kitfit1

direraptor22

I think we should increase the LinX runs requirement... I've learned that 20 passes doesn't mean you're completely stable. In my case I ended up having to lower QPI PLL by a huge amount, lower VTT by a bit, and up VCore a few bumps to be completely stable. My previous 20 passes at those settings crashed around 30-40 passes in LinX and failed P95 in just under 4 hours.

But then again, only a suggestion. Maybe it's only the X58 stability club and not the X58 super mega 200% on steroids stability club

It really wouldn't make any difference if runs were increased.
Linx, despite what a very large amount of people believe, is no great tester of system stability and never has been.
All it can and does test is the ability of the cpu to come up with the same result time after time.
As you have found out, Prime Blend is a much better all round test. Simply because, if you have any ram instability it will find it at some point. Most fails in Prime Blend, especially after 2 to 3 hours can normally be laid at the door of ram instability or IMC subsystem instability.
As you can see from the pic below, my 5ghz run failed Prime blend 21 hours in, but passed 20 rounds of Linx.

 
This is good info but at the same time I can't help to wonder what would have happened had you tried to run LinX for 21hrs. I am pretty sure you would have failed too and probably much sooner. The instability in the system can have several different sources. LinX has the problem of allocating memory and operate within that allocation. Only the allocated memory gets testes. It means that if you have some memory chips which are bad or could be unstable, you would miss them. In that sense you are correct. However Prime is really not stressing the CPU temperature wise. Temperature is the one of the most likely cause of failure of the CPU. When you are able to run long enough on LinX your temperature reaches a steady state. testing beyond that only raises the other possible cause of instability power fluctuations (current or voltage).
Prime95 blend I think (not 100% sure) rolls the memory allocation and allows you to test your entire memory outside of the one already allocated to the OS but it does not stress the CPU enough to bring it to its peak temperature. To me it is not really a good stress test for the CPU. If you fail after many hours you cannot be sure of what failed: Did the temperature go up at that time of the day? Did your current fluctuate? Did it hit a memory allocation which was bad? I have wondered many times in the past. I have often been Prime stable for days and failed miserably within the first loop of LinX because of temperature.
That's why to test the memory Memtest is best. It runs the entire memory through random writes and reads.
For the CPU LinX is best. Combining the two and you should have a pretty good system. Prime blend tries to do both but is not very helpful. The only remaining element of stability are occasional gross power fluctuations which can cause the system to reboot or BSOD. Not much you can do here besides raising some voltages to get out of the voltage low spikes.
Just trying to rationalize the logic of the stress test here. 
 
 

Hmm I will try it. I don't know what version I am running right now and will check when I get home. The funny thing is Prime is the recommended software to "reflow" the indigo extreme because it loads the CPU continuously and not as hard as LinX. LinX is discontinuous and causes a lot of variation. LinX would freeze my computer though when I try to reflow because it heats up too quickly. Under normal condition my current Prime95 runs my CPU 10-15C lower than LinX. We'll see.
 
Thanks for the input.

You are extreme cooling though and I am ready to bet the integrated thermocouples are not calibrated for this range.

kitfit1

If you are on air, the only way to find out then is to run both and compare. Your right though that they arn't callibrated at these temps. But even if they're not, the difference is still only 2c.

2c difference in a range of 14c is a pretty significant percent change really.

direraptor22

I'm using Prime95 version 26.6 Build 3, just came out this month.

CPU temps still look lower than what LinX was giving me, instead of 91 max I saw around 87-88 after 2 hours.

Let it run for 24hrs and see what the temps are at that time.

What's funny is I can run LinX all day long and Fail UT3.
 
UT3 for stability testing!

So I finally got around to doing this, even though I've been folding at 4.2 for a month now.  In any event, when I ran LinX the first time I was at 73 GFlops, but then the power savings kicked in and crashed the test.  I turned off the feature and reran, it ran fine, but lower GFlops.  Here's the info.
 
Core: i7 970
CPU Clock: 4.213
Voltage: 1.3375
Idle: 29 30 25 29 31 29
load: 75 75 67 72 77 75
GFlops peak: 70.8548
 
Oh well, let me know what you guys think.  By the way, I used e-Leet in lieu of CPUID as I didn't reinstall because e-Leet is based on CPUID.
 
Let me know if you can't read the attachment, I've been working on screen shots and I know I'm still not doing something right.
 
 

@kitfit1, I've been on Prime95 blend for 10 hours now, the MAX temp is still 87 degrees. I doubt it's gonna get any higher. I hear prime95 completes its full round of testing in 10-12 hours anyways.
 
According to your GFlops, I'd say they aren't as high as they can be. One or more of your voltages could be off, it's what happened to me before I managed stable settings. At one point my GFlops dropped to the low 40s and yet I was running same clock and RAM speeds as now (4.2GHz).
 
I'd suggest you slowly experiment with your voltages to get the best GFlops. Aim to pass 24 hours of Prime95!

@Kitfit,
 
I just tested LinX tonight Vs. Prime 95 26.5 temperatures:
 
Ambient 20C:
Prime95 after 1 hr
60,61,55,66,66,60
LinX
70,71,66,74,76,71
I still stand by my statement that Prime does not stress enough.

made it to 4.2 booted to windows ran linx for 30 seconds seen my temp go past 90 and said hell with this lol. So i've dabbled and I don't even remember what i did.....

Thank you for the welcome.  This morning I decided I didn't like my GFlops so I turned off the computer, removed my two "storage" drives (hot swappable on the 800D) and reran the test.  HUGE difference.  Here's the new info:
 
Core: i7 Batch: 3012B129
CPU Clock: 4.213
Voltage: 1.33750
Idle: 27 29 24 26 29 23
Load: 77 77 68 72 79 79
GFlops peak: 78.2717
 
So I'm much happier with this run.  I'm guessing that there is some tweaking I can do in my voltage settings that are going to be beneficial to my OC.  Now I'm going to have work on the sigs to get them all to fit...
 

Here's mine. With ambient temps at 28C it gets a little warm!!!
 
caniba
Core i7 920   C0 stepping
CPU clocked at:  4.21
Voltage: 1.399
Idle temps: 48 46 48 45
Load temps: 96 92 93 88 
GFlop peak: 54.1652
 

Damn, 1.4VCore at idle... 96C max temps... you were literally 4 degrees from thermal throttling... DANGEROUS!!
 

Which Prime test are you guys running?  Blend or Small FFT?

Blend. I checked later after 2hrs running and my temps did go up by another 3 degrees accross the board. Still it is 7deg lower than what I get on LinX after 2min... (My room temperature did go up by 2deg in the process though)