Are Prime 95 blend results always reliable for quad cores?

[Renegade227]

Hi I'm a bit new to Prime95 testing and had some questions:

I have version 25.5 of Prime95 (I run blend) and was wondering if it's normal for a stable system to crash on one of the cores every couple of trials anytime after 10 min to 3 hours of testing? For a quad core, are Prime 95 Blend crashes 100% telling of an unstable system or are they sometimes just a fluke?

Also, does simple web browsing negatively affect prime95 results?

 

[Idontcare]

Originally posted by: Renegade227
Hi I'm a bit new to Prime95 testing and had some questions:

I have version 25.5 of Prime95 (I run blend) and was wondering if it's normal for a stable system to crash on one of the cores every couple of trials anytime after 10 min to 3 hours of testing? For a quad core, are Prime 95 Blend crashes 100% telling of an unstable system or are they sometimes just a fluke?

Also, does simple web browsing negatively affect prime95 results?

Any crash of prime95 ought to be viewed as being induced by system-instability unless you can prove to yourself otherwise.

Blend is NOT a good test to run if you want to know which part of your system is the likely cause of the instability.

Run small-FFT for testing CPU and on-die cache.

Run large-FFT for testing ram, FSB/HT/QPI, NB chipset stuff etc.

You should be able to consistently pass both small and large FFT tests for more than 12hrs each. If you can't then that is usually an indication that something is too hot or under-volted or both.

In some instances being too hot can actually be caused by being over-volted, so the solution to an instability can sometimes entail reducing your voltages on a given component. (I over-volted my ram too much once and discovered this the hard way)

 

[Elganja]

LinX and OCCT are good at testing stability as well

 

[VirtualLarry]

The newest version of prime is 25.9, there were some bugs in older versions. Then again, I've used 25.6 for testing many a time, and never encountered software bugs. If prime errors, your system is likely to be unstable.

 

[BonzaiDuck]

I'm often mildly impressed with Russian programmers. This "Stasio" fella has seemingly done a great job in taking the old LinPack IntelBurnTest and putting it into a very clean and informative windows interface.

Generally -- nobody has challenged my speculations much here -- it would seem that errors under stress-testing programs are Poisson-distributed. Chucking quantitative notions for a moment as opposed to relative ones -- if your system passes PRIME95 10 minutes without error, the probability of its going to 20 minutes is higher. If it passes one hour without error, the probability of its going 2 hours is higher. This logic applies for longer and longer time periods.

I have seen PRIME95 fail after 9 hours, thus requiring a very slight increase in VCORE voltage or RAM (voltage/latencies) to eliminate the instability that caused it. So most people who use PRIME95 exclusively advise stability testing for 12 hours or longer.

IntelBurnTest -- or Linpack or LinX -- will find instability in a much shorter period of time. It will also stress your processor more than PRIME95.

In the process of over-clocking, it is useful to find the settings that will fail within 10 or 20 minutes. This suggests some small adjustments, and noticeable increases in time-without-error after such adjustments are telltale.

Once I've found stability that lasts 1 or 2 hours, I test for 4 to 6 hours. If that results in 0 errors, 0 warnings before I terminate the test, I'll submit it to IntelBurnTest for 10 iterations.

Once I get IBT (LinPack or Linx) to go 10 iterations without fail, I'll run it for 20 to 25 iterations, and if it passes, I give the system a pass for rock-solid stability. 5 iterations of IntelBurnTest takes about 36 minutes.

 

[Dadofamunky]

Also, I believe Prime has a native 64-bit version. Correct me if I'm wrong, though. too lazy to check right now (expend too many ergs running the search). Definitely get the newest one.

I've had Prime small-ffts fail after nine hours recently! Not anymore, however. And nothing I do will EVER approach Linpack levels. I have sailed through 10 iterations lately, however. I don't think Linoack is all that realistic for real-world indications. Just my personal view.

It's great that those guys are continuing to maintain Prime. I like to use OCCT and Prime as my primary stability tests, and I just put my Penryn box in complete lockdown after the last round of successful testing. No more BIOS changes for this puppy. It's running perfectly now, a whole year after I started the build.

 

[n7]

Small FFTs = CPU test; useless for checking stability of anything else.

Large FFTs = board test, particularly when you are using a high FSB & putting lots of strain on the NB.
It's very good for testing VTT/NB + GTL stability.

Blend = CPU + overall system, though i don't trust it for large amounts of RAM as it will miss things.

LinX/IBT/LinPack = good for extreme CPU testing (use max RAM; sets larger calculations which = more stress = more heat) & RAM testing.

HCI Memtest = great for RAM testing; will also show NB-related errors if that's not stable.

Memtest86+ = okay for basic RAM testing; useless for stressing NB though.

Play some games as well...i've had P95 pass & LinX pass yet still have had board related instability.

Renegade227, if a core if failing, you aren't stable.
9x445 is still a lot of stress for a board handling a quad core.
You likely need to work on your FSB Termination/NB Voltage + GTLs, & possibly vCore.

CPU is easy to figure out.
Drop the multi to 7x, & run 7x444.
If that's not stable (run Large FFTs + LinX), you've got some board stability to work on, or possibly RAM.

 

[Elganja]

Originally posted by: Dadofamunky
Also, I believe Prime has a native 64-bit version. Correct me if I'm wrong, though. too lazy to check right now (expend too many ergs running the search). Definitely get the newest one.

I've had Prime small-ffts fail after nine hours recently! Not anymore, however. And nothing I do will EVER approach Linpack levels. I have sailed through 10 iterations lately, however. I don't think Linoack is all that realistic for real-world indications. Just my personal view.

It's great that those guys are continuing to maintain Prime. I like to use OCCT and Prime as my primary stability tests, and I just put my Penryn box in complete lockdown after the last round of successful testing. No more BIOS changes for this puppy. It's running perfectly now, a whole year after I started the build.

And what do you do that will approach the levels of Prime95?

The whole point of LinX, Prime95, and OCCT is to test stability under any circumstances. You should not discount one if your PC fails and just say "well i will never stress it that hard, so who cares about the results".


My stability test consists of the following (may be a little overkill, but i wanted to make sure 100%):
-Prime95 small fft's - 18 hours
-Prime95 large fft's - 18 hours
-Prime95 blend - 18 hours
-OCCT CPU 1 hour test
-OCCT Memory 1 hour test
-OCCT GPU 1 hour test (just stresses the hardware in a different way)
-LinX small memory test, 50 passes
-LinX max memory test, 50 passes

Yes it takes time, but patients is needed when over clocking. Plus I can say my system is 100% rockstable

 

[SunnyD]

Originally posted by: Elganja

Originally posted by: Dadofamunky
Also, I believe Prime has a native 64-bit version. Correct me if I'm wrong, though. too lazy to check right now (expend too many ergs running the search). Definitely get the newest one.

I've had Prime small-ffts fail after nine hours recently! Not anymore, however. And nothing I do will EVER approach Linpack levels. I have sailed through 10 iterations lately, however. I don't think Linoack is all that realistic for real-world indications. Just my personal view.

It's great that those guys are continuing to maintain Prime. I like to use OCCT and Prime as my primary stability tests, and I just put my Penryn box in complete lockdown after the last round of successful testing. No more BIOS changes for this puppy. It's running perfectly now, a whole year after I started the build.

And what do you do that will approach the levels of Prime95?

The whole point of LinX, Prime95, and OCCT is to test stability under any circumstances. You should not discount one if your PC fails and just say "well i will never stress it that hard, so who cares about the results".


My stability test consists of the following (may be a little overkill, but i wanted to make sure 100%):
-Prime95 small fft's - 18 hours
-Prime95 large fft's - 18 hours
-Prime95 blend - 18 hours
-OCCT CPU 1 hour test
-OCCT Memory 1 hour test
-OCCT GPU 1 hour test (just stresses the hardware in a different way)
-LinX small memory test, 50 passes
-LinX max memory test, 50 passes

Yes it takes time, but patients is needed when over clocking. Plus I can say my system is 100% rockstable

But what IS that level? What happens if Prime95 would hypothetically fail on you after 18 hours and 1 minute? You would never know? What happens if OCCT would fail on you after 2 hours? What happens if your tests would fail on you after 50 hours? Therein lies the issue - stability is relative. Sure after 18 hours you have a good indicator of stability, but likely only marginally better than 2 hours.

 

[Elganja]

Originally posted by: SunnyD

Originally posted by: Elganja

Originally posted by: Dadofamunky
Also, I believe Prime has a native 64-bit version. Correct me if I'm wrong, though. too lazy to check right now (expend too many ergs running the search). Definitely get the newest one.

I've had Prime small-ffts fail after nine hours recently! Not anymore, however. And nothing I do will EVER approach Linpack levels. I have sailed through 10 iterations lately, however. I don't think Linoack is all that realistic for real-world indications. Just my personal view.

It's great that those guys are continuing to maintain Prime. I like to use OCCT and Prime as my primary stability tests, and I just put my Penryn box in complete lockdown after the last round of successful testing. No more BIOS changes for this puppy. It's running perfectly now, a whole year after I started the build.

And what do you do that will approach the levels of Prime95?

The whole point of LinX, Prime95, and OCCT is to test stability under any circumstances. You should not discount one if your PC fails and just say "well i will never stress it that hard, so who cares about the results".


My stability test consists of the following (may be a little overkill, but i wanted to make sure 100%):
-Prime95 small fft's - 18 hours
-Prime95 large fft's - 18 hours
-Prime95 blend - 18 hours
-OCCT CPU 1 hour test
-OCCT Memory 1 hour test
-OCCT GPU 1 hour test (just stresses the hardware in a different way)
-LinX small memory test, 50 passes
-LinX max memory test, 50 passes

Yes it takes time, but patients is needed when over clocking. Plus I can say my system is 100% rockstable

But what IS that level? What happens if Prime95 would hypothetically fail on you after 18 hours and 1 minute? You would never know? What happens if OCCT would fail on you after 2 hours? What happens if your tests would fail on you after 50 hours? Therein lies the issue - stability is relative. Sure after 18 hours you have a good indicator of stability, but likely only marginally better than 2 hours.

I understand what you are saying, but there has been documented proof of prime failing after 9 hours, linx failing after 20 passes (occt is a variation of linpack IIRC), etc... I have never heard 1 case of someone's stability failing after 18 hours or ~50 passes of linx. If there were I would double those numbers.

 

[dallev79]

Actually I got a Prime error on core2 on my AMD 720BE after 21h 48min when running the blend test. I was running everything on default and the RAM@1066 with EPP enabled (5.5.5.15 2T 2,1v). When running the RAM with EPP disabled and @DDR800 everything works fine.

I have the Prime results if u want to take a look at it.

 

[Triton67]

IBT (Linpack, 2GB/High) would go on for 50 passes, P95 Blend would freeze at 1024K (less than 15mins) with same config.

I just have unstable settings and trying to find a mix of atleast 20 runs IBT and couple of hours P95 Blend...if games don't crash after that, I'm satisfied.

 

[Idontcare]

Originally posted by: dallev79
Actually I got a Prime error on core2 on my AMD 720BE after 21h 48min when running the blend test. I was running everything on default and the RAM@1066 with EPP enabled (5.5.5.15 2T 2,1v). When running the RAM with EPP disabled and @DDR800 everything works fine.

I have the Prime results if u want to take a look at it.

Welcome to the forums! :thumbsup:

Failing prime after nearly 20hrs suggests the system's cooling abilities as a whole (the case airflow and PSU) is inadequate relative to the power dissipation your HSF is attempting to dump into its local environment (inside the case, and eventually inside the room that your computer sits in).

If you had a CPU temperature trace for the 22hr run (use coretemp and capture the output to a file in 60s intervals) I bet you your temps ever so slowly kept creeping upwards hour after hour until it finally got to the point where the signal/noise ratio on your xtors fell below the critical threshold and prime started detecting computational errors.

 

[dallev79]

Maybe so, but I doubt it. Cuz I was checking in on it every once in a while. I´ve also run other tests with OCCT and LinX (Linpack), and remember my sysytem isn´t overclocked. Anyhow my CPU temps never exceeded 34 celsius.

I have a Antec 300 with two 120mm fans. Schyte Ninja2 cooling my CPU, only one harddrive. A Corsair 520HX PSU, with fan monitoring and the fan on it never goes over 50%.

 

[Idontcare]

Originally posted by: dallev79
Maybe so, but I doubt it. Cuz I was checking in on it every once in a while. I´ve also run other tests with OCCT and LinX (Linpack), and remember my sysytem isn´t overclocked. Anyhow my CPU temps never exceeded 34 celsius.

I have a Antec 300 with two 120mm fans. Schyte Ninja2 cooling my CPU, only one harddrive. A Corsair 520HX PSU, with fan monitoring and the fan on it never goes over 50%.

If your system is not overclocked then you should be able to run prime95 for months without issue...something is wrong with your rig - be it PSU, ram, cpu, etc - so your observations regarding 9hr vs. 20hr stability aren't exactly apple-to-apple comparison with what the other folks are discussing in this thread (wherein they have hardware that is otherwise known to be stable in default config prior to OC)...unless I am missing the point of your post?

 

[dallev79]

Something is wrong, yes. But my point is that one should run the stability tests as long as possible. I could have settled at 8h, 12h or 20h, thinking my system was stable. But just around the corner there might be an issue, as it did this time. I think people in general that overclock and tinker with hardware, don´t do thourugh stability testing. 12h or 18h of Prime sometimes just isn´t enough. Or is Prime wakie?

Come to think of an old rig I had. I OC:d the living crap out of it and when stability testing it just didn´t like Prime, at any setting. At that time Prime didn´t have the multicore support. So I had to direct two separate worker windows to each core, which really doesn´t matter. But when using SP2004/Orthos, that is a version of Prime, everything worked just fine. And I never had any issues with that rig. In that case Prime wasn´t reliable. Why? Don´t know. And my system was stable IMO.

As a sidenote, is there any differance between OCCT and Prime?

 

[lopri]

Originally posted by: n7
Small FFTs = CPU test; useless for checking stability of anything else.

Large FFTs = board test, particularly when you are using a high FSB & putting lots of strain on the NB.
It's very good for testing VTT/NB + GTL stability.

Blend = CPU + overall system, though i don't trust it for large amounts of RAM as it will miss things.

LinX/IBT/LinPack = good for extreme CPU testing (use max RAM; sets larger calculations which = more stress = more heat) & RAM testing.

HCI Memtest = great for RAM testing; will also show NB-related errors if that's not stable.

Memtest86+ = okay for basic RAM testing; useless for stressing NB though.

Play some games as well...i've had P95 pass & LinX pass yet still have had board related instability.

Exactly the way I test my systems and I think n7 is spot on here.

I mix a few quick tests before above mentioned tests. For example. 3DMark05's CPU test seems to be pretty good at catching instability of dual-core CPUs. (not quad-cores) For AMD quad-core CPUs, Cinebench 10 (multi-threaded rendering) is great. A full loop of PCMark Vantage is also pretty decent to test the whole system stability. (though it costs $$) Multiple instances of Super Pi 32M with and without affinity set per core catche some odd errors as well. (maybe it's got to do with cache coherency?)

Prime Blend isn't a bad test but is not the most intense test. My favorite has always been LargeFFT. But to be frank Blend is usually 'good enough' for most usage scenarios. (including mine)

 

[OILFIELDTRASH]

I have a core i7 920 oc'ed to 3.4 on stock voltage. I ran prime95 8 threads for 9 hrs 20 min no problems and did a 75 pass test of IBT no problems. Is that enough stress testing to call my rig stable?

 

[ShawnD1]

Originally posted by: dallev79
Actually I got a Prime error on core2 on my AMD 720BE after 21h 48min when running the blend test.

I hate when this happens. My phenom 9600 was failing the large FFT test after 5 hours, and it did this 3 or 4 times after making little voltage changes each time. It shouldn't take a whole day to overclock something! Grrr!!!

Failing prime after nearly 20hrs suggests the system's cooling abilities as a whole

Good to know. The phenom was running at 80C for a few hours without failure so I was starting to think maybe it just works at that temperature. My graphics card can run at 90C without a problem, so why not. I eventually did turn it down quite a bit and it did become stable.


While I know Linpack is supposed to be the most hard core testing, I still don't trust it. It only ever seems to use 25% (1 core), then it shoots up to 100% for maybe 3 seconds, then back to 25%. What the hell is wrong with it? Why can't it do the most basic thing that any stability program can do? If I run OCCT or Prime95, it immediately uses all 4 cores. Linpack rarely does.

 

[betasub]

Originally posted by: OILFIELDTRASH
I have a core i7 920 oc'ed to 3.4 on stock voltage. I ran prime95 8 threads for 9 hrs 20 min no problems and did a 75 pass test of IBT no problems. Is that enough stress testing to call my rig stable?

As explained earlier in the thread, you can call it stable because it hasn't failed a test - yet. How long to test is down to personal preferance.

Better to say it isn't unstable

 

[OILFIELDTRASH]

good advice. Until it gives me problems Ill call it stable.

 

[ShawnD1]

Originally posted by: OILFIELDTRASH
good advice. Until it gives me problems Ill call it stable.

Keep testing your system every once in a while. If your computer is balls to the wall overclocked and heat is your only limiting factor, a couple months of dust buildup can be enough to cause data corruption. The corrupt might not be bad enough to cause crashing, but it's enough to damage files. I also heard that processors can degrade in a relatively short time if they're too hot or the voltage is too high or something like that, so the system's stability can change even if the CPU is running at the same temperature and voltage as it was when you first tested it.