• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Presler Stability Testing - Try This !

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

iamjcl

Member
Joined
Aug 4, 2003
I've been going over several 920s, and have seen some strange behaiviour. Without going into a whole bunch of detail, I'll throw out a very abbreviated version of what I've found:

After finding Prime95x2 being generaly stable, while other things would cause crashes (like opening IE, or something very mundane, while the primes, as long as you were not really right on the edge, didn't error much if at all), I decided something odd was going on.

Firstly, Primex2 is great for keeping the system loaded, and has been a great tool. But, something is going on with these 9xx chips.

PCmark04 I think may be discovering very quickly how stable an OC on one of these 9xx chips is. Let me explain.

I noticed that while trying to load my system as much as possible, running my primes (each on its own core), and then running PCmark04, it would often times run fine, but not turn out a score -and say "must select system suite" or something like that, upon completing all the tests. Tests seem to complete, and no error messages or prime failures were recorded. Googling for this did not reveal anything other than PCmark was flakey, or needed to be updated, re-installed, etc...

Here is where it gets interesting: Up the Vcore a bit, or lower the FSB a bit, and you get score every time. And you don't need to wait for the entire test to run to observe this:

You can see, often in the 1st test, that the 2 progress bars do not proceed at the same rate. In my case, the upper one lags behind the lower one, and then may stop. Test finishes, and goes to the next one. If you didn't see the bars get "out of sync" on the first test, you might see it on the 2nd or 3rd. If it goes all the way through and gets a score, great. But you'll know (usually by the end of the 2nd test, ~40 seconds) quickly that you need more vcore or less FSB.

My point in all this is that Prime95x2 can run for 24 hours (It did on mine, numerous times), and you can run all kinds of stuff w/out problems (including PCmark04). But, if you run PCmark04 but DONT get a number at the end, even if the tests seem to complete, somethings unstable.

I guess the main point is that if my chip behaves like anyone elses, this gives a really QUICK way to see how stable you are, without waiting hours or days.

This holds true going way down in FSB / Vcore. I've been able to quickly plot FSB / required voltage on this chip in under an hour what had taken me about 2 days to gather.

Further Prime95x2 etc... should clearly still be a part of testing, but this process sure gives a pretty clear picture very quickly.

I'm going to put one of the other 920s back in and see if it behaves the same way.

You need to make sure that you PRIME 95 X2, each on its own core (set affinity) WHILE doing the PCmark04 runs, or you'll have to run PCmark04 alot more times to begin to see the pattern. You can just run PCmark04 without the primes, but often it will be OK for several runs (or more) before doing what I explained. Also, if doing it without the primes, put up the TASKMGR CPU load graphs - when the issue starts, you will see both CPU loads drop from 100% down into the 50 something range. When this happens, you are guaranteed no score will be displayed at the end, and somethings wrong. The progress indicators will also get "out of sync" with each other at this point.

Hope this makes sense.

Let me know what some of you find.

- Chris
 
Yea, my system could dual prime for hours on ends - but most of my system instability was exposed through other means - i.e inserting a CD into the drive and having a system reboot.. or other means.
 
I agree, i could prime and prime and prime but open up AIM while it was diong that, and i'd get a blue screen.


I've found that running one instance small FFTs and the other on large seems to be more stressful, and the small FFT always is the first to crap out when nearing instability
 
You can find countless threads here were I say use PCMark04 first, then run your Primes if you want and then run PCMark04 again after Primes without rebooting. Some people do them concurrently like you did, but PCMark04 is so fast and so good for quick stability testing, P95 is almost a waste unless you fold and need to test it loaded for a LOOOONNNNGGGG time first. PCMark04 + couple hours of 3DMark loops + PCMark04 again is also very good.

The first test in PCMark04, the top moves in steps of 4-5 compared to the bottome one, unlike the other multithread tests that should move more or less "smoothly" together.

The bench can/will continue even if a test fails and that's when you get the "Must select system suite..." at the end...one failed. If during the multithread tests (first 3 tests), one bar completely stops and the other keeps going, that test failed...might as well abort and make adjustments.
 
Guess I just haven't put that much time on PCmark, but I've used it by itself, and it will often be fine, but will have issues with either prime or BSODs, but concurrently 2xPriming AND PCmarking has really shown me that you can zero in on OC parameters REALLY quickly. Also, these 9 series chips are very different in how they fail vs Northwoods or Prescotts. At least it sure seems that way to me.

- Chris
 
my newark cant pass prime at 2.8, no how, no way. it folds like a champ 24/7 though. i can play all games, do any work i want, surf the web, open all kinds of apps, but fails after just 1 min of prime. iver never used prime to test my systems, and i never really will. as far as im concerned, if its stable folding, its stable for me. =0)
 
dmatko said:
my newark cant pass prime at 2.8, no how, no way. it folds like a champ 24/7 though. i can play all games, do any work i want, surf the web, open all kinds of apps, but fails after just 1 min of prime. iver never used prime to test my systems, and i never really will. as far as im concerned, if its stable folding, its stable for me. =0)

you can't be serious, stop folding! If the box can't pass prime95, then it's going to yeild all kinds of bad WUs.
 
Thats what surprised me so much about PCmark quickly identifying issues, when I could (did) primex2 for as much as 36 hours at a clip, error-free.

I agree - if you cant prime, it aint stable.

- Chris
 
Back