How important is Prime 24/7 stable, really?

[The Sauce]

As I have been overclocking and testing things out here I have wondered how valid the practice of stability testing with Prime is for real-world use. I mean, under real world situations you will never, ever see sustained, 4-core 100% utilization for extended periods. Hence you will never see those kind of temps, or need those kinds of voltages that we move up to in order to get the holy grail of 24h Prime stability.

Under regular computing conditions for the average user, what is the longest you may possibly see 4-core 100% utilization in situations like gaming, encoding, compression/decompression? Maybe 1-2 minutes? If so wouldn't Prime Small FFT for 15-30 minutes be more representative of the maximum severe stress a CPU might see? Or maybe the Blend test is more representative of real world usage. It seems to me that doing this is like buying a new car and running it at 140 mph for 5 days before saying it's ok to occasionally run it at 80 mph when passing another car.

So I ask, is there any rationale for funning small FFT 24h? Is there a good correlation with real-world stability or is this setting the bar way too high?

 

[VirtualLarry]

What is this, a noob question? Of course Prime95 24-hour stability is important. You don't want your computer crashing or corrupting data, do you?

In my case, I demand stability measured in months at a time. So even though my overclock was "stable" (according to every 24hr stress-test I could throw at it), it would still spontainiously reboot about a week after it was left on, running SoB (a DC program) 24x7. That was unacceptable, so I had to downclock from 8x400 to 8x350.

 

[WildW]

Long periods of stress testing really aren't about testing if the cooling can cope with that duration (not beyond the first few minutes anyway) or if the processor can handle an hour of hard work. You're testing to see if the processor makes mistakes.

An unstable overclock can cause cpu instructions to give the wrong result - that is, get its sums wrong. By testing for a longer time you're more certain that this isn't happening. An unstable overclock might get only one instruction wrong in a billion, but then again your processor is doing millions every second. Testing for 30 minutes doesn't prove that your CPU will always work fine for at least 30 minutes - only that in 30 minutes no (detected) errors occurred in the calculation of whatever Prime95 actually does. (There could still have been errors occurring in other system processes.)

For a lot of things this doesn't really appear to matter. An error will give a slightly wrong answer to a sum that happens somewhere in a game, or somewhere in the background of the operating system, and nothing obvious will go wrong. But sometimes that mistake will cause your program to crash out, or the whole system to reboot.

You may shrug and be happy enough that your system doesn't randomly lock up too often to bother you. Other folks may need to rely on the numbers their computer generates, for real work perhaps, or during an important clan match when you really don't want to crash out.

 

[The Sauce]

Well I don't consider myself a noob. I've been overclocking for 20 years starting with the Intel 80286. I have always done the Prime thing...at least as long as it has been around, without questioning it. Now I am questioning it. Just wondering what people think about this thing that we have all been doing...without thinking about it or if it makes sense.

 

[The Sauce]

An unstable overclock can cause cpu instructions to give the wrong result - that is, get its sums wrong. By testing for a longer time you're more certain that this isn't happening. An unstable overclock might get only one instruction wrong in a billion, but then again your processor is doing millions every second. Testing for 30 minutes doesn't prove that your CPU will always work fine for at least 30 minutes - only that in 30 minutes no (detected) errors occurred in the calculation of whatever Prime95 actually does. (There could still have been errors occurring in other system processes.).

I hear you, but Prime puts a much greater stress on the CPU in terms of heat than it will actually ever likely see, hence making it more likely to produce an error then under ordinary conditions. That is what I am wondering about. If your cores are running at <50&#37;, as they most likely usually do, they are running cooler, more stable, and they may never produce that error that they would at 100% for 24 hours when they are hotter and under artificially high stress that will never be produced under ordinary use.

Take the car analogy. Is it necessary to run your Taurus at 140 mph for 5 days straight in order to say that it's ok to pass at 80mph for 1 minute?

 

[biostud]

it depends on how often you're willing to run into a crash, most would like to minimize the chances to crash significantly by sacrificing 1-2&#37; speed increase. You can ask: will the difference going from 3.8Ghz to 4Ghz in terms of speed be noticeably, or will the crash of the computer during an intense battle be worth the speed increase.

 

[The Sauce]

it depends on how often you're willing to run into a crash, most would like to minimize the chances to crash significantly by sacrificing 1-2&#37; speed increase. You can ask: will the difference going from 3.8Ghz to 4Ghz in terms of speed be noticeably, or will the crash of the computer during an intense battle be worth the speed increase.

Oh, I agree with you completely. Crashes are bad. You should never sacrifice stability for a few extra MHz. The question is, does running Prime for 24h give you that information. Or...is it creating an artificially extreme environment that is causing the errors in the first place?

 

[Zensal]

I hear you, but Prime puts a much greater stress on the CPU in terms of heat than it will actually ever likely see, hence making it more likely to produce an error then under ordinary conditions. That is what I am wondering about. If your cores are running at <50%, as they most likely usually do, they are running cooler, more stable, and they may never produce that error that they would at 100% for 24 hours when they are hotter and under artificially high stress that will never be produced under ordinary use.

Take the car analogy. Is it necessary to run your Taurus at 140 mph for 5 days straight in order to say that it's ok to pass at 80mph for 1 minute?

Your car analogy is a poor one for various reasons that I won't go into, but suffice it to say that your computer is not a car, nor do we use them like we use cars.

Your processor runs at full load more often then you would think. Things like loading a web browser, starting/playing a game, media encoding, Flash games/movies, etc. All these things can use the full extent of your processor for small or large amounts of time, and all can stress different parts of your processor or MCH/IMC. How do you know your RAM overclock is safe if it's never been filled up? How do you know that your processor won't return bad results to your game if you haven't tested it?

And the worst result would be data corruption. Windows is using your computer all the time to adjust and move data around. What if, because of your bad overclock, it flips just a bit or 2, and now your computer won't boot. Or opening IE now crashes your computer. Or moving your music to another place corrupts all of it. It's happened to me and many others.

If you really are somehow against running Prime for 24 hours, try the IBT. It seems to error much faster and I've been considering taking my burn-in test down to 12 or 6 hours using it, because it seems to error out so much faster and more reliably.

 

[fffblackmage]

I only test with Orthos for maybe 10 minutes or so only. I'll know the overclock is unstable if I get crashes or other weird problems.

And the worst result would be data corruption. Windows is using your computer all the time to adjust and move data around. What if, because of your bad overclock, it flips just a bit or 2, and now your computer won't boot. Or opening IE now crashes your computer. Or moving your music to another place corrupts all of it. It's happened to me and many others.

Hmmm... It somehow never occurred to me bit flips could occur, but that does sound rather likely. I have yet to have boot issues from a corrupt os, but I have caught some files that failed crc32 checks. I''ll certainly have to keep this in mind from now on.

 

[motsm]

I can't say I really feel the need to get into a debate about stability, as it's a rather slippery, and subjective slope to where people want to draw the line. I just say use your best instinct, whatever has been giving you stable results through the years surely can't be invalidated by someones post here.

Personally, I use Intel Burn Test and have found it really doesn't take long to get a read on how stable your system might be. Plenty of people who can do 24 hour runs of Prime fail IBT within the first few minutes. I bet the vast majority of those people don't care about that either, they passed their own standards, maybe years ago on that same overclock, and likely haven't run into issues. That is why I just see it as a slippery slope, who's to say your system isn't stable if you can't pass 30 minutes of IBT, when you already have years of stable use under your belt under any and all other conditions.

I suppose there is a little something to be said about passing hours of IBT, but I have never needed to. My system has been problem free from any issues related to an overclock from doing IBT runs minutes long, rather than hours.

/2 cents.

 

[thefreeaccount]

Or...is it creating an artificially extreme environment that is causing the errors in the first place?

IMO, it's a good question. Does anyone really know how error rate scales with heat and processor load? Is it like a 1 / 1B chance of an error at 100% load goes to 1 / 10 * 1B errors at 90% load? Or is it more like 1 / 1B * 1B at 90%?

 

[bunnyfubbles]

the friends of mine who don't go the extra lengths to test for stability are always the ones to have problems such as games crashing when mine rarely, if ever, do (when we're playing in the same games).

At the end of the day how hard is it to leave a stress test run over night and then while you're at school/work?

From what I've seen, people who don't like the 24/7 tests are those who can't achieve a dream overclock stably for such a length of time. Don't think with your e-penis.

 

[ShawnD1]

Hence you will never see those kind of temps, or need those kinds of voltages that we move up to in order to get the holy grail of 24h Prime stability.

It's not like Prime is hard on the processor. My Athlon II X4 620 passes Prime Small FFT testing at stock speed when it's undervolted to 1.050V (stock is 1.300). My E6600 passes when overclocked by 400mhz and undervolted by 0.1V at the same time. Being prime stable is like the most basic test possible. If it can't even pass Prime or Linpack, games like World of Warcraft will make your system crash a lot.

during an important clan match when you really don't want to crash out.

I remember lowering my overclock by 100MHz and setting my RAM to stock speed just so I was absolutely certain I wouldn't crash in World of Warcraft. Crashing in a raid is like skipping work or skipping school; you'll eventually get kicked out.

 

[Spikesoldier]

so you wouldnt mind if you are in the BIOS and set CPU vcore to 1.35, yet a value of 2.25v gets programmed into the EPROM because a 0 got turned into a 1 somewhere down the line?

would you like windows to automatically download updates and corrupt your OS with updates while you are asleep?

would you like to download a 8 gig file only to find out when you try to open it that it has been corrupted?

you see where im going with this.

yes 24/7 stability is important; everyone can get a CPU to post at a certain speed, others can get it to the desktop, but its not a real OC until we get screenshots of prime or another comparable burnin running for a period of time that casts no doubt that the system is stable.

 

[frostedflakes]

I'd rather test the worst-case situation, that way I'm positive anything I do on my computer will be stable. IMO you don't mess around when the potential problems are data corruption and stuff like that.

Plus if your CPU is just barely stable in the winter when it's cool, what do you think will happen in the summer when things get warmer? That might be enough to push it over the edge. You always want a bit of a buffer for stability.

 

[BD231]

Don't listen to the people who like to be assh*les around here they just hate themselves.

Depends entirely on what you're doing, as you've stated CPU's are hardly EVER put under full load by today's standards. If you're not crunching, Prime is not a big deal if normal use is not effected by your overclock. Visual clues would be all over the place if the problems that come from an unstable overclock that people are referring to in this thread are really giving your system issue's.

Some people around here we'll rant to no end about file corruption like they've never had to re-install an OS on a non-overclocked system for any reason. That's complete BS and you don't exactly want to scare yourself away from maxing out your overclock just because it MIGHT do something you don't like. Computers can be problematic for a multitude of non cpu related user actions so it becomes a senseless argument in that multi-core CPU's downclock themselves a better part of the time the computer is on anyway.

Longevity of a chip due to overclock is not compromised due to this fact alone so the risk factor in maxing out a CPU these days is all the more safe (unless you're using an Intel chip, in which case voltage/degradation is a bit more of a concern).

Personally I feel I'm smart enough with a computer to realize when it's not working correctly, if you feel the same way, then the associated risk is no different than a person downloading a file via torrent. You might find yourself a little trojan, you might not.

If you don't see crashes while gaming or using your OS/typical software's, chance's are anything else you throw at your computer is going to work just fine *FOR YOU*. Not even encoding suffers to terribly from an unstable overclock so there's really not a whole lot of things you need to consider outside of the fact that your computer works with the programs you throw at it.

I don't waste my electricity so crunching is out of the question.

 

[ShawnD1]

Some people around here we'll rant to no end about file corruption like they've never had to re-install an OS on a non-overclocked system for any reason.

That's because overclocking really is 99% of the reason anyone would ever need to reinstall. School and work computers run the same copy of Windows for 5+ years and it never seems to screw up. Never blue screens. Never freezes. My 80386, Pentium 1, Pentium 2 Celeron, and Athlon had the same copy of Windows installed for their entire life span.

 

[BD231]

A registry file could be marred by many things shawn, power outage, trojan, botched uninstall that corelates with other programs. Windows would not have such a bad rap were it not for this truth. The problems people are speaking of in this thread severely impare a computer, to the point where simple tasks such as gaming would not be possible without a crash or some form of implication.

 

[TJCS]

The last thing you would want is losing valuable work due to an unstable system. Prime95 is by no means an absolute test to stability (I have passes hours of it, yet computer crash in video editing sessions), but rather it gives you an idea whether your is PC can remain stable if pushed.

 

[TJCS]

There are many great free benchmark tools on the web today, and I don't see why you wouldn't want to make good use of them if they are readily available. Instead of betting everything on prime95, you should be trying a multiple of different tests to check your system's stability.

 

[manimal]

Silent data corruption, bitrot, etc and so forth...

 

[ShawnD1]

A registry file could be marred by many things shawn, power outage, trojan, botched uninstall that corelates with other programs.

This stuff is surprisingly rare. As an example of how strong a computer is, the computers in my design class are turned off by instructors switching the breaker on the wall. They do this so people stop looking at facebook and surprisingly the computers still work after all these years. The computers are all HP Pentium 4s running Windows XP.

The virus problem is also a lot less of an issue than people think. The average computer idiot does stupid things like run iloveyou.exe, but average computer idiots also buy Norton and McAfee.

I haven't seen an install or uninstall destroy Windows so i can't comment on that.

 

[Cattykit]

As I have been overclocking and testing things out here I have wondered how valid the practice of stability testing with Prime is for real-world use. I mean, under real world situations you will never, ever see sustained, 4-core 100&#37; utilization for extended periods. Hence you will never see those kind of temps, or need those kinds of voltages that we move up to in order to get the holy grail of 24h Prime stability.

Under regular computing conditions for the average user, what is the longest you may possibly see 4-core 100% utilization in situations like gaming, encoding, compression/decompression? Maybe 1-2 minutes? If so wouldn't Prime Small FFT for 15-30 minutes be more representative of the maximum severe stress a CPU might see? Or maybe the Blend test is more representative of real world usage. It seems to me that doing this is like buying a new car and running it at 140 mph for 5 days before saying it's ok to occasionally run it at 80 mph when passing another car.

So I ask, is there any rationale for funning small FFT 24h? Is there a good correlation with real-world stability or is this setting the bar way too high?

Hours and hours of 92-100% 4-core load is rather common and that was I was only encoding several min. of h.264 clips on 3.7ghz I7.

Anyway, the point is not about how you'll use the system in real-life but about how stable the system would be in extreme conditions. It's about making sure it never fails due to overclocking issues. Even in real world-usage isn't set to be intensive, there'll a brief moment where CPU ticks and your whole system crashes. By Bullet proof testing with Prime or Intelburn would let you know such things won't happen.
Also, passing Prime, Intelburn several hours in default setting doesn't mean much. You gotta let it run in higher priority with maxium RAM. So many times, I see people saying how their system passed Prime for 10 hours only to witness system crashes from here and there.

 

[Ben90]

Your processor runs at full load more often then you would think. Things like loading a web browser, starting/playing a game, media encoding, Flash games/movies, etc. All these things can use the full extent of your processor for small or large amounts of time, and all can stress different parts of your processor or MCH/IMC. How do you know your RAM overclock is safe if it's never been filled up? How do you know that your processor won't return bad results to your game if you haven't tested it?

And the worst result would be data corruption. Windows is using your computer all the time to adjust and move data around. What if, because of your bad overclock, it flips just a bit or 2, and now your computer won't boot. Or opening IE now crashes your computer. Or moving your music to another place corrupts all of it. It's happened to me and many others.

Quoting this because I dont want to type it, and its dead on.

Also prime isnt that great of a stress program... Im:
24 hours prime stable
24 seconds IBT unstable

 

[LoneNinja]

I believe it is important(though I usually only test for 3-5 hours) Couple years back I overclocked an Athlon X2 5000 BE to 3.2Ghz, I believed it was stable like that for months and the occasional program issue was just Vistas fault(probably didn't even crash a program weekly). The games I was playing at the time worked fine too, until I got Left 4 Dead. It took me days of trouble shooting, but no matter what I did Left 4 Dead wouldn't run for more than 30-45 minutes at a time, finally discovered it was my overclock. Decided to run Prime95 for the first time since I assumed it was stable, I wasn't able to pass more than an hour even after upping the voltage. Dropped my cpu to 3.1Ghz and never saw a problem again and was prime95 stable as well.

Just my little story, as to why I always check Prime for a few hours now.