Maths/SIMD tests "No operations reported in timeout period"

[JamesBulpin]

In approximately 1 in 5 1 hour runs of BurnInTest 5.1 or 5.2 on a particular class of hardware I get the error "No operations reported in timeout period" for either the CPU Maths or SIMD test with a negative number of operations reported. E.g:

Test Name                   Cycles   Operations      Result Errors   Last Error
              CPU - Maths   3        -1996516420     FAIL   1        No operations reported in timeout period
               CPU - SIMD   23       609 Million     PASS   0        No errors
             Memory (RAM)   1        1.561 Billion   PASS   0        No errors
               Disk (C: )   0        72.581 Million  PASS   0        No errors
                Network 1   4        36000           PASS   0        No errors

An old thread http://www.passmark.com/forum/showthread.php?t=474 reported a similar error and it was suggested that this was a BIT bug and was fixed in a new version. I assume that new version was 5.2. I see these errors with 5.2 - is there a known bug? (I want to check before I start chasing hardware/platform issues.)

[Ian (PassMark)]

This is not a known problem.

On the surface this would appear to be an operation counter overflow in BurnInTest. The question is why there is an overflow, as BurnInTest was changed in v5.1.1012 to overcome any possibility of this occurring (even though we could not reproduce any problem, we changed it anyway). In hindsight, I now wonder in the post you reference if there was an overflow in BurnInTest because of some problem with the system hardware timer that BurnInTest uses to measure the duration of the individual maths tests.

Interestingly in the post referenced, there was a problem caused by the CD drive on the system (possibly related). I wonder if this was a problem with the CD/ IDE controller interfering with the interrupt priorities for the system timer or a more general system timer problem??? I would check your interrupt settings and ... hmmm not exactly sure at the moment.

For the results you posted, what was the duration? Are the results you posted a 1 Hour test run as to have only 3 cycles (much much too little) of the CPU maths test would seem to point toward a problem with the system timer (where each individual test runs much too long)?

If you run the test with only the CPU/SIMD tests and logging set to activity trace level 2, this will show more information about what is going on. Please try this, and if you see the error, please email me the trace files(s) for the test to help [at] passmark [dot] com.

Maybe 6 months ago we changed PerformanceTest to work around some unpatched AMD systems, due to a system timer problem on some of AMD dual core systems, mainly the X2. There is an AMD patch for this. I do not believe this problem would have any effect on BurnInTest as the problem was quite specific. What CPU is in the system and how many CPU's/cores are there?

Regards,
Ian (PassMark)

[JamesBulpin]

Thanks for your reply Ian, and apologies for my delayed reply, I've been on vacation.

The run duration was 1 hour. This was with dual core AMD systems but we suspect the problem is elsewhere in the system. My colleagues who have been investigating have found a problem with time handling by the platform. We have a fix and will be rerunning tests this week.

Thanks,
James

[Ian (PassMark)]

It would be great if you could post the problem you find.

Thanks.
Ian (PassMark).