Should prime95 really be used?

[LegolasElf]

"Back to the original topic*

Its a smiliar deal with me, i have a stable overclock as seen in my sig, i pass 10 loops of 3dmark05, PCmark04. and laos passes 32M on SuperPi32, but i still fail Prime95 after 17 minutes.IMHO, its no big deal, it still games and tasks perfectly and i have never ran into a single problem so far *knocks on wood*

 

[roYal]

What torture tests do you guys use? I used the default blend yesterday and when I left my computer both cores were at 100% (I run 2 instances). Then when I come back, I notice my pagefile is full and my cpu's are at 0% but the test was still running.

 

[SuperFarStucker]

I think the blend test makes the assumption it will be the only program consuming memory so if you run 2 instances you have to adjust the amount of memory each instance can use otherwise it will enter a page fault deadlock(disk thrashing). I wouldn't use a clock that isn't prime stable. It just isn't worth the small gain.

 

[Michael Warren]

actually thats because its running at a low priority so when you go to do other things prime just gets put on the back burner while your doing whatever else your doing. if you set your priority that prime runs at higher it will put more stress on the cpu and all but cripple your system as the prime calculations will take priority over anything else.

I know that because I normally go to sleep while running prime. Even at high priority it still doesnt use as much bandwidth as a CPU burner would use such as Hot CPU Tester.

 

[Xstatic]

Yes I was being sarcastic. I was saying that because he said his single core can multitask just as good as there is no point in haveing dual core. I disagree fully. My main point is that if you cant pass a 24h stess test then your not stable If it fails then it fails. Thats the point. I can understand that people want the most speed out of thier machines, that is why we are all here. But dont make yourself believe that your stable when your not. Its going to bite you in the end. I wish all of us could reach the speeds we wish for stable. I really do. But that is not the case most of the time. I'll be happy with 2.6-2.7 stable with my 165 at lower volts than I'm at right now. Ill keep pushing till I find out if I can or cannot. Im going to run this system at the sweet spot. More than likely not at the highest stable speed but at a high enough speed with low enough volts. I'd love 3g but its probably out of reach with my cooling setup.

 

[rangi]

Hi all,
RoadWarrior said it all IMO, I have tested quite a few A64 rigs with many different stress testing tools, prime95 allways had issues especially with the Winchester cores, but still was at that time a tried and true method of testing a system for stability.

Now though I know it is unreliable, the testing processes I use are alot tougher than what p95 offers.

I now use superpi32 to test initial clocks speeds and memory stability(one core only at a time as two instances running one will fail in seconds even at stock speeds), cpu burn to burn in for 24hrs then stressCPU using affinity for both cores for 24hrs. SuperNade has a great rational for testing and these methods have provided a better means to indicate stabilty IMO.

I would not use P95 as it would not be as consistant as I would require for a real true method of system stability, a range of different stress tests would be far more consistant in its results than one program that fits all.

Rangi

 

[hawtrawkr]

im not meaning to sound rude here but i seriously dont understand how you guys can say prime is just some made up thing. it does math on your cpu and if it gets the wrong answers back it says so. if i had a program that 90% of people could run error free and i could too but if i clocked my cpu high enough i couldnt pass it i wouldnt say the program lies, id turn down my overclock....

now if you say "i can run everything i do just fine @ speed X but pass prime which is only stable @ speed Y and i choose to run @ speed X" i can understand that (dont agree with it because the gains are so minimal in real world aps but can still understand why someone would say that) but how do people justify in their own minds that prime is nonsense the base of the program is so simple it makes your cpu do advanced calculations and checks them against known correct answers to those problems.... that doesnt change after a year or two the answers will always be what they are.

in the end i run a slew of stability tests not just one or two (id hope we all do) if ANY of them fail or turn up errors or crash i turn down my clock not call that particular bench/stress test "outdated" an error is an error its not primes fault your cpu cant do the math for whatever reason.

the people saying theyve tested multiple a64s and always had issues need to test some more. ive gone through 20 or so a64s in the past couple years for my own personal use and havent had a trouble with any of them and prime (including my dualcores) until my overclock was too high. I do know some of the winchesters were having problems passing prime even at stock settings but aside from the winnies ive yet to experience (or even read about) mass problems with prime and a64s that arent overclocked (or are overclocked and actually stable)

 

[Xstatic]

im not meaning to sound rude here but i seriously dont understand how you guys can say prime is just some made up thing. it does math on your cpu and if it gets the wrong answers back it says so. if i had a program that 90% of people could run error free and i could too but if i clocked my cpu high enough i couldnt pass it i wouldnt say the program lies, id turn down my overclock....

now if you say "i can run everything i do just fine @ speed X but pass prime which is only stable @ speed Y and i choose to run @ speed X" i can understand that (dont agree with it because the gains are so minimal in real world aps but can still understand why someone would say that) but how do people justify in their own minds that prime is nonsense the base of the program is so simple it makes your cpu do advanced calculations and checks them against known correct answers to those problems.... that doesnt change after a year or two the answers will always be what they are.

in the end i run a slew of stability tests not just one or two (id hope we all do) if ANY of them fail or turn up errors or crash i turn down my clock not call that particular bench/stress test "outdated" an error is an error its not primes fault your cpu cant do the math for whatever reason.

the people saying theyve tested multiple a64s and always had issues need to test some more. ive gone through 20 or so a64s in the past couple years for my own personal use and havent had a trouble with any of them and prime (including my dualcores) until my overclock was too high. I do know some of the winchesters were having problems passing prime even at stock settings but aside from the winnies ive yet to experience (or even read about) mass problems with prime and a64s that arent overclocked (or are overclocked and actually stable)

It doesn't get any better than that.^^

 

[damarble]

I'm gonna side with hawtrawkr on this. Everything he's said so far I agree with. No, Prime is not the ultimate stabilty test, but it's part of a mix of tests that should be run.

It's very simple. If your CPU fails Prime, it's not calculating correctly. If you can live with the possiblity of wrong answers once in a while, don't worry about Prime stable.

 

[rangi]

Perhaps p95 under some conditions, the programing that makes prime do what it does errors not the hardware, I know many apps that have issues with XP 98SE etc. Just for one moment can anyone be perfectly sure if prime errors its a hardware issue? I mean if someone were to have a dodgy installation, perhaps even a little corruption cause by other things other than hardware, i.e. Viri, malware, drivers, firmware, incombatibilities, etc.
So many variables to consider with each rig, to use statistical observations p95 is inconsistant when considering other testing programs.
Sure you all could be on the money with p95 being an awesome program to ferret out the dodgy clocks, but up to now it has been my experience to get a second opinion especially with dual core setups.

 

[Xstatic]

I never rely on just 1 program to see if my system is stable. I always use a few to check it out. But alot of the systems that prime is being run on are new installs. I dont install alot of stuff when im oc'n or stressing my system. For one I dont have any reason to be doing alot of other stuff while its stressing. And two I dont see any reason to install anything not needed at the time as if anything happens to the install I dont want to have wasted all that time installing everythinjg and have to do it agian. So I wait till its stable and then proceed to install what Im going to use from that time on.

 

[Stratcat]

A question: Will p95 eventually crash no matter what clock you're using?

No!

Or, perhaps more accurately: "It shouldn't".

But, as Murphy always eventually wins, yeah, somewhere down the months or years, sooner or later, a computer crash inevitably will happen, just as computers that don't run Prime95 eventually crash, sooner or later, too, even though they "shouldn't".

But let's not lose sight of the fact that Prime95 is foremost a GIMPS Distributed Computing Client!

It is intended to be able to run 24/7/365, while searching for Mersenne Primes. Prime95 should "technically" run w/server app type "uptime", given a suitably stable machine. As I stated earlier, the keyword is "technically". It's impossible to keep Murphy at bay, indefinitely, for an infinite period of time.

The "torture test" subset of the Primwe95 software is simply a utility for a prospective GIMPS participant to test/validate the stability of their machine(s), prior to deploying GIMPS, or, for a current GIMPS participant to validate a new or questionable machine, for whatever reason, for GIMPS deployment, without actually running a live Work Unit, possibly producing bad science.

The fact that it has become a somewhat de facto stability test piece of s/w, is actually an unintended consequence.

Many ppl run Distributed Computing 24/7. I AM one of them.

Many ppl run GIMPS 24/7. I am NOT one of them.

However, a team on another tech board that I am associated with, does have a very large GIMPS team w/several hundred machines running GIMPS, many of them 24/7, and they do often run 24/7 for quite a few months, until s/w updates, system maintenance, or relocation, etc., require a shutdown or reboot.

BTW - If it seems Distributed Computing peeps are stability freaks, you're right, they are (at least the ones that care about producing good science, rather than simply being a stat's wh**, at all costs). The last time I popped in on my Distributed Computing team's GIMPS thread (approx one year ago), it took ~ 3 weeks to check a single GIMPS exponent (Work Unit) on an average computer, to determine if it's prime, or not. If it is (lucky you - your name will be recorded in mathematical sciences for posterity - & you may receive a substantial sum of money!), it will then be sent out to other machines w/different architectures & OS's, for verification (called "re-checks"). But, during those 3 weeks you were crunching that exponent, all it takes is 1, yes ONE, bit to be off, to throw the whole show down the tube. Imagine missing finding "the big one" (250K U$D Reward - BTW ), due to your machine being off one bit!!!

Anyway,

FWIW

Strat

 

[MadHyeNa]

I figured i'd mention that my rig right now is dual prime stable but I can't run 2 instances of superpi. It errors immediately when I run 2 (each assigned to a separate core), so I think there might be some problem running 2 at once with a dual core. Anyone else run into this problem?

This happens same for me too, but for me it's due to the HT. That I don't have 2 processors, no dual core, only HT. It didn't even starts to calculate pi, it stops, and gives an error. But my rig is completelly stable, I dual fold 24/7, benchmark .. everything. Can prime95 torture test for days, pcmark for days... etc

About prime95, YES it is outdated, but still it's about maths. Personally I'm using spi 32MB, and then pcmark/3dmark01se (because 01se uses the most of cpu), for over a night or so, and then start folding. If your system is unstable you would see it that's for sure!