E8400/8500 degradation myth possibly busted?

[CrazyNutz]

Wolfdale degradation myth possibly busted?

Im sure a lot of you have had this happen, you clock your new e8X00 up and run the usual torture test
until your find the max stable OC @ some vcore, and temps you are comfortable with. That's the
Idea right. Right!!

By now your prime stable for >8 hours, and your happy because your at 4ghz (or more) and you
didn't even break the bank when you made that sweet CPU purchase. Still with me?

Ok now at some point ( Next day/Next week) you'll run that torture test again and then OMG!!!
WTF??? Prime95 FATAL ERROR!! But it used to be prime stable for hours and now it's failing in
less than 5-10 minutes? Instantly that horrifying thought comes to mind "Degradation".

Ok so that's the scenario for most. At least it was for me. Iv'e had this happen with two E8400's
so far. The first one I upped the vcore passed 1.4 and after a couple of days when prime started failing
I thought for sure this is the symptoms of the dreaded degradation problem. The second E8400 I decided
I would never go past the 1.36 safe (or safer) limit (actually it's never been passed 1.33), and I
would not see this problem again, Wrong!! I did, but this time I was determined to figure this out!


And here is what I found:

If my system is either off, or has been at Idle for a long period of time (i.e 18 - 24 hours) and I run prime
(especially the 10k FFT's) It will fail in < 5 minutes sometimes < 1 min. If I then run prime again it will take
prime a little longer to fail but maybe still < 5 mins. If I run It again it will take a lot longer to fail > 20 mins.
And then I run it once more Bam!!! Smooth sailing no more failing, It's back to the stable state it was
initially. 10k priming for hours on end.

So why?
Answer: Not sure but it seems apparent that it's needs to get nice and warm to be prime stable, and
maybe It's not warm in a literal sense, I just don't know. However It does seem to need this so called
"warm up period" more often to become stable when the ambient room temps are lower than 70F or so.

Please note these are just my findings on my rig, I would like to ask you guys to put this to the test
to see if this "Warm Up Period" holds true with others. If you would like to participate I urge you
to use Prime95 with 10k FFT to test with. I have found the 10k FFT's to put the most stress on the
wolfdales!

Thanks



UPDATE/CONCLUSION

Well guys, I can confirm now that my CPU's did not actually degrade.

It is apparent now that the 45nm CPU's have a burn-in period of about
1 - 2 weeks, after this time the CPU's will need a small bump in vcore
(.024mv - .050mv) to regain initial prime stability.

My CPU's are rock solid stable now, and are not showing any other signs
of further degradation, I even had the vcore on one up close to 1.5v for
some suicide runs, and still did not hurt it.

[kup]

Interesting find, it may also explain why my Quad goes in and out of being rock solid stable @ certain clocks...

I'll keep an eye on this thread.

[~aoe~]

/subscribed

I've just got one of these and I'll be installing it shortly. Have read many posts regarding the supposed 'degradation', so am clearly interested in any findings.

On a side note, how come all these 'Wolfdale' threads never include the E8200 .

[hstuehmeyer2000]

are you sure it is not just a bug in orthos

[zlojack]

Hmmm... interesting bit of data. I'll keep an eye on this...

[mrcape]

The warm up didn't help my situation, but I'm not looking at just passing prime, I'm looking at what vcore it takes to pass prime at a specific clock. Crazy Nutz, can you provide some more detail about what vcore was needed at a certain clock to pass prime? Did you ever find that low spot during your early testing?

For example, before taking an e8400 over 1.3vcore I could run prime for hours at 4ghz with 1.22vcore. I started with everything low out of the box to try and identify this very issue. I was able to run Orthos blend priority 9 for several hours at 1.28vcore at 8.5x500 or 4.25ghz. I had several successful runs documented with all settings recorded.

After getting a handle on what the chip could do, I had a night of moderate benchmarking on air with the same chip. After that night, it now takes +.05v to be stable at the same exact clocks as before that night. The additional vcore amount (+.05v) is needed at all of the previously recorded clocks no matter how warm or cold the chip is.

The benchmarking never went over 1.42vcore and didn't cause high temps. Max temps ever were 56-58c. I tried re-flashing, fresh set of ram, just about everything I could think of to rule out other issues.. Nothing I did could get me back to prime stable with that low vcore.

Still a mystery to me. I appreciate you starting this thread Crazy Nuts. Hopefully we can get some other user experiences here and learn something about this.

The big challenge I see in debunking anything is the fact that barely anyone will record their settings on the way up or shoot for low vcore to begin with. So if they jump right to 1.45vcore, they might never know that the chip they're using has degraded lol.

[CrazyNutz]

Originally Posted by ~aoe~

/subscribed

I've just got one of these and I'll be installing it shortly. Have read many posts regarding the supposed 'degradation', so am clearly interested in any findings.

On a side note, how come all these 'Wolfdale' threads never include the E8200 .

84/8500 are more common I guess, however the 82's are the same core so most of the things we discuss pertain to 82's as well.

[CrazyNutz]

Originally Posted by mrcape

The warm up didn't help my situation, but I'm not looking at just passing prime, I'm looking at what vcore it takes to pass prime. Crazy Nutz, can you provide some more detail about what vcore was needed at a certain clock to pass prime? Did you ever find that low spot during your early testing?

For example, before taking an e8400 over 1.3vcore I could run prime for hours at 4ghz with 1.22vcore. I started with everything low out of the box to try and identify this very issue. I was able to run Orthos blend priority 9 for several hours at 1.28vcore at 8.5x500 or 4.25ghz. I had several successful runs documented with all settings recorded.

After getting a handle on what the chip could do, I had a night of moderate benchmarking on air with the same chip. After that night, it now takes +.05v to be stable at the same exact clocks as before that night. The additional vcore amount (+.05v) is needed at all of the previously recorded clocks no matter how warm or cold the chip is.

The benchmarking never went over 1.42vcore and didn't cause high temps. Max temps ever were 56-58c. I tried re-flashing, fresh set of ram, just about everything I could think of to rule out other issues.. Nothing I did could get me back to prime stable with that low vcore.

Still a mystery to me. I appreciate you starting this thread Crazy Nuts. Hopefully we can get some other user experiences here and learn something about this.

My Q748A144 takes 1.328 (real CPU-Z reported) to be 8 hours Prime Blend Stable @ 3960mhz.
My Q747A takes 1.368 (real CPU-Z reported) to be 8 hours Prime Blend Stable @ 3780mhz

As for the "low spot" you speak of, I found it on both CPU's, and
I thought that I lost it due to degradation but so far It turns out it was
due to the needed warm-up period.

Also what criteria do you consider passing prime?


Thanks for your appreciation, I just think we need to shine some light on the issue

[cadaveca]

me, 24 hrs. iwas one of the first to report degredation..however, prime is not what showed degredation...required voltages for boot increased, never mind running prime.


I mentioned your findings, in my own form, in another thread...you cannot just start prime, have it fail, and have to start over!


Although, from what you describe, it's like a capacitor needs to build up a charge, and when it is not fully charged, you have no stability.

More testing must be done, as no single cpu can "confirm or deny" existenec of any problem.

[mrcape]

I don't think a huge block of time matters to test for this. Maybe to confirm stability, at a given clock, but not to gather data on performance vs voltage. I think passing a 1.5-3 hours at a few different clocks is enough of a baseline. Then if you go back and re-test and you need the exact same amount of increase to hit each clock, I think that says something.

[Speederlander]

Higher degradation risk with 45nm is real. The physics of the smaller shrink and the new gates makes it inevitable. Initially people were clocking their 45nm chips using 65nm voltages, often bench voltages. And they killed or degraded their chips. But, many other people mistook failures for degradation when they simply were never stable in the first place.

[CrazyNutz]

Originally Posted by cadaveca

me, 24 hrs. iwas one of the first to report degredation..however, prime is not what showed degredation...required voltages for boot increased, never mind running prime.


I mentioned your findings, in my own form, in another thread...you cannot just start prime, have it fail, and have to start over!


Although, from what you describe, it's like a capacitor needs to build up a charge, and when it is not fully charged, you have no stability.

More testing must be done, as no single cpu can "confirm or deny" existenec of any problem.

Your case sounds more like actual degradation.
Basically I started this thread because I think many people are Misdiagnosing
there CPU as Degraded. Maybe a different title would have been appropriate.

And why cant I just start prime over after it has failed?

Also I have found this same problem with 2 e8400's, and I have a third coming next week. However this is not to say there could not be something
associated with my rig (i.e. mem/mobo/psu) that is causing this, that's why I'm
asking others to try the warm up test.

[cadaveca]

sry, i meant that after a failure, you should restart as you have suggested. reading over it, i was not quite clear. I was more saying that overclock process need not be restarted. Nevermind we(me) are not sure how extra cache may affect prime's stability either, or whether some noted bugs in whitepapaers already cover the issue...hard to say just yet.

From everything i have heard, I think my own cpu is one of few that have really degraded, apart from a couple others. I think we can safely say 5-6 have degraded/died on this forum, others were not stable to begin with.

Thread title says "possibly", that's good enough for me. I posted for discussion, only, not to attack your position on this issue.

[mrcape]

The more feedback the better Crazy Nutz. Like I said though, many people will never record those early baseline settings. That means that there is a lot of bad data.

I've made the same point as Speederlander in other discussions, that people just dove right into these as if they're using 65nm and hosed them from the get go. No way to tell anything then.

[cadaveca]

We had Intel officals say to go slow on voltage increases, and that's what I did, still degraded, tho. My cpu shows 40c temps via DTS, 35c via "cpu temp", when subzero, so my cpu was a strange one from the start. I merely think a few bad ones slipped through the binning process in the rush to get these cpu's out for launch.

I also know it's probably something I myself did that hurt the cpu...I don't care. I did'nt give it too much pll, FSB, vcore, or nothing, but i am thinknig that the difference required between these new cpus and current chipsets may paly a role...something has to covert 1.45v chipset to 1.2v cpu....(example only)

[mrcape]

Yeah I too am sure that the superpi benching I did, even though it was only 1.42vcore, was what effected my chip. And I don't care about the chips either. Hell, I want more of them asap. At this point I just want to find a good method to find exactly where that edge is and not cross it. It may be different on every chip, but maybe there's a ballpark edge based on your vid.

I'm guessing the higher 545fsb+ I was pushing up with 1.42vcore and running superpi32m was the culprit in my case.

About chipsets and bioses, I'm real curious to see what happens with overclocking the 45nm chips on the p45, or better yet the next iteration after that.

[Cronos]

Concerning "warming", it will take some several seconds for CPU to warm nearly to it's final temperature. Especially on water, there is usually only few C increase after the initial jump.

CrazyNutz and others,

can you try to run Linpack (64 bit if possible) ?
I know at least one person with 8400 who said he was having "unstable Prime instability", when he tried to run Linpack (though only 32bit, which is nothing comparing with 64b), he had nearly 100% Residual failure rate! I have never seen such bad pattern before.

Like this:

CPU frequency: 4.050 GHz
Number of CPUs: 2
Number of threads: 2
Parameters are set to:

Number of tests : 1
Number of equations to solve (problem size) : 11500
Leading dimension of array : 11500
Number of trials to run : 20
Data alignment value (in Kbytes) : 4

Maximum memory requested that can be used = 1058234096, at the size = 11500
============= Timing linear equation system solver =================

Size LDA Align. Time(s) GFlops Residual Residual(norm)
11500 11500 4 55.419 18.3002 1.138041e-010 3.036202e-002
11500 11500 4 54.993 18.4419 1.271550e-010 3.392393e-002
11500 11500 4 55.208 18.3702 1.181119e-010 3.151132e-002
11500 11500 4 55.196 18.3743 1.082662e-010 2.888457e-002
11500 11500 4 55.206 18.3709 1.080989e-010 2.883993e-002
11500 11500 4 55.075 18.4146 1.197248e-010 3.194163e-002
11500 11500 4 55.128 18.3969 2.028002e-010 5.410548e-002
11500 11500 4 55.199 18.3732 1.173892e-010 3.131849e-002
11500 11500 4 55.121 18.3991 1.292362e-010 3.447918e-002
11500 11500 4 55.205 18.3713 1.302878e-010 3.475976e-002
11500 11500 4 55.608 18.2380 1.525874e-005 4.070909e+003
11500 11500 4 55.309 18.3366 1.314611e-010 3.507278e-002
11500 11500 4 55.344 18.3251 1.271538e-010 3.392363e-002
11500 11500 4 55.311 18.3361 1.191961e-007 3.180058e+001
11500 11500 4 55.436 18.2947 1.475248e-010 3.935845e-002
11500 11500 4 55.353 18.3222 1.328986e-010 3.545627e-002
11500 11500 4 55.345 18.3248 1.186223e-010 3.164748e-002
11500 11500 4 55.730 18.1982 1.252908e-010 3.342659e-002
11500 11500 4 55.282 18.3455 1.191869e-007 3.179812e+001

[Leeghoofd]

I have degradation for sure, CPU was rockstable at 4ghz with 1.36 volts, after a weekend of benching sessions at nite ( with 3°C ambient ) and speeds over 4.5GHz and 1.55 volts max.. my CPU needed already 1.38 to be stable at 4ghz again... A month later I'm requiring already 1.6 volts to bench at 4.5Ghz, same tests etc... Intel will get this one back for sure...I already mailed them about the bogus temp readings, secondly this CPU never ran at 1.125volts like the VID tells me , not on 3 different mobo's, at least 1.17 volts was required for 3ghz, pretty bad for a retail chip... I just think we need better cooling for these chips than a high end watercooling kit to get them clocking high and keeping them a long time...

45nm is fast and Oc'able but doesn't seem to be good value for money in the long run... at least not with one of my QX9650... other one is from ebay and does 4ghz at 1.26volts and I never run it over that voltage... seems way better...

Also it's pretty clear that there isn't a program available that can guarantee 100% stability.... some prefer Orthos, some Prime, some OCCT,...

I do a mix of them and if all works out it's stable enough for me... surely as it does my gaming,folding and co each nite without giving any errors... and in my humble opinion too many cycles are wasted with these stability tests...

[Kondik]

I found this ... might be usefull for this thread

http://www.anandtech.com/cpuchipsets...spx?i=3251&p=6

[Leeghoofd]

Tells nothing new, just using expensive, technological typical anandtech high tech talk and graphs ... we all know the risks of overclocking though it seems that with the 45nm and surely with the quads it seems to be happening faster than we expected...and not even corresponding to Anand's graph... I don't think many peeps give the CPU max 1.25 volts...

These CPU's just clock higher but thx too the many borked sensors peeps have no clue what the real temps are... which is really a disadvantage big time... Yet we push them and on top of that some forget that we have to add 0.1 volts to compare the voltage settings from our beloved 65nm ones...

[jonny_ftm]

Originally Posted by CrazyNutz

Wolfdale degradation myth possibly busted?

Im sure a lot of you have had this happen, you clock your new e8X00 up and run the usual torture test
until your find the max stable OC @ some vcore, and temps you are comfortable with. That's the
Idea right. Right!!

By now your prime stable for >8 hours, and your happy because your at 4ghz (or more) and you
didn't even break the bank when you made that sweet CPU purchase. Still with me?

Ok now at some point ( Next day/Next week) you'll run that torture test again and then OMG!!!
WTF??? Prime95 FATAL ERROR!! But it used to be prime stable for hours and now it's failing in
less than 5-10 minutes? Instantly that horrifying thought comes to mind "Degradation".

Ok so that's the scenario for most. At least it was for me. Iv'e had this happen with two E8400's
so far. The first one I upped the vcore passed 1.4 and after a couple of days when prime started failing
I thought for sure this is the symptoms of the dreaded degradation problem. The second E8400 I decided
I would never go past the 1.36 safe (or safer) limit (actually it's never been passed 1.33), and I
would not see this problem again, Wrong!! I did, but this time I was determined to figure this out!


And here is what I found:

If my system is either off, or has been at Idle for a long period of time (i.e 18 - 24 hours) and I run prime
(especially the 10k FFT's) It will fail in < 5 minutes sometimes < 1 min. If I then run prime again it will take
prime a little longer to fail but maybe still < 5 mins. If I run It again it will take a lot longer to fail > 20 mins.
And then I run it once more Bam!!! Smooth sailing no more failing, It's back to the stable state it was
initially. 10k priming for hours on end.

So why?
Answer: Not sure but it seems apparent that it's needs to get nice and warm to be prime stable, and
maybe It's not warm in a literal sense, I just don't know. However It does seem to need this so called
"warm up period" more often to become stable when the ambient room temps are lower than 70F or so.

Please note these are just my findings on my rig, I would like to ask you guys to put this to the test
to see if this "Warm Up Period" holds true with others. If you would like to participate I urge you
to use Prime95 with 10k FFT to test with. I have found the 10k FFT's to put the most stress on the
wolfdales!

Thanks

For me, all your symptoms are also real degradation. It's not a car to warm up (here it's electronic, not mechanic). The temperature makes electrons move more "crazely", let's say it like this, exactly as vcore does. When you increase temperature or vcore, the electrons will dig new paths outside of the normal path in the silicone. It's a complex mechanism ending in a physical degradation of the chip circuits. Now, when you increase temperature (warm), you hopefully reproduce a closer situation of vcore increase, at the electron level. I'm sure, if you give it small vcore increase, you won't need that "warm period", for a while at least.

Increasing the frequency up to 4GHz is a major thermal producer. People do still believe that temperature is only related to vcore because of these "thermal guides for C2D/C2Q" that you all know and that were written for 65nm CPUs. If you really well test your CPU, you'll know that increasing frequency without touching the vcore or by increasing it only a marginal part, it will lead to a dramatic temperature increase. This is because of the smaller 45nm build. Making transistors run 50-100% of their rated frequency will produce a major heat. Less heat than with 65nm chips, but... at 45nm technology, that lower heat is more harmful than higher temperatures on 45nm chips

Your story is just a confirmation that these 45nm chips die easier than most think. I have no doubt, in few months, we'll see newer 45nm revisions giving what G0 gave to B3 stepping

After the death of "Power = Frequency" theory, we see again the limits of silicone. Intel won't be able to push it to 35nm and lower building process. That's why they're going multicore. This will give them few years up to some 8-16-24 multicore plateforms... But the real question is, for how long? A new technology should come soon, or a replacement to the silicone that reached really its limits

[CrazyNutz]

Originally Posted by jonny_ftm

For me, all your symptoms are also real degradation. It's not a car to warm up (here it's electronic, not mechanic). The temperature makes electrons move more "crazely", let's say it like this, exactly as vcore does. When you increase temperature or vcore, the electrons will dig new paths outside of the normal path in the silicone. It's a complex mechanism ending in a physical degradation of the chip circuits. Now, when you increase temperature (warm), you hopefully reproduce a closer situation of vcore increase, at the electron level. I'm sure, if you give it small vcore increase, you won't need that "warm period", for a while at least.

Increasing the frequency up to 4GHz is a major thermal producer. People do still believe that temperature is only related to vcore because of these "thermal guides for C2D/C2Q" that you all know and that were written for 65nm CPUs. If you really well test your CPU, you'll know that increasing frequency without touching the vcore or by increasing it only a marginal part, it will lead to a dramatic temperature increase. This is because of the smaller 45nm build. Making transistors run 50-100% of their rated frequency will produce a major heat. Less heat than with 65nm chips, but... at 45nm technology, that lower heat is more harmful than higher temperatures on 45nm chips

Your story is just a confirmation that these 45nm chips die easier than most think. I have no doubt, in few months, we'll see newer 45nm revisions giving what G0 gave to B3 stepping

After the death of "Power = Frequency" theory, we see again the limits of silicone. Intel won't be able to push it to 35nm and lower building process. That's why they're going multicore. This will give them few years up to some 8-16-24 multicore plateforms... But the real question is, for how long? A new technology should come soon, or a replacement to the silicone that reached really its limits

Interesting, based on what you're saying at some point the "warm up" to stability
for my cpu will eventually over time become more difficult at the same vcore possibly
indicating real degradation. So to test this I will keep my vcore at the 1.328v and
frequency at 3970mhz for the next week or so to see If this degradation will occur.
I will also run SuperPI, Prime, OCCT, games, and benchmarks as much as possible
to place a lot of stress on my system.

[jonny_ftm]

Originally Posted by CrazyNutz

Interesting, based on what you're saying at some point the "warm up" to stability
for my cpu will eventually over time become more difficult at the same vcore possibly
indicating real degradation. So to test this I will keep my vcore at the 1.328v and
frequency at 3970mhz for the next week or so to see If this degradation will occur.
I will also run SuperPI, Prime, OCCT, games, and benchmarks as much as possible
to place a lot of stress on my system.

Or just increase a bit your vcore and see if it solves your "warm-up" issue. I'm sure, at some vcore, the problem disappears indicating that your CPU is in auto-suicide mode
If you confirm it enough soon, you could save some overclock headroom of it. If you continue killing it, you'll loose it + its resell value
In anycase, your feedback is welcome, as it is another proof of Intel ing us without competition (no need to flame me Intel blind lovers, ok?)

[v0dka]

Originally Posted by CrazyNutz

Interesting, based on what you're saying at some point the "warm up" to stability
for my cpu will eventually over time become more difficult at the same vcore possibly
indicating real degradation. So to test this I will keep my vcore at the 1.328v and
frequency at 3970mhz for the next week or so to see If this degradation will occur.
I will also run SuperPI, Prime, OCCT, games, and benchmarks as much as possible
to place a lot of stress on my system.

Two things there:

1. Stress your CPU 100% constantly (while not losing any performance you need) by crunching WCG. Click the link in my sig.

2. A week is not a good time for testing things like this.

[jonny_ftm]

Originally Posted by v0dka

Two things there:

1. Stress your CPU 100% constantly (while not losing any performance you need) by crunching WCG. Click the link in my sig.

2. A week is not a good time for testing things like this.

With my method, he'll sort it out in few minutes
I'll take a look by the way at WCG too, looks to be interesting