Sandy Bridge maximum vcore for 24/7 use

[Movieman]

Lee,
You were right dead on back during the core 2 days, when you talked about degradation. The Yorkies would experience a little bit of degradation even at lower voltages (like 1.36), sometimes as much as 0.04, and then would stabilize. Which is exactly what you said.
But repeated abuse and linpack and small FFT stress testing only helped make them degrade more, and faster...

I feel that all this 12 hour linpack/prime testing at high voltages does nothing but help degrade the chips...they're going through abuse they really werent meant to withstand ...

I sometimes wonder where common sense has gone.
I would like all of you guys to stop and just think about what your subjecting your chips to when you run these linpack and prime 95 apps for hours on end.
Now you guys know I do DC work and it loads the cpu's to an indicated 100% load and my machines run at 100% load 24/7 for years on end.
All on good air cooling and the maximum temp I will allow a machine to run at is 65C.
The difference in loaded temps between the DC apps and these bench apps is HUGE. These apps are not real world and you are subjecting them to stress that they would never normally see so why do this?
You pay hunndreds of dollars for your cpu and then set out to prove it will do X with these apps and then a few months later wonder why your cpu is degraded?
Run SP32m, in most cases if it will run that then it is stable for almost anything you will use it for in daily life.
End of rant.

[ugotd8]

+1, well said movieman.

I used 5-10 minutes of prime blend custom to get the general ballpark. Then I use SPECapc for Lightwave and if I can get thru that, I'm stable. Killing my chip trying to join an e-peen club is not my idea of fun.

[koc]

my chip like 1.42v for 4.8Ghz and 1.48v for 5Ghz , anything under my system crash , but i saw some nice chip that need only 1.27 for 4.7ghz

I think 1.5v is the max safety Vcore

[Sam_oslo]

I sometimes wonder where common sense has gone.
I would like all of you guys to stop and just think about what your subjecting your chips to when you run these linpack and prime 95 apps for hours on end.
Now you guys know I do DC work and it loads the cpu's to an indicated 100% load and my machines run at 100% load 24/7 for years on end.
All on good air cooling and the maximum temp I will allow a machine to run at is 65C.
The difference in loaded temps between the DC apps and these bench apps is HUGE. These apps are not real world and you are subjecting them to stress that they would never normally see so why do this?
You pay hunndreds of dollars for your cpu and then set out to prove it will do X with these apps and then a few months later wonder why your cpu is degraded?
Run SP32m, in most cases if it will run that then it is stable for almost anything you will use it for in daily life.
End of rant.

You see, some of us needs to push these CPUs to the limits and distinguish it from the rest. Now it's not possible to put these CPUs under LN2 and get a 7GHz+ SuperPi1M-shot. Everybody is going to get stuck around 5.7Ghz +/- anyways.

This seams to be creating some kind of fenestrations, and then some keep LinXing and Priming at the limit of these CPUs to distinguish it from the rest, but you are right (gain) that's not a good idea at all.

[Movieman]

Listen for a minute.
We're all different with different wants and uses.
All I want to get across is use some good old fashioned common sense when you test your chips as the thing I hate to see most is some 17 year old kid reading these threads then going out and blowing up parts that he can't afford to replace.
I run at 4634. Can I get more? Yes.
Ran 32m on this at 5088MHz on air and I mean real 70F air not minus 15F on my back porch.
BUT to me what I want is a solid running system at a speed that my mind says is a good long term setting with temps I'm comfortable with and I think I have that.

[ugotd8]

In my testing on the SPECapc lightwave render tests my chip can do:

451 seconds @ 4.8GHz max temp 71c
431 seconds @ 5.0GHz max temp 76c

5 percent gain for 5c ? I don't think it's worth it. 71c should get better this weekend tho, going under water.

This magical 5GHz number is making usually rational people insane.

[Movieman]

In my testing on the SPECapc lightwave render tests my chip can do:

451 seconds @ 4.8GHz max temp 71c
431 seconds @ 5.0GHz max temp 76c

5 percent gain for 5c ? I don't think it's worth it. 71c should get better this weekend tho, going under water.

This magical 5GHz number is making usually rational people insane.

Yes, and it's not ending there.
The guys know that in theory the 2600K's have a max 57 multi and they want that speed so you will see people killing these to try to get there.
Then who will be the first to get to 6GHz..
That's 57x105.26..
See what I mean?
It never ends..

[zalbard]

Some kind of stability testing is needed for a 24/7 machine. It's not going anywhere.
The amount and methods are subjective, however. Nobody is forcing anyone to run Prime or LinX for days. Everyone has their own measure of stability, it's them who is picking methods, and it's them who is responsible for the consequences.

[Eldonko]

Absolutely. FAH is still going to stress all cores 24/7 but it is going to be a hell of a lot cooler than running stress tests that nothing comes close to replicating in the real world. Plus there is a reason to do it.

Would Folding@Home 24/7 be considered a form of "stress testing"? I find it pointless to overclock a CPU to 4.8+ only to let it sit idle 99% of the time.

[sin0822]

i gave up doing long term stress testing on these chips because i did see voltage degradation at an early stage at 5.2ghz and 1.5v then upto 1.55v, but with LLC level2 so 1.53v to 1.58v, its pretty terrible, but that is what happened over time, now my chip gets to hot to OC on air, and on water or cold air the CPu won't do 5.2ghz, because it has a 20c cold bug, took me weeks to figure it out, but it drove me nuts.

Hondacity 2 chips over 2 months at least. Both D1 stepping and i did run into voltage degradation as well as much different OCes, and these chips were heavily used before i got them.

[Eldonko]

Do you suspect that it could be related to D1 and perhaps D2 is better?

[rge]

I will always make sure I am prime or linx stable with my 24/7 settings, as I dont like BSODs/corrupt OSes, but again stability testing is one of those to each his own. Also, some run prime 24/7 to search for prime numbers, just like others fold 24/7. And, I can do 32m spi at 4.9 ghz on my i950, but it is not close to stable.

That being said, I do agree that many of us run prime/linx at almost benching mhz/vcore and that may contribute to or cause degradation, but just like benching at crazy vcore, the user just needs to be aware and accept the consequences of having fun. And for some of us, myself included, it is fun seeing what you can get stable, just like it fun seeing what you can get max spi 1m, sp32m, SS, etc.

[flesheatinvirus]

wow "some" of you guys are just blurting out numbers...

sheeesh

an idiot/kid/idiot(again) would just stick to 1.45v...then prime95 for 1-4 weeks..and confirm your fhucking recommendation and kill their chip.....or it will survive...or it will degrade.

Yup, an awful lot of that going on here.

[deaffob]

why are so many people biyatching about the fact that because they don't have access to LN2, they can't reach higer? Asus already stated that with SBs, it is actually worse to go subzero. If you can't go higer with decent temp, you won't get anymore from LN2. So stop biyatching.

[Sam_oslo]

Yup, an awful lot of that going on here.

Hey you guys take it easy. No reason to use offensive language. People have different preferences and different ideas about how they want to test their setup. Addressing them as "idiot/kid/idiot" is idiotic.

[flesheatinvirus]

I always take it easy. Thank You for caring though.

[roadie]

Personally it will be 1.3V or over 65 degrees Linx/Prime95 load, which ever comes first.

[Falkentyne]

But people need to be responsible for what they do, instead of saying stuff (like was posted on hardforums) that you HAVE to prime to make sure your cpu is stable. I had to troll away in that thread, PROVING that you do NOT need to small FFT, showing that my Rybka Chess engine is quite enough for me.

Now, Prime Blend *IS* a useful test. The temps don't get anywhere NEAR the small fft test, and are about a couple of C higher than Rybka, and is a great test to show RAM stability as well. If you can NOT do prime blend, then you are NOT stable at all, and you also risk OS corruption. (I had prime blend give me an error on ALL cores after one test (didn't even get to one pass), when overclocking my RAM (might have needed 1.65v) and I bluescreened right when a round of the game Crossfire ended (yes, the game). At the very least, maybe people can back off the small FFT abuse a bit and try blend to see if their systems will pass that. If they will, then try some gaming or folding and see how it holds up. You don't need to take your system to the edge to show you're stable.

And if your RAM isn't stable, Prime Blend will usually CRASH within the first few minutes. So if you can pass even ONE full pass of blend, your OS won't get wiped due to RAM errors. No need to let it loop overnight...

When I was having problems at *STOCK* on my other system (freezes in games even at stock), I ran prime blend and was getting crashes in about 30 seconds. Small FFT ran forever...(a reseat of the 12v 4 pin fixed that).

About LinX:
Lin X is NOT even supposed to be used as a stability test. It's an internal Intel app designed for binning CPU's and QA testing them. When Lin X came out, people praised it for giving you the option of running it for 30 minutes to test stability, instead of having to run Prime for 12 hours to do the same thing, so they would put less stress on the chips. But the temps got oh-so-much hotter than small fft's. Now, people are running LinX for hours...

[flesheatinvirus]

Absolutely correct and well said IMO.

[psyside]

But people need to be responsible for what they do, instead of saying stuff (like was posted on hardforums) that you HAVE to prime to make sure your cpu is stable. I had to troll away in that thread, PROVING that you do NOT need to small FFT, showing that my Rybka Chess engine is quite enough for me.

Now, Prime Blend *IS* a useful test. The temps don't get anywhere NEAR the small fft test, and are about a couple of C higher than Rybka, and is a great test to show RAM stability as well. If you can NOT do prime blend, then you are NOT stable at all, and you also risk OS corruption. (I had prime blend give me an error on ALL cores after one test (didn't even get to one pass), when overclocking my RAM (might have needed 1.65v) and I bluescreened right when a round of the game Crossfire ended (yes, the game). At the very least, maybe people can back off the small FFT abuse a bit and try blend to see if their systems will pass that. If they will, then try some gaming or folding and see how it holds up. You don't need to take your system to the edge to show you're stable.

And if your RAM isn't stable, Prime Blend will usually CRASH within the first few minutes. So if you can pass even ONE full pass of blend, your OS won't get wiped due to RAM errors. No need to let it loop overnight...

When I was having problems at *STOCK* on my other system (freezes in games even at stock), I ran prime blend and was getting crashes in about 30 seconds. Small FFT ran forever...(a reseat of the 12v 4 pin fixed that).

About LinX:
Lin X is NOT even supposed to be used as a stability test. It's an internal Intel app designed for binning CPU's and QA testing them. When Lin X came out, people praised it for giving you the option of running it for 30 minutes to test stability, instead of having to run Prime for 12 hours to do the same thing, so they would put less stress on the chips. But the temps got oh-so-much hotter than small fft's. Now, people are running LinX for hours...

Sorry to bump this old thread guys, but this is just great post