The Prime 95 fallacy

[cbehnken]

Yes, but the topic here is whether Prime95 gives errors on a stable system. It does not. Run 3dmark AND Prime95. A system must pass both before it is stable.

 

[ectx]

Don't fool yourself. If you run a proven software like Prime95 and it errors out, there is only one conlusion: there is something wrong with your setup.

Stop fooling yourself and face the problems. Just claiming your pc is stable does not mean it is.

Sigh!! This is all Micorsoft's fault. They make lousy software that has all kinds of problems so people are used to having some problems.

 

[Lithan]

Personally I'd consider not crashing every hour more important than 100mhz, but that's just me.

 

[RealityTime]

Originally posted by: Lithan
If you can't pass prime it's not because somethings wrong with prime it's because your computer ISN'T STABLE. It doesn't matter if you can run 3dmark, if you can't pass prime (or anything else that's properly coded), then your machine ISN'T STABLE. You may be happy with a machine that ISN'T STABLE. But you can't call it stable because it ISN'T STABLE. You might as well say that a table missing a leg is perfectly stable because you never bump it or put anything on it, so it doesn't tip over. Always funny seeing people sell cpu's as stable @ XXXXmhz, then someone buys the cpu and it can't even get close to those speeds. The sellers excuse? Well it ran so and so a program just fine. You can't call something stable unless it runs EVERYTHING. And so far Prime is the best program for testing cpu stability. Get the picture?

relax

 

[SinfulWeeper]

Originally posted by: belazeru
has any one had there system crash on P95 while running everthing stock?

Plenty of times. I do it all the time when I assemble a new system. I like to see what kind of difference my OC's make.

 

[StrangerGuy]

If a PC fails in a P95 torture test, IMO then it is not 100% stable, no matter how much you run other apps with no apparent errors or crashes whatsoever, since P95 explicitly states that the torture test is based on correct and proven calculation results.

 

[imported_NoGodForMe]

Originally posted by: belazeru
has any one had there system crash on P95 while running everthing stock?

Yes, that's what all my testing is being done at.
Have you seen my web site? It has all the details.
http://www.nogodforme.com/MyBabyTera.htm

I received my memory back from Corsair last night. They replaced my PC3200LLPro (2326) with PC3200XLPro (2225). Thanks guys. I was all excited, until I put it in the AMD machine and tested.
Failed memtest on the first pass at 2225, one stick is bad. Passes memtest and Prime95 at 2336 timings.

I think Corsair has a bad testing machine. Because in both cases of my memory, one stick fails, while the other passes.

 

[big4x4]

Ok, I am testing all my computers with Prime 95 as we speak. What test is the best to use? Right now I am maximun power consumtion and I will see how it goes!

 

[keichan82]

Here's some results along with the type of Prinme95 ran, length or Prime95 ran, CPU info, etc.

http://forums.anandtech.com/me...9790&enterthread=y

Please post your results so we can compare. Thanks

 

[big4x4]

Prime fails on 2/3 of my computers

 

[VirtualLarry]

Originally posted by: big4x4
I don't buy into that prime stuff. I used it once, and never used it again. I would just loop 3dmark01 tests 1 and 3. Then, just do your everyday stuff and if it doesn't crash / freeze you are golden! Prime might be useful to some, but I know many people who use it and their computer fails it eventhough it is stable in everything else they do!

I guess you didn't read the README.TXT included with Prime95 regarding the torture test. If the computer crashes or returns the wrong data - IT'S NOT 100% STABLE.

Anyone who thinks otherwise, needs to step back and evaluate what's going on.

BTW, Prime95 tends to stress the FPU more than the integer unit, and most "ordinary" applications don't use the FPU much at all. When I first got my AthlonXP and started toying with OC'ing it, I discovered that. Leaving the OpenGL screensaver running, would at times, and worse after a longer period of letting it run, start to occationally display missing and strange polygons, every once in a while. So I let Prime95 run overnight, and got a few bad results. Diagnosis? System not stable - needed a few more fractional volts for vcore.

And yet, through all of that, my "normal" apps never skipped a beat.

 

[Goi]

^^^ What he said.

Prime95 is extremely FPU intensive. It works well as a stability tester, but just because a system is Prime95 stable doesn't mean it's stable. There's still the integer part to handle.

 

[VirtualLarry]

Originally posted by: gururu
I stopped using prime95 a long time ago. games and movies have always let me know when something was amiss. fine tuning an overclock usually takes me a good couple of weeks as I believe it should.
IMO, the reality is that prime95 only tells you that your system is stable enough for prime95 but doesn't guarantee that your system is absolutely stable. as for me, my system only needs to be stable enough for the apps I use.

I'm going to guess that you are P4 user, correct? Prime95 tends to be FPU-heavy, which is great for testing for Athlon stability, since that is the "hot spot" of the CPU, and the part that is likely to fail first. On the other hand, with a P4, the "hot spot" is the double-pumped ALU, used for integer calculations. Some media-encoding/decoding tasks fall under that category, which is why in your case you might be "Prime stable", but not 100% stable overall.

So in other words, if Prime95 shows an error, then you aren't 100% stable, but if it shows no errors, it's also not proof that you are, either. For an Athlon I would say that's a safe bet, but for a P4, I'd run some other tests too.

Not to mention, as some have also pointed out, running looping 3DMark often can be a good test as well, it tends to test a lot of things in the system, including the GPU, and necessarily, the stability of your power plane, both the PSU and the mobo/CPU/AGP, as well as the internal case temps of both the CPU and GPU.

 

[gururu]

Originally posted by: VirtualLarry

Originally posted by: gururu
I stopped using prime95 a long time ago. games and movies have always let me know when something was amiss. fine tuning an overclock usually takes me a good couple of weeks as I believe it should.
IMO, the reality is that prime95 only tells you that your system is stable enough for prime95 but doesn't guarantee that your system is absolutely stable. as for me, my system only needs to be stable enough for the apps I use.

I'm going to guess that you are P4 user, correct? Prime95 tends to be FPU-heavy, which is great for testing for Athlon stability, since that is the "hot spot" of the CPU, and the part that is likely to fail first. On the other hand, with a P4, the "hot spot" is the double-pumped ALU, used for integer calculations. Some media-encoding/decoding tasks fall under that category, which is why in your case you might be "Prime stable", but not 100% stable overall.

So in other words, if Prime95 shows an error, then you aren't 100% stable, but if it shows no errors, it's also not proof that you are, either. For an Athlon I would say that's a safe bet, but for a P4, I'd run some other tests too.

Not to mention, as some have also pointed out, running looping 3DMark often can be a good test as well, it tends to test a lot of things in the system, including the GPU, and necessarily, the stability of your power plane, both the PSU and the mobo/CPU/AGP, as well as the internal case temps of both the CPU and GPU.

yea I'm a P4 user. I never would have thought that to be the case. very interesting! thanks!

 

[VirtualLarry]

Originally posted by: Goi
Prime95 is extremely FPU intensive. It works well as a stability tester, but just because a system is Prime95 stable doesn't mean it's stable. There's still the integer part to handle.

I was going to ask what your preference for a good integer-based test would be, and then I remembered the config that I use.

This really only works if you have some large well-compressed .ZIP files, bigger than 10MB or so, but small enough to fit entirely into SMARTDRVs' 32MB cache limit. If you don't have some files like that on the HD, then you could, I suppose, put some on a CD, and make sure to set up SMARTDRV to be able to also cache CD-ROM contents, I think it takes another switch. Anyways, the idea is to run repeated loops of PKUNZIP version 2.04g for DOS, over some fairly large ZIP files. Usually I test for a while without SMARTDRV, as this exercises the IDE channel hardware, and then using SMARTDRV, such that the entire file is in RAM. Possibly you could use RAMDRIVE to set up something similar.

But basically, the very tightly-coded (in assembly) loops in PKZIP for DOS, tend to really stress your CPU (integer only), cache, and memory subsystems, such that any weakness usually shows up within a few hours, or the system just freezes solid. (It may take a few hours for the CPU to really heat up to max temps.) I use a batch file to do the looping, and print a message if there is a shell error-code returned. If the system hangs or otherwise seriously glitches out, that's another good indicator that the test failed.

I might just have to set up a bootable CD with the right stuff on it, for testing the integer part of P4 systems, now that I think about it. Could be another handy tool for the toolbox.

 

[big4x4]

Which of the 3 prime 95 stability tests shoud I use?

 

[skyking]

I've only ever had prime fail while overclocking, or when I had a bad component.
I use memtest 86+ to find the apparent limits of the ram, then prime to the highest stable processor speed. has not failed me yet

 

[rogue1979]

Originally posted by: skyking
I've only ever had prime fail while overclocking, or when I had a bad component.
I use memtest 86+ to find the apparent limits of the ram, then prime to the highest stable processor speed. has not failed me yet

I use a similar test, but add one more. Run Prime while web browsing, downloading music, playing the media player and burning CD's at the same time. If that holds up for a few hours then you are good to go.

While Prime95 isn't the end all of stability tests, the fact remains.

NO Prime95, NO stable computer.

 

[dug777]

I started a similar thread in general hardware, and posed the following question, without a satisfactory answer

I can't run prime95 at my oc (2000+ to 2600+), but have no problems whatsoever (literally...none) with my stability with any of the other programs i use, far cry, d3, sims2, simcity4, IE/Excel/Word etc, and can happily encode dvds while running winamp and using IE at the same time, benching with 3dmarks from 01-05 doesnt reveal any instability, as doesn't memtest, superpi, or the sandra burn-in test (ran the cpu tests/mem bandwidth/cache/file system combo 100 times the other day after i discovered i wasn't prime stable with no probs, and have left it running overnight with no probs b4), and my machine runs for 5 days or more between restarts (tho hibernated every nite, too noisy, and fan lights 2 bright for sleeping ).

So what other programs will show up the prime instability?

EDIT: the restarts are mainly to put new drivers on, or when it gets shutdown proper when i go away for a day or two, and it has been running at this level of o'c for the past six montsh or so.

 

[htmlmasterdave]

Originally posted by: Lithan
If you can't pass prime it's not because somethings wrong with prime it's because your computer ISN'T STABLE. It doesn't matter if you can run 3dmark, if you can't pass prime (or anything else that's properly coded), then your machine ISN'T STABLE. You may be happy with a machine that ISN'T STABLE. But you can't call it stable because it ISN'T STABLE. You might as well say that a table missing a leg is perfectly stable because you never bump it or put anything on it, so it doesn't tip over. Always funny seeing people sell cpu's as stable @ XXXXmhz, then someone buys the cpu and it can't even get close to those speeds. The sellers excuse? Well it ran so and so a program just fine. You can't call something stable unless it runs EVERYTHING. And so far Prime is the best program for testing cpu stability. Get the picture?

This guy is totally correct. Anyone who doubts this either doubts the program or doesn't know what stable means. You are essentially betting on the fact that when the cpu screws up it won't be something important. My A64 2800+ for example was posting up to 2.4ghz and booting into windows. However Suse wouldn't boot properly and Prime95 wasn't stable. Just because something appears to work doesn't mean it's not slowly recking things in the background or making mistakes. It can be really bad if it miscalculates at the wrong time. If any of you who think a few extra mhz is worth the stability, I cannot imagine you doing anything remotely imporant on that system at all. (I guess none of you have lost work to some hardware failure or software glitch, losing many hours of work, possibly losing grades or money)

I think people are in two camps. One camp actually uses computers to produce things. The other camp might as well have an xbox that supports PC games

 

[gururu]

Originally posted by: htmlmasterdave
This guy is totally correct. Anyone who doubts this either doubts the program or doesn't know what stable means. You are essentially betting on the fact that when the cpu screws up it won't be something important. My A64 2800+ for example was posting up to 2.4ghz and booting into windows. However Suse wouldn't boot properly and Prime95 wasn't stable. Just because something appears to work doesn't mean it's not slowly recking things in the background or making mistakes. It can be really bad if it miscalculates at the wrong time. If any of you who think a few extra mhz is worth the stability, I cannot imagine you doing anything remotely imporant on that system at all. (I guess none of you have lost work to some hardware failure or software glitch, losing many hours of work, possibly losing grades or money)

I think people are in two camps. One camp actually uses computers to produce things. The other camp might as well have an xbox that supports PC games

if you can't run prime95 you aren't stable, I don't think anyone doubts. But the same goes with any other application as well. the question in my mind is, and already addressed by VirtualLarry, is that if you CAN run prime95, you STILL aren't necessarily stable.

 

[htmlmasterdave]

Yep, I agree gururu, stable -> prime95, but prime95 doesn't imply stable (just some people are questioning whether or not prime95 is reliable when it reports errors)

 

[dug777]

well...give me one other app that will replicate the prime instability my machine exhibits?

As far as my situation is concerned (read my post b4 for details) saying that my computer is unstable is a bit like saying that my table isn't stable because if i try to set it up on a vertical cliff face in a tropical cyclone it won't stay where its put (crap analogy i know ), but perhaps u take my point? You wouldn't call a Ferrari Enzo unstable, but whack it into a corner on the wrong surface 2 hard and you'll still end-up on your head

I certainly wouldn't make the claim of 100% stability because i fail prime, but i can't see its relevance to the stability of my rig in ANY other situation, but show me the light and i'll give up talking crap on the forums forever (well, crap about prime95 anyway )

(reading back i realise the analogy is kinda bs, but there is a point behind it-what, i'm not so sure )

 

[Goi]

As others and I myself have mentioned, it's fine if you're happy with your system and you don't feel the need for 100% stability. We aren't here to diss you or your system. You also seem to be clear that your system isn't 100% stable so that's fine. I don't see the point of any further argument since both sides are pretty clear.

Your analogies, as you've mentioned, are pretty whack. Here's my take on it. If your Ferrari has a recall on an airbag that doesn't inflate properly, would you wanna drive it? Sure, 99.9% of the time you won't end up in a situation needing the airbag, and you might be perfectly comfortable of taking that risk. That's ok. Saying that the Ferrari is 100% reliable however is not, because it isn't. Why sin't it? Because something that *should* work, *isn't working*.

In this case, Prime95, being a perfectly well coded piece of software, *should* work, but it *isn't working*.

Again, if you're comfortable with the risk, more power to you, but many others aren't, which is why the original Pentium errata (and subsequent ones) was such a big deal.

 

[dug777]

meh...ok


(but like a little kid who wont shut up -i would be interested to see my problem recreated under another piece of software, it just interests me since i am clearly completely prime unstable (illegal sumout immediately ), but the the other apps aren't screwy. As pple have mentioned i have unknowingly been running the risk of hdd corruptions etc etc, but i find it hard 2 believe that i have just been very, very lucky for the past 6 months. (i'm not trying to argumentative here, genuinely interested)

Could it be my psu (i can't even connect a burner or drop in another 512mb stick without random crashes in games)?