E8400/8500 degradation myth possibly busted?

[RunawayPrisoner]

there should be no difference between 445 x 9 and 500 x 8.

Yes, there should, because 445 x 9 = 4005, and 500 x 8 = 4000. Believe me or not, there might even be a drastic difference in voltage requirements between 487 x 8 and 488 x 8 already... talking about that. Of course, it might not matter, and it might just have been my board.

[dnottis]

I've got a 4GB kit of OCZ 1066 DDR2 coming in for review, I'll see if I can run less vcore @ 500 than 445.

[mrcape]

It seemed to be a sweet spot for my setup and others on the DFI have had similar experiences. It's not to say that you couldn't get even better results with optimal gtl settings at 9x on your board. I just find that most people use 9x and leave gtls on auto, and when I did that I needed significantly more vcore to run stable.

Does onepagebook post on here? He's had a bunch of e8400s able to prime for long hours at ultra low vcore. I wonder if those chips have been taken up high and how they are now.

I hope you don't mind butting heads a little, it brings on more good discussion. I still think the case is not closed!

[dnottis]

I hope you don't mind butting heads a little, it brings on more good discussion. I still think the case is not closed!

Not at all, this is a discussion forum, thats what it's all about I'd like to see a paste of OPB entire bios during some of his OC's.

What are your recommendations for voltages, GTL, etc?

[Big Lar]

Not at all, this is a discussion forum, thats what it's all about I'd like to see a paste of OPB entire bios during some of his OC's.

What are your recommendations for voltages, GTL, etc?

I am Overly interested to hear of others GTL settings here, for some reason I am having decent luck with 63/63/63 so far,( 8.5 x 500 ).

Larry

[mrcape]

Okay this is for 500x8.5

I have had the most luck with VTT (fsb term) at 1.15v (set in bios) and gtls 59/59/59. From there, scaling the GTL settings up with increased NB and VTT works, so if I bump up to 1.20v vtt I use 61/61 for those gtls. If I raise NB I raise those GTLs as well, but the increment is different.

Other voltages -
Vcore - 1.325v
PLL - 1.55v
NB - 1.37v
VTT (fsb term) - 1.15v
DRAM - 2.23v
VDroop Control - Disabled

I have more I can share but I'm late for Poker. Will be back later.

[dnottis]

1000 watt PSU ftw.

[Big Lar]

Looks just like my graph, bout the same voltage too. Nice supply you have there dnottis.

Larry

[dnottis]

Looks just like my graph, bout the same voltage too. Nice supply you have there dnottis.

Larry

I'd imagine the vdroop pencil mod and having load line calibration helps to stabilize voltage too though

[rge]

Occt vcore for 10 mins, but would be same if ran it all day. All this shows is that vdroop/voltage fluctuations are less than the sensor is capable of reading...sensor reads in steps of .016v. This is with loadline disabled, GB P35 DQ6. Would need a multimeter to actually read the fluctuations, or a sensor that is more sensitive than reading rounding of .016v steps.

On another note, I had been running 4.2 at 1.39 CPUZ vcore for a while now, but had not checked on stability recently. My game froze for first time, then the desktop froze on me after that. Checked orthos and it stopped in 2 seconds, then 3 seconds, had been stable 15+hrs blend and 17+hrs 8-32 small ffts. Prime 95 did the same. Turned it off overnight, and it ran just over 2 hours then stopped, ran it again for 7 hours (and I stopped it) then ran 10ffts and it stopped at 23 mins. I backed down to 4.05, using two notches in bios higher than previous to get 12+hrs stable. These are the screwiest cpus I have ever dealt with. I am upgrading when Nehalem gets here, I just hope they improve the 45nms by then to at least not have the weird variabilty/stability/degradation whatever you want to call it issues at vcores ~1.39 and less...or maybe i need to keep it at 1.36 or less.

[RealTelstar]

The state that I'm at right now is giving me very reproducible results:

1. I let the machine idle for 18-24 hours.
2. I then run prime95 10k FFT. Error in < 2-5 minutes
3. Continue restarting prime95 10k until stable at > 2 hours.

This thread reminds me of the old "burn-in theory" that has been recently abandoned and may do a comeback with the 45ns intels.

[dnottis]

This thread reminds me of the old "burn-in theory" that has been recently abandoned and may do a comeback with the 45ns intels.

I thought those days were done, ugh.

[mrcape]

That's kinda what it seems happened with mine. It seems to have been broken in, and since then it hasn't got any worse.

[CrazyNutz]

That's kinda what it seems happened with mine. It seems to have been broken in, and since then it hasn't got any worse.

Mine has not (yet) gotten any worse either. It's does at this point appear to be a burn-in phenomenon, however in the past burn-in was supposed to make
the cpu more stable at lower vcore. If the burn-in is what is causing our
45nm's to exhibit this behavior, it is the exact opposite of what we would normally expect from a burn-in. Heh, lets just hope It does not go from
burn-in, to burned-out!

I think this burn-in theory may be a good direction to go in, however lets
not "nail the coffin", or "close the case" until we know for sure.

[nullface]

So to sum up, there IS a degradation problem, maby not for all but some ppl are experiencing degradation, right?

Should rev.C1 fix this?

[mrcape]

Supposedly C1 (on the quads at least) addresses some issues with the GTL reference balance. Not sure of any details or performance effect but I'm sure they will be common knowledge soon. If you read anything post it up, I'm super curious about c1 too, especially because of the huge list of errata for C0.

[CrazyNutz]

So to sum up, there IS a degradation problem, maby not for all but some ppl are experiencing degradation, right?

Should rev.C1 fix this?

From what we can tell at this point, the 45mn's seem to degrade to a point.

For example, my cpu was initially 8hours prime stable @ 3960mhz @ 1.328
vcore,and within a week it started gradually failing to the point that it would
fail prime in <1min. Now for the same 3960mhz it needs 1.352vcore to be 8
hours prime stable.

So what is unknown at this point is will our CPU's degrade further, or
is this a negated burn-in effect, that after a burn-in period the cpu
just needs a little more vcore to be stable again,
and the degradation WILL NOT continue to worsen.

Both of my e8400's have exhibited this behavior, neither one has degraded
any further yet. If you read more post you'll see mrcape also claiming no
further degradation.

So keep checking the thread, if anyone has any further degradation we'll
post it here.

[CrazyNutz]

Supposedly C1 (on the quads at least) addresses some issues with the GTL reference balance. Not sure of any details or performance effect but I'm sure they will be common knowledge soon. If you read anything post it up, I'm super curious about c1 too, especially because of the huge list of errata for C0.

BTW: Any further degradation on your cpu's yet?

My 748 is banging out 10k fft's like a champ, no errors @ 3960mhz & 1.352vcore.

[mrcape]

None. Running and working 24/7 4.25ghz with weekly OCCT or Orthos (10k fft too) re-checks. Seems super solid where it's at.

[dnottis]

I went back to the quad. Gaming is smoother for me.

[deathman20]

Well just saw this thought I'd drop my 2 cents in here.

I ran 1.37v @ 4.05Ghz right when I got the CPU. It was nice and fast no doubt. I know I tried higher at 4.2Ghz with 1.45V but was unsuccessful for benching purposes right away. Now just a few weeks ago I tried higher clock at 4.32Ghz w/ 1.43 or 1.45V can't remember and was able to run MANY benches at that speed. I haven't tried to drop the voltage on my 4.05Ghz clock to see if I gained something or not but I'd say I haven't had an issue with it.

[slugzkea]

After supplying 1.66v to my e8400 to stabilize 4.43ghz-4.50ghz, I decided to go Q6600.

The cpu seems stable for gaming but I still can't do prime or orthos for the life of me. So screw that.

If it can't last 30 mins orthos or prime, it's a sign of true instability.

Bottom line is, the wolfdales are not stable past 4.2-4.3ghz unless you apply a LOAD of voltage. And for what, a few extra hundred points?

I'll take an extra thousand points instead for half the hassle.

Thanks

[MosIncredible]

I'm still having the same situation I stated earlier in the thread. My 2nd core (never the 1st) fails at 1hr 18-22 mins every single time I prime. Same clock and voltage ran for 15hrs+ originally. One voltage notch up and it primes for 17+ hrs. What would cause it to fail at the exact same spot every time? Seems odd that only one voltage notch up would cause it to go for an extra 16+ hrs.

[mrcape]

Wait, I don't get it slugzkea, you're not happy with 4.3ghz? I'm not being a jerk I just don't get what you're saying.

[mrcape]

I'm still having the same situation as I stated earlier in the thread. My 2nd core fails at 1hr 18-22 mins every single time I prime. Same clock and voltage ran for 15hrs+ originally. One voltage notch up and it primes for 17+ hrs. What would cause it to fail the exact same spot every time? Seems odd that only one voltage notch up would cause it to do for an extra 16+ hrs.

It's normal for it to fail in the same part of the test. At least it's consistent. just give it a bump or two and call it stable. Or try a different stability test.