[Iroh]

Has anyone else has issues with IBT/LinX where the results all match but the time elapsed / gigaflop speed does not?

If I leave the board at auto settings, everything matches as it should. If I select any other CPU multiplier in the BIOS, even keying in stock setting, the results will match (and the program will claim it's stable with about the same total time) but the time on about 20% of the lines will be doubled.

I know when it's unstable from insufficient voltage because the time will return 0.000 (again with the proper result), but half the time at stock voltage and speed just doesn't make sense.

I guess the question is what kind of system instability this could be.

 

[sn0man]

You're gonna need some screenshots of both your LinX results as well as your CPU-Z settings (side by side), with an example of situation 1 vs. situation 2.

Your description is kind of unclear/confusing.

 

[Iroh]

Quote:

Originally Posted by sn0man You're gonna need some screenshots of both your LinX results as well as your CPU-Z settings (side by side), with an example of situation 1 vs. situation 2. Your description is kind of unclear/confusing.

Alright, I'll do a couple runs and post them in a few minutes.

 

[Iroh]

It wasn't cooperating today. Yesterday about every other pass was much extra displayed time but it didn't have any slowdown in reality, when I keyed in stock cpu and northbridge speeds and volts.

On Automatic All

Selected x16 multiplier in BIOS, all other settings automatic. Note wacky vcore and gflop rating.

Selected x16 multiplier in BIOS, vcore manually keyed at 1.35v where it should be, all else on automatic. Note gflop speed normal this time; yesterday this setting gave the extra time listed and abnormally low gflop speed (<20 gflop) on the third pass and every other thereafter. Today it behaved.

I had to push it to make it do it today, by going to x17 multiplier and 1.375v as set in BIOS. If this multiplier/voltage is software-set it'll run linpack all day without a cough or hiccup. This is what it looks like when it gives a wacky time.

It comes and goes. Some days I can key in x18 multiplier and it'll pass everything I throw at it, and some days like yesterday it won't pass unless I leave it on auto and set what I speed/voltage I want in software.

Why would linpack show this instead of a failed result?

 

[sn0man]

If you're stable but not quite close enough to 'optimal' voltage (for a given overclock), you may find that LinX will pass, but the GFlops will be inconsistent or low.

Another thing is, the time it takes to run each pass is highly dependent upon processor speed, memory speed, AND problem size / memory used...

In looking at the last screenshot, not only is your processor running at a faster speed - but you're also running a larger problem size in LinX (and using 1275MB of memory rather than the 768MB used in the other tests).

 

[Iroh]

I forgot I set it to mem all as a last-ditch effort to make it do it, but yeah I understand what you're saying about time. Bigger datasets take longer.

The weird thing is, at the auto vdroop, I can set 1.45v at x17 in the BIOS and have this happen, or use AMD Overdrive/ PhenomMSR/ K10Stat at 1.42v and x18 running an effectively lower voltage and be fully stable in linx/prime/occt.

Oh well, so much for this board being much of an overclocker, even with the new BIOS revision. Just wanted to make sure I wasn't running into a known bug. Thanks for the input.