E8400/8500 degradation myth possibly busted?

[mrcape]

no mods

I suppose it's dangerous as you go higher if you don't keep an eye on that. It's always behaved like that.

[CrazyNutz]

People, the nail is not in the coffin yet, the case is not closed! There are
still people that cannot run there initial stable clocks without
more vcore added, mrcape, and myself included, just to name a few.

I have tried other high wattage PSU's and that has not fixed anything.

One theory that got started was the voltage spike, some people have
posted graphs (myself included) showing enormous vcore spikes 1.7v and greater, and some have posted graphs that are "smooth as baby skin".
I have graphs that are smooth as baby skin too, and in fact it took
me 3 hours of running OCCT in 10 minute intervals to get the 1 graph
with spikes. Like I said earlier the sample rate of the voltage monitors
are low, (i.e. 1hz, 1 sample per second) literal hundreds of vcore spikes
could occur between samples.

So this leaves some new questions:
1. Do these 45nm cpus degrade initially to a point and then stop degrading?
2. or, are the huge vcore spikes the culprit (ultimately causing degradation)?


Also I created, and built this:



This filter should make the 12v going to the CPU ATX power very clean, and
hopefully arrest any spikes comming from the PSU.


This is what my vcore graph looks like now:


So Im hoping to not get anymore spikes, however even with the vcore this smooth I still cannot run my initial clocks without more vcore.

[ownage]

After reading this thread i think these CPU's just sucks. These chips are only on the market for a few months now, and people already have degradation problems. I did some tests with my 1,5 year old 65nm @ 1.61v and no degradation at all. The poll says 40.31% of 45nm owners have degradation problems, i wonder how many will have these problems after one year lol.

A friend of mine gave me an interesting link.
Maybe these problems have something to do with This? It looks like Intel took a different approach how to make these silicons.

[mrcape]

So this leaves some new questions:
1. Do these 45nm cpus degrade initially to a point and then stop degrading?
2. or, are the huge vcore spikes the culprit (ultimately causing degradation)?

Nice filter!

I'm on the same page as you with your deductions/outstanding questions. I'm thinking that what I'm experiencing is a combination of the two. Good news is that mine has really leveled off and isn't still going down. I'll continue to monitor my chip and report back if I find anything.

It will also be interesting to see how things shake out with the new stepping and newer boards/bioses.

Ownage, nice link. I think that the differences between these 45nm and 65nm paired with using an older approach to overclocking is the root of the issue. I still love the new chips though. I can't beat sitting pretty at 4.25ghz on air for $200.

[CrazyNutz]

Nice filter!

I'm on the same page as you with your deductions/outstanding questions. I'm thinking that what I'm experiencing is a combination of the two. Good news is that mine has really leveled off and isn't still going down. I'll continue to monitor my chip and report back if I find anything.

It will also be interesting to see how things shake out with the new stepping and newer boards/bioses.

Ownage, nice link. I think that the differences between these 45nm and 65nm paired with using an older approach to overclocking is the root of the issue. I still love the new chips though. I can't beat sitting pretty at 4.25ghz on air for $200.

Thanks.


Yeah, If these CPU's level out and don't continue to degrade I would feel
better. My cpu does seem to be leveled off also. Could be that the new
metal gates get conditioned over time, which as a side effect the performance gets dampened by a small margin (Ha another theory).

[sfdmalex]

Well what I notice too is that some people were able to initialy run their chips at much lower Vcore then I ever could.

"4.25ghz 1.28vcore, now requires 1.325vcore "

I could never cross 4.0 with 1.28 when hte chip was just out of the box.

And there is no way I can run stable 4.25 @ 1.325



I can play around some more and drop my voltages further but I could never achieve those clocks with voltages that low in the first place.

[mrcape]

Did you try to out of the box?

[CrazyNutz]

Well what I notice too is that some people were able to initialy run their chips at much lower Vcore then I ever could.

"4.25ghz 1.28vcore, now requires 1.325vcore "

I could never cross 4.0 with 1.28 when hte chip was just out of the box.

And there is no way I can run stable 4.25 @ 1.325



I can play around some more and drop my voltages further but I could never achieve those clocks with voltages that low in the first place.

There is never a guaranty that you can reach 4.0 with 1.28, you'll have to
have a gem chip to get there at that vcore, neither one of my CPU's
would ever get to 4.0 with 1.28vcore

[mrcape]

That's true. My other chip needs 1.36vcore for 4.0ghz in the same exact setup.

[dinos22]

Well what I notice too is that some people were able to initialy run their chips at much lower Vcore then I ever could.

"4.25ghz 1.28vcore, now requires 1.325vcore "

I could never cross 4.0 with 1.28 when hte chip was just out of the box.

And there is no way I can run stable 4.25 @ 1.325



I can play around some more and drop my voltages further but I could never achieve those clocks with voltages that low in the first place.

there are a few reasons there

not all chips are the same
some will run at low vcore and some won't

second thing is cooling. many ppl don't report the cooling or ambients or whatever....we had one jackass telling ppl how he has a 4.6GHz stable machine but initially forgot to mention he had his windows open with zero temps coming though the window directly hitting the PC

also the last point and probably most important one is most ppl rely on SOFTWARE readings............ppl need to pull out a DMM (digital multimeter) and check whether the readings are correct

[KTE]

So this leaves some new questions:
1. Do these 45nm cpus degrade initially to a point and then stop degrading?
2. or, are the huge vcore spikes the culprit (ultimately causing degradation)?

One of the major engineering problems with high-k CMOS was reliability and degradation after transistor stress. Notably, high current stress causing a degradation in PMOS drive current (which directly affects your clock speeds). That's why you have a low voltage/temperature threshold, smaller chip area pumping out higher heat.

Many challenges with high-k integration have included VT pinning, mobility degradation due to soft optical phonons, and poor reliability

[Issues in High-k Gate Stack Interfaces - MRS Bull., vol. 27, no. 3, pp. 212-216, 2001]
[Fermi Level Pinning at the Poly-Si/Metal Oxide Interface - Symposium VLSI Tech. Dig., pp. 9-10, 2003]
[Review on High-k Dielectrics Reliability - IEEE Trans. on Device and Materials Rel., vol. 5, no. 1, pp. 5-19, 2005]

TDDB and bias-temperature reliability of high-k + metal transistors have been a concern

[Review on High-k Dielectrics Reliability - IEEE Trans. on Device and Materials Rel., vol. 5, no. 1, pp. 5-19, 2005]

Bias temperature instability has been investigated in high-k stacks with metal gate C13397, C11049. The impact of nitrogen on NBTI in HfSiO(N)-based pMOSFETs with TaN metal gates study showed that nitrided stacks either in NH3 or by DPN exhibit much more NBT degradation as compared to O2- or N2- annealed gate stacks. The additional degradation was attributed to the generation of nitrogen related bulk traps.

[(100) and (110) Si surface orientation: impact on bias temperature instabilities in HfSiON/TiN MOSFETs - 37th IEEE Semiconductor Interface Specialists Conference - 2006. pp.14]

Bias Temperature Instabilities were also investigated on (100) and (110) surface orientations on HfSiON TiN metal gate. Charge pumping current as function of stress time showed that more interface states are generated on (110) orientation for both N and PBT stress. However, a higher interface state contribution to the total degradation is observed for NBTI stress compared to PBT stress. Therefore, NBTI depends on the surface orientation while PBTI does not due to the fact that PBT degradation is dominated by bulk electron trapping rather than interface state generation. On the other hand, PBTI degradation was found to be a more severe issue than NBTI in terms of expected lifetime.

[Impact of nitrogen incorporation in SiOx/HfSiO gate stacks on negative bias temperature instability - 44th Annual IEEE International Reliability Physics Symposium Proceedings 2006. pp.317-324]

But we need more data.

Is it only Penryn 45nm that degrades this way or do other CPUs too?
How quick is this degradation from initial overclocking?
How is the degradation characterized?
How severe is the degradation?

[yes I don't need to know, I have read the thread but in the end, that's what needs to be answered and better, a post with names/level of degradation with specific CPU all aggregated for better viewing and reading]

[dinos22]

hard to say it is degradation if one user here reported those issues and then changed the power supply and found stable settings at old clocks again provided no setting was changed to initial testing

[KTE]

True that, the testing individual would have to test it accuratey to find out - like at least have everything else working good. If they change a piece of hardware and it isn't acting degraded anmore then it obviously can't have beeen, because degradation isn't a reversible proces for us

[Warpcore]

Hi all.I don't have the 45nm chip myself.But a friend has one and is reporting the same problems as the threads here.He or I didn't have any answers.But I did suggest to him to do what I used to do if I had voltage fluctuations on my 3.3volt rail.Shut off the pc,Turn off the power,unplug the big 24 pin connector.Then plug it back in and fire up pc.Then my 3.3 v. was back to normal.Now what does that have to whats happening here?I don't know if it'll work or if will do anything at all.But there was mention of a power supply change by someone and the problem went away.So all I can say is maybe try it and see if it works.My friend is going to try and report back to me if he sees any differance.Good Luck Guys.

[dnottis]

Well what I notice too is that some people were able to initialy run their chips at much lower Vcore then I ever could.

"4.25ghz 1.28vcore, now requires 1.325vcore "

I could never cross 4.0 with 1.28 when hte chip was just out of the box.

And there is no way I can run stable 4.25 @ 1.325



I can play around some more and drop my voltages further but I could never achieve those clocks with voltages that low in the first place.

I think the mobos might be playing some strange tricks with vcore. I have my pencil vdroop modded and load line calibration on so there is no varience with my vcore. If you get to a failed OC, clearing the CMOS and restoring the profile works very well. I think some kind of hardware reset is neded to flush the mobo. Seeing these guys at <1.3 and 4.2Ghz just doesn't even seem realistic. Mine is a good clocker and I need 1.32v for 4.1Ghz and 1.37 for 4.2Ghz. Pushing 4.2 on 1.28v just doesn't sound right to me, thats not going to be enough voltage imo.

[CrazyNutz]

Hi all.I don't have the 45nm chip myself.But a friend has one and is reporting the same problems as the threads here.He or I didn't have any answers.But I did suggest to him to do what I used to do if I had voltage fluctuations on my 3.3volt rail.Shut off the pc,Turn off the power,unplug the big 24 pin connector.Then plug it back in and fire up pc.Then my 3.3 v. was back to normal.Now what does that have to whats happening here?I don't know if it'll work or if will do anything at all.But there was mention of a power supply change by someone and the problem went away.So all I can say is maybe try it and see if it works.My friend is going to try and report back to me if he sees any differance.Good Luck Guys.

FYI: The 3.3v rail does not power the CPU anymore, It's the 12v rail that powers the CPU these day, and for that matter has been for quite some time
now.

And, I was (like many) very skeptical of degradation but here are the facts:

1. I have 2 e8400's now that WILL NOT run at the initial (>8 prime) stable
clocks at the initial vcore (i.e. 3940mhz @ 1.32v). They started gradually failing within a weeks time.

2. On one e8400 the vcore was never over the 1.32vcore, when it started
degrading.

3. With the exception of my CPU's I can say with certainty that my rig is stable, and good for over 500mhz FSB, I however clock my CPUS in the lower to mid 400's.

4. No PSU's or any other hardware has made my CPU's return to the initil stable clocks @ said vcore.

5. Heat is not an issue with my rig as I have high-end water cooling, my rig is
in a wide open space, the ambient temps stay under 72, and my Core's rarely get past 55c (according to real temp).


The reason why I was skeptical is when the instability initially started, It would be unstable one day, and the next day back to normal.
However eventually over a few more weeks, It would not return to stable ever, failing prime/occt/orthos immediately.

[Warpcore]

Well yes I know the 3.3 v doesn't run the CPU.That wasn't the point.But all the voltages are off that big plug reguardles of what you run.It was a suggestion for someone to try.It can't hurt anything if other might want to try.I don't know if there's degredation or electromigration or what ever.It may be that these chips have just reached the limit of what you can do being they are so small.I would hate to see anyones CPU fail at this stage of the game.I'm sure they will figure it out.Or at the very least get it fixed with a newer steping cpu,bios update or a bigger hammer.Lets just hope the degredation at least stops after awhile.

[Brahmzy]

was going to suggest a better PSU

i don't trust any PSUs after this HX620 i have....awesome psu

those spikes are scary

Amen brutha! I LOVE my HX620 and my P5E with Vdroop mod - VERY stable on the voltages...

[mrcape]

Seeing these guys at <1.3 and 4.2Ghz just doesn't even seem realistic. Mine is a good clocker and I need 1.32v for 4.1Ghz and 1.37 for 4.2Ghz. Pushing 4.2 on 1.28v just doesn't sound right to me, thats not going to be enough voltage imo.

Have you tried 8.5x495-500 and manually setting gtls? Reason I ask is that I had a much easier time getting stability with lower vcore using those settings vs. 9x4xx.

I think my 745a is a very good chip, one of the best I've seen looking at tons of results. Onepagebook at OCX and a few others with 743a have similar or better performance out of theirs. And like I said earlier, my 748a took 1.36vcore to be stable at 4ghz using the same test method.

Re: ambient temps, mine are pretty low - mid 60s.

[dnottis]

Have you tried 8.5x495-500 and manually setting gtls? Reason I ask is that I had a much easier time getting stability with lower vcore using those settings vs. 9x4xx.

I think my 745a is a very good chip, one of the best I've seen looking at tons of results. Onepagebook at OCX and a few others with 743a have similar or better performance out of theirs.

Re: ambient temps, mine are pretty low - mid 60s.

I'm wondering if that is part of the falacy. Why should you be able to run lower volts with a higher FSB? Does that really make sense to anyone? I dunno but I wouldnt expect 500 x 8 to run stable at any less voltage than 9 x 445.

However - maybe at really low volts chip runs a stable orthos for 12 hours maybe out of pure luck or the motherboard is overvolting on that particular startup. Then it never does again.

Personally, I wouldnt expect 4Ghz from anything less than 1.3 volts though. The thought that 4Ghz at anything less just doesn't really seem feasible to me, in fact I started with 1.36v to get 4.050 stable and then moved back from there to see that 1.328v was stable. Any less and I fail Orthos after awhile though. Maybe starting low and working up is providing less than accurate results. Ambient temps will make a large difference too - if someone had their window opened to winter air or if the ambient is a few degrees different this can all affect stability...

I guess what it comes down to is expecting a 3.0Ghz chip to run 4.0Ghz at anything less than 1.3v is slightly foolish. I'm wondering if anyone with a chip that needed high volts to hit 4.0ghz experienced this degredation or not?


oh and to answer you question no I havent tried 800 x 5 because at 445 my Mushkin will run 5-4-4-12 still, at 500 I'd have to pump more volts into it and drop to 5-5-5-15.

[mrcape]

I'm wondering if that is part of the falacy.

I don't think there's any falacy here, a lot of denial though. : )

Why should you be able to run lower volts with a higher FSB? Does that really make sense to anyone? I dunno but I wouldnt expect 500 x 8 to run stable at any less voltage than 9 x 445.

It's not only lower vcore and high fsb. It's that and fine tuning GTL/FSB term.

However - maybe at really low volts chip runs a stable orthos for 12 hours maybe out of pure luck or the motherboard is overvolting on that particular startup. Then it never does again.

And maybe not, we can maybe all year. Maybe things are working as they should and the results are real. I believe the latter.

Personally, I wouldnt expect 4Ghz from anything less than 1.3 volts though. The thought that 4Ghz at anything less just doesn't really seem feasible to me, in fact I started with 1.36v to get 4.050 stable and then moved back from there to see that 1.328v was stable. Any less and I fail Orthos after awhile though. Maybe starting low and working up is providing less than accurate results. Ambient temps will make a large difference too - if someone had their window opened to winter air or if the ambient is a few degrees different this can all affect stability...

I guess what it comes down to is expecting a 3.0Ghz chip to run 4.0Ghz at anything less than 1.3v is slightly foolish. I'm wondering if anyone with a chip that needed high volts to hit 4.0ghz experienced this degredation or not?

Well you shouldn't expect anything at all until you test a particular chip. There's a big range with these. But I don't think you can talk about "chips" in general anymore, especially with the differences in the architectures of 45 and 65nm. I think a lot of people are blasting these chips, treating them as if it's another e6600. Once you do that, and assuming it dergades there's no way to see what potential it could have had.

oh and to answer you question no I havent tried 800 x 5 because at 445 my Mushkin will run 5-4-4-12 still, at 500 I'd have to pump more volts into it and drop to 5-5-5-15.

I see. My team and gskill kits are doing fine at 4-4-4-12 with 500 in all my stability tests. Definitely helps.

[mrcape]

As I said a few pages back, we need a more comprehensive data set to get any real answers, and I don't see how we can get that data without fresh chips. Opinions and speculation won't help. I've tried to submit the data I've gathered and conclusions about it.

I'm planning on getting more of these chips so if anyone wants to propose a testing methodology going foward, so that we can provide the same data, I'm all ears.

[CrazyNutz]

Perhaps we should ask FUGGER to kindly add his input on the degradation issues. He's pumping lots of voltage in these 45nm's to brake records, so
then what's the state of his 45nm cpus after that kind of abuse? Are they
degrading? if so anymore than 65nm?

[dnottis]

Perhaps we should ask FUGGER to kindly add his input on the degradation issues. He's pumping lots of voltage in these 45nm's to brake records, so
then what's the state of his 45nm cpus after that kind of abuse? Are they
degrading? if so anymore than 65nm?

not a bad idea.

[dnottis]

I don't think there's any falacy here, a lot of denial though. : )

Maybe but it's not mine, my Xeon 3110 runs the same voltage at the same OC's. I think the power issues are alot of it. I have a 1000 watt PSU and my mobo is pencil droop modded, my voltage doesn't spike or dip at all during OCCT. Not a single notch


It's not only lower vcore and high fsb. It's that and fine tuning GTL/FSB term.

Same thing, why would you need less vcore at the same speed? Once the GTL/FSB is tuned there should be no difference between 445 x 9 and 500 x 8.