PRIME Don't Place Much Faith In It?

[DasFox]

Someone told me this:

Originally posted by: AruisDante
That's a CPU error IRRC. If you get a RAM error it gives you a much longer message.

But really, prime stability means nothing to me. If it can do what I want without crashing, I don't care if it fails instantly. So don't place too much value on prime stability, test it by doing what you'd actually want the thing to do.

My reply:

I thought people SWEAR by PRIME, that if you get any errors then there is a miscalculation on the hardware, and it's not running correct.

See being somewhat of a PRIME newbie in understanding it this way I don't really know, but I'm not a newbie as far as the program is concerned, I've known it for years.

Well the thing is I ran PRIME on my box one time without any overclock, and with it all stock, and from what I can tell if someone has a stock system then PRIME isn't going to give errors, and if it does that means you have bad hardware.

I don't want to ruin my CPU if it's giving me errors. I thought the hardcore overclocking geeks swore by PRIME, and if you get an error, then it's not correct, and you fix it till it doesn't give you an error.

Anyhow as far as killing my CPU is concerned the volts are at 1.5 and it runs cool, 43c idle and 50c loaded with PRIME. So maybe as far as PRIME is concerned even though it gives an error if I can game and never crash and the box seems to run stable, and I have the CPU cool enough maybe it's ok?

THANKS

 

[Duvie]

It really is a toss up....Quick prime errors showing a really flaky system will eventually even in most of the programs you run creep up....

NOw if you game for 2 hours at a time make sure you are at least 2 hours prime stable....2 hours prime stability will weed out most errors of a truly flaky systems and chances are only gaming you will never see any errors...

If you encode movies of HD content an dthey take 4-5 hours then prime for those durations....


I do DC..which is distributed or heavy number crunchings. I dont want to take the risk of EEUs wasting my time into a 40hour work unit or the chance I am sending back bad data. I make sure my systems are 24hours stable of prime95 or orthos (which is prime95 anyways)


Bottomline is that prime95 and orthops basically stress your system in ways more extreme then most real world apps we run (this may not apply to some mission critical server and heavy computation apps). IE the heat...Notice how most anything else you run will never come close to the heat prime95/orthos produces? power is in the same boat....

It becomes a safety blanket. we wrap ourselves into the security we have ran it to weed out imperfections in out hardware....




ONe correction to above....For gamers I would run it 2 hours with small FFT to stress cpu and then another 2 hours of blended to stress memory or about 4-5 passes of memtest dos bootup version...

Times I have done that the system has never crashed for basic gaming and what not apps.

Upped it to 4-5 hours and I have never had a system crash doing any encoding or any apps period....I could still get an occasional EEU (early end units) in Folding at Home (maybe like 1 every few weeks) which is why it is strictly 24 hours prime95 testing in the beginning when I set system up. NOw not all EEU's are wrong. Some are actually the correct results. Some are bad proteins. I have had ones and posted the specs on the unit to have the mods confirm they are known problem WUs....

Often times I find limit with stability of prime95 and then take it down a few notches anyways to deal with less heat

 

[Shimmishim]

i used to do 6 hours of small fft.

now i try to do 12+ hours of blend test. at least with the c2duo systems i've run, blend seems to be the breaking or making point of stability for me.

if it'll run blend, it'll run small fft.

 

[Roguestar]

I do synthetic tests and real tests. If can run Orthos blend for >12 hours on it overnight without a single error, that's good. That stresses the CPU and RAM but for overall "am I going to use this ok" testing there's nothing like playing two or three hours of a high-demand game. Besides, playing battlefield 2142 is a hell of a lot more entertaining than waiting for prime95 to finish .

In conclusion, passing stress-tests = good, but always give it a whirl at whatever you're going to be using it for. If I got a few errors in orthos after a long time it wouldn't bother me much if I could still play games for hours without experiencing any trouble .

 

[Paratus]

When using Memtest does a single error indicate to high of an mem OC? Or is there an acceptable number of errors. (I'm guessing even one is too much)

 

[Duvie]

Originally posted by: Paratus
When using Memtest does a single error indicate to high of an mem OC? Or is there an acceptable number of errors. (I'm guessing even one is too much)

No errors are acceptable...any ram error can produce an error in an application...

You want error free for multiple testa dn multiple passes. i would usually run it overnight (8+ hours)...It is multiple passes but really depends on the amount of memory in the system being tested...

You cannot always blame it on the ram though...It is as much a test on your whole memory subsystem...

 

[cryptonomicon]

what I do:

Memtest x86+: two passes of test #5 with no errors. yes you can run it for more, but I use a quicker method which is finding the point that I get the first errors, then back it down 5mhz and it is usually stable.

prime95: This program needs to run for hours, not minutes. The more the better, with 12 hours being a good point to call it stable enough for almost anything.

--> Also the main reason people use PRIME is because you don't want to discover that your OC is not stable while using an application with live data.. a game.. software.. ect. Of course you can OC without using PRIME but that is just taking a risk of failure at some later point when you don't expect it.

 

[Shimmishim]

i remember someone who ran prime95 for over 800 hours (forgot he was running prime on a back up pc) and it generated an error...

so i agree with those that you should do other tests such as gaming, encoding, whatever else you do on a regular basis just to be doubly sure.

 

[Roguestar]

"Error: no more prime numbers to calculate. Please press F8 to reboot the universe."

 

[Yellowbeard]

I'm of mixed thoughts on this. First, I totally agree on the need for 100% stability, not 99.9%, on a computer; especially if the machine is mission critical. However, if the machine is 100% percent stable doing what you built it for, then it works perfectly, right?

I have seen a few odd cases where a person has an issue that the computer will do EVERYTHING perfectly except 1 specific benchmark and they seem to think it is the end of the world. So, I guess what I am getting at is that there is no 1 specific perfect test for stability. However, I do believe that benchmarks are great tools.

 

[dwell]

That's bull. If it fails Prime/Orthos your system is not stable, period. Especially if it fails in minutes. If it fails either quickly there's a good chance games, etc are going to crash. It fails after hours there's a good chance your DC computations could be off or an encoding could mess up.

Some people think just booting up at "X"Ghz is good enough. Not so.

 

[Yellowbeard]

Originally posted by: Chris
That's bull.

What's bull?

 

[dwell]

Originally posted by: Yellowbeard

Originally posted by: Chris
That's bull.

What's bull?

Sorry, was addressing the OP's assertion, not your's

 

[Yellowbeard]

It's cool, no hurt feelings. I was just curious as to which one you were addressing.

 

[gerwen]

If it's stable when torture tested, (orthos/prime/whatever for 12+ hours) then you can be reasonably sure that it's stable for anything you want to do with it.

If it's not stable under a torture test, then you open yourself up to problems. Sure it may run X game for 2 hours without crashing.... this time. The error chance is there and it may catch up with you.

You may run something else at a later date and have problems. If it's much later, there's a good chance you'll forget about your little torture test instability and start troubleshooting elsewhere, and end up blaming the program itself for instability. Meanwhile, it was a general system instability that you ignored when testing specifically for it.

Leaving known instability in your system is asking for headaches later.

 

[Roguestar]

I've just reduced my vCore from the 1.36 in my sig (1.3625 actually) to 1.35 and am testing it. I've run an hour of orthos blend but I'm getting bored so I'm going to play some Half-Life 2: Episode One with everything turned up as high as it will go . I'm going to have it run orthos overnight but needless to say if I start getting problems I'll know I'm not as stable as I'd like to be :beer:.

 

[aka1nas]

What I like to do is run Orthos/Prime overnight alongside something like 3dMark set to loop infinitely. That way you at least are stressing most if not all the components of your rig at the same time and testing if it is stable with max power consumption, etc.

 

[nyker96]

I think some errors show up sooner in prime but some show up later. For instance I built a system for a friend 939 based and ran prime which became unstable after 2 hr but ortho caught it in 20min. So I think it makes sense to run a few of them just to be sure. I in fact run prime, ortho, folding, 3dmark, another memory testing mem86 etc sometimes together.

 

[aka1nas]

Unless the system is extremely unstable it's pretty much a statistical event. It may happen in the first 20 minutes or after 20 hours, but one is not neccessarily more stable than the other.

 

[dwell]

Originally posted by: nyker96
I think some errors show up sooner in prime but some show up later. For instance I built a system for a friend 939 based and ran prime which became unstable after 2 hr but ortho caught it in 20min. So I think it makes sense to run a few of them just to be sure. I in fact run prime, ortho, folding, 3dmark, another memory testing mem86 etc sometimes together.

My system will run Prime95 for hours under 1.4v but fail on Orthos in minutes. At 1.4v I had Orthos running for 11+ hours. Called it a day at that point.

 

[Duvie]

Originally posted by: Chris

Originally posted by: nyker96
I think some errors show up sooner in prime but some show up later. For instance I built a system for a friend 939 based and ran prime which became unstable after 2 hr but ortho caught it in 20min. So I think it makes sense to run a few of them just to be sure. I in fact run prime, ortho, folding, 3dmark, another memory testing mem86 etc sometimes together.

My system will run Prime95 for hours under 1.4v but fail on Orthos in minutes. At 1.4v I had Orthos running for 11+ hours. Called it a day at that point.

So by what you are staing you ran 2 instances of prime95 or the multicore version? Older versions like 24.14 onluy stresses one core...Orthos is just a fancy frontend to load 2 instances of prime95...

There really should be no difference...unless you were not stressing 2 cores and thus it makes sense the discrepancy between the two...

 

[dwell]

Originally posted by: Duvie
So by what you are staing you ran 2 instances of prime95 or the multicore version? Older versions like 24.14 onluy stresses one core...Orthos is just a fancy frontend to load 2 instances of prime95...

There really should be no difference...unless you were not stressing 2 cores and thus it makes sense the discrepancy between the two...

Yes, I was running the single-threaded version of Prime. Taxing both cores is what shakes out the instability in my setup.

 

[Stumps]

last time I ran Prime95 on my rig (A64 3000+@2.87GHz) I was able to run it for more than 52hrs with out any errors...I got sick of waiting for it to crash so stop running prime95...I think it's pretty safe to say that my rig is stable at 2.87ghz.

As a bonus I was able to back of the Vcore 0.05 from 1.55v to 1.5v...it knocked temps down quite a bit, I rarely get any higher than 4c above room temp.

 

[Roguestar]

52 hours is a bit extreme . I'm happy with 12-18 hours of orthos blend (leave it on when I go to sleep, then when I go out to work) and not getting any errors in high-demand games. :beer:

 

[Duvie]

I can give you one the other way...Had mysterious reboots, BSOD...system wasn't stable for more then 3-5 hours...Even taken back to all stock settings would still eventually BSOD about same 3-5 hours....

I ran orthos blend for near 24hours without incident at the OC'd speed...thought fine!! I will go back to working....3 hours into Folding a cpu core and a gpu core I get a BSOD....Ultimately I swapped every part out. CPU, PSU, HDD, video card.....still it would creep up time and time again. Again orthos runs fine for 12+ hours....I then swapped the Gskill HZ sticks for my ballistix memory sticks and 3 weeks later not a peep...

So orthos ran fine yet I still had instability...

Do not put all you eggs in one basket and solely trust this application. this isn't my only story...I had a P4 that couldn't finish a 3 hour encoding yet passed prime testing for 24 hours. ultimately it had something to do with memory timings or chipset timings....

I think it is a good cpu tester. However I don not think it is a good system tester, memory tester (Better with AMDs and the integrated memory controller) , or power tester.

Beyond cpu stressing I dont really care for this app.