure). I think dielectric breakdown is always a hard failure.

Wikipedia says: "In modern consumer electronic devices, ICs rarely fail due to electromigration effects. This is because proper semiconductor design practices incorporate the effects of electromigration into the IC's layout. "

edit: taltamir, do you have a source for that Intel lifetime claim?

I apologize for the late response, as I just read this, but I did want to ask about this subject. I personally tested probably about 1000 zener diodes for their reaction to high current loads in a previous job. I found that nearly 100% of them would have a higher breakdown region after testing than before. They were mostly still in spec, but the phenomenon of the higher output voltage was very repeatable after a large current rush.

I never quite knew for sure why they did such a thing, but I figured that the dopant likely changed with the additional current flowing through it. Perhaps it was electromigration in a sense, in that maybe the current litterally removed some doped particles, but I never did any testing to find out why. (It wasn't important why, it was just important to know what happened and design the circuitry accordingly)

Since you have quite a bit of knowledge in this area, maybe you can suggest some reasons why this may have happened? Obviously a transistor is very similar to a Zener diode in composition and function, so the increased need for 'switching' voltage after prolonged use is likely the same phenomenom.

(I apologize if my terminology is incorrect. I haven't done any actual electrical design work now in about 7 years, and I have always had bad memory when it came to terminology anyway.)

 

[Any_Name_Does]

If you are overclocking the heck out of something it isn't running as a "proper design" now is it?


Nope. Saw a video tour of intel along with a tech article in some review site where they explained it, but until I get a computer attached to my brain (I can't wait) I am unable to properly hyperlink my knowledge. It is my most greatest of wishes that I could. I do, however, try to "tag" memories with a memorized metadata tag, such as "dream", "personally verified fact", "unknown source", "known to be unreliable source" etc. For example, I have a vivid recollection of breathing underwater with a memorized metadata tag that says it is a dream I had from when I was 11, and living in bet arye. The really annoying thing is that I know for a fact I have some corrupted memory metadata tags, it pisses me off to no end at how imperfect human memory is!

I do not have a superman complex; for I am God, not superman

???