Intel Burn Test has Bug

[dave_74]

Recently overclocked my rig to 4ghz, and was testing stability.
Poster on forum here recommended intel burn test.....seemed great program.

But one bug u all need to be advised of, is that any i7 920 will fail this test under certain conditions.
For demonstration, set stress level = custom = 64MB. Set threads = 8. You will auto fail the test instantly, every time.

I have system that has passed testing 5+gig sizes on 8 threads, only to fail them at other times.
Spent long time trying to figure out why. And the answer was (after scouring the internet) was that the program gives inconsistent results with certain MB + thread combinations. The reason my system would pass sometimes (and if it passed the first test, it would pass the next 20) is that the program would lock in the MB's to test based on the amount of free memory my system had (the standard test sizes take a fraction of you available memory). So if it liked the MB + thread combination of that it was initially locked at, i would continue to pass all tests.

Just thought i'd pass along the info, before anyone spends too many hours, or does crazy things with their voltages tying to pass this test. P.s. standard size + 8 threads (on my system, is always 1024 MB) and that seems to work. My system always passes the standard test, but given the flaw in either the test (or perhaps an unavoidable flaw in the processor), hard to know exactly what a fail means....

confirming posters found here....
http://www.xtremesystems.org/forums/showthread.php?t=197835&page=35

 

[B-Unit]

Or you could just run Prime95...

 

[dave_74]

Intel burn claims to find instabilities / errors in 30 mins that would take 40+ hours on prime. So the fact that i was passing prime on 1-2 hour tests seemed somehow consistent with my failing intel burn tests, and thus my desire to root out the cause as i'm running a custom app sensitive to errors.

 

[ac3144]

FWIW Thanks for going to the effort of letting everyone know, respect to you Davo!
AC

 

[htoonthura]

Recently overclocked my rig to 4ghz, and was testing stability.
Poster on forum here recommended intel burn test.....seemed great program.

But one bug u all need to be advised of, is that any i7 920 will fail this test under certain conditions.
For demonstration, set stress level = custom = 64MB. Set threads = 8. You will auto fail the test instantly, every time.

Why would you want to set memory to 64mb? It does not make sense at all. You are supposed to use maximum amount of memory.

 

[dave_74]

U set it to 64 to give the comp something it can evaluate fast, and test the algorithm for stability. If the same comp passes a standard 1024MB test 50x in a row, but fails a 64MB test every time, how are we suppose to know that a semi-random amount of memory: max-currenlty-available-memory is also not subject to the same faults?

Moreover, the way i interpret the fail at 64MB is consistent with the fact that if my comp passes the first test, it will pass every test thereafter at the same amount of memory. It suggest the algorithm is sensitive to specfici amounts of memory and threads, or alternatively i7 processors are defective is in such a way that we can't hope to pass this test at certain MB/Thread combinations (hyperthreading defect, who knows?). It'd be interesting to found out the ultimate reason why the test fails, but for now i just wante to pass along what i learned (credit to the linked posters for discovering this) to save others overclocking frustration.

 

[htoonthura]

U set it to 64 to give the comp something it can evaluate fast, and test the algorithm for stability. If the same comp passes a standard 1024MB test 50x in a row, but fails a 64MB test every time, how are we suppose to know that a semi-random amount of memory: max-currenlty-available-memory is also not subject to the same faults?

Moreover, the way i interpret the fail at 64MB is consistent with the fact that if my comp passes the first test, it will pass every test thereafter at the same amount of memory. It suggest the algorithm is sensitive to specfici amounts of memory and threads, or alternatively i7 processors are defective is in such a way that we can't hope to pass this test at certain MB/Thread combinations (hyperthreading defect, who knows?). It'd be interesting to found out the ultimate reason why the test fails, but for now i just wante to pass along what i learned (credit to the linked posters for discovering this) to save others overclocking frustration.

It is not unusual to have software bug. But, you are supposed to use maximum amount of ram.

If your test is stable with max amount of ram, you can ignore the error message you get at 64 MB. Your system will crash in real time if the error message you get at 64mb is true. Since your system is not crashing during test with max amount of ram and in real time, you can conclude that the error message you are getting at 64 mb is simply not true.

 

[flyin15sec]

I stopped using Intel Burn test on my i7 system. I would pass Prime95 for extended hours and LinX 20 loop, but failed IBT within the first 2-3 loop.

 

[htoonthura]

I stopped using Intel Burn test on my i7 system. I would pass Prime95 for extended hours and LinX 20 loop, but failed IBT within the first 2-3 loop.

It makes you feel good and safe when your system passes all tests. I am running i7 at 3.8. Yes, it is stable.

 

[dave_74]

This was reported by the linked-to posters, the intel burn test seems to perform correctly when it is set to 16 threads. My system (and the linked to posters) passes any intel burn test at 16 threads, any MB amount, but its hit and miss when using 8 threads. Seems pretty strange why this is the case.

 

[B-Unit]

Intel burn claims to find instabilities / errors in 30 mins that would take 40+ hours on prime. So the fact that i was passing prime on 1-2 hour tests seemed somehow consistent with my failing intel burn tests, and thus my desire to root out the cause as i'm running a custom app sensitive to errors.

If your app is THAT special, I think you could find the 48 hours to test stability.

Or you could just try running at stock.

 

[dave_74]

If your app is THAT special, I think you could find the 48 hours to test stability.

Or you could just try running at stock.

Cause doing a 40+ hour test before doing various 30 min tests and wasting 2 days before learning my system might be unstable is so appealing....

 

[B-Unit]

'Intel claims' is never a good thing to base things on.