Prime95 Poll

[Sovereign]

How long do you let your CPU burn? I understand people can get "hot" (pun intended) when talking about the validity (or lack of validity) of stability testing with these programs. I used to be a "Damn Prime95, my computer runs fine!" person. Truthfully, the only problem I have had with my machine was a mobo failure that the maker (MSI) took full responsibility for and basically admitted it was a QC slip that allowed a board with a bad PCI lock out of the factory. Many of my friends subscribe to this view, which makes logical sense in the following way:

If my system always is able to do what I use it to do when overclocked, and I don't experience any problems, there isn't a problem.
That is, until your system flames out. There's no guarantee that this will happen. In fact, many will never push their system as far as Prime95 does. People complain about "synthetic" benchmarks that reveal very little if anything about how a computer will perform when playing an actual game or running a real application. In a way, Prime95 represents a synthetic system load. No game or application that I know of short of some distributed computing projects (such as F@H or SETI@Home) push a CPU this hard. Even still, a computer can be "Folding Stable" but "Prime unstable." So Prime still requires more cajones than Folding...

If you're not running those applications and you've transcoded movies and played games, you'd think your system is fine in its overclocked state. However, the Prime people (and I am increasingly one of them) come back with this:
If your system can't survive this load, then it is more than likely that a lesser load will be the straw that breaks the camel's back. Of course, Prime95 could be the only straw that breaks the camel's back, but it is possible that some "lesser" task will bring your computer to its knees. However, if your computer can survive Priming then it will definitely survive everything else you throw at it.

I think some people dislike Priming or other testing like it because it takes so much time. With some hardcore overclockers recommending 24+ hours of running an application that chews up all your cores and makes your computer virtually unusable, it's understandable why people would grow impatient. "I didn't build this computer to watch it error-check itself all day!" they say. Some people say that 8 hours is sufficient, that anyone else is crazy. Others opine that tests often fail in the 16-20 hour range and those failures will NEVER show up earlier. Generally, very few people will argue that 24 hours is too little.

Others dislike Prime95 because it tells them something they don't want to hear: your OC is wrong or too aggressive and won't hold.

What do you all do? My Q6600 is currently Priming at 3GHz and has been going for seven hours strong. I personally find 24 hours to be excessive...

[ibleet]

I typically prime in the 1-4 hour range for stability.

If I stress my cpu to 100% using Prime for an hour and it throws no errors, I figure it can pretty much hang stable with any application or game I throw at it.

[DMB2000uk]

Well usually I only run prime (ORTHOS stressprime edition) until I am beginning to reach the limits of an overclock, or am nearing my target ghz.

As for extended stability testing 8 hours/overnight is all I ever do. If my machine can handle 100% load for that long, I'm not likely to leave it running loaded 100% for that long again, so for everything I do it is stable.

The trouble with deeming a CPU stable if it can just fold is that folding may not have as strict error checking as prime95, so it could still be throwing up errors but being within the tolerance of the folding project, leaving stanford with incorrect results. It's better for everyone if overclocked CPUs are at least checked overnight with prime95 (or equiv) if folding is the aim.

I'm the kind of person that is obsessive compulsive about things like this (not quite to an extreme degree though). If my CPU isn't stable then I don't want to run it that way.

Slight tangent, but could running a CPU that isn't 100% stable for long periods of time cause some damage to it?

Dan

PS. You shouldn't use your computer AT ALL when it is stress testing, and switch off screen savers too, as any CPU cycles that aren't being used by prime95 aren't pushing your CPU hard enough (I know you can't catch every single one, but using it drastically increases the missed cycles)

[Sovereign]

No, if the CPU is unstable it should crash/err and Prime95 will terminate. Thus, even if the stress were to cause damage, the error would prevent the stress from continuing and the CPU would fall back to idle. Heat, on the other hand, can kill even a "stable" CPU. Heat and excessive volts are an overclocker's worst enemy. Also most modern chips can take quite a bit of abuse before failing. Finally, Intel's Thermal Protection Circuit should save your CPU from massive overheating (although it may be damaged and only work at stock or lower speeds afterward; I've never had this happen or heard of it happening but it's entirely possible that you can "partially" damage a CPU).

[DMB2000uk]

I didn't mean if it could die during stress testing, but if an unchecked error/nonstable CPU could be in some way damaged by running too fast/with not enough volts.

Dan

[Sovereign]

My belief is (and I'm not entirely sure) that if a CPU is going to be damaged in any way by stress tests, the CPU is faulty or overclocked to an extreme (but then I think it still has more to do with over-volting and heat than the stress test; the test just causes the heat which really could come from anything). Otherwise, the stress test should err and the CPU will return to idle as the stressing program shuts down/pauses due to the error.

Again that's just my opinion, I'm not sure of it so please don't quote me on it ;)

[ibleet]

No, if the CPU is unstable it should crash/err and Prime95 will terminate.

In my experiences, if the stress test fails, it typically, if not always dumps on one core first, and the other core continues to run the test.

[stopthekilling77]

you should add a "i have no friggin clue" button to the vote thing...

[Zertz]

Prime for about 20 minutes, then F@H.

[ibleet]

Whoever says they run it for 8 hours or longer is lying.

[skier]

i just do superpi 32million, so like a 45min test

[martini161]

i just do superpi 32million, so like a 45min test

well it depends on your system ;)
as for prime 95, when i first get a chip, before i start overclocking i make sure its a good chip by burning it in on prime 95 for at least 120 hours. (5days) its may be a little overkill, but if a chip fails i want it to fail while i can still RMA it to newegg (as i ussally buy oem chips) then when i start overclocking, if when i can boot windows at my target overclock, i run it for 12 hours and when it fails i mess around with the voltage and all that crap until i can get it stable. after that point (usaly around 3.5 ghz or so) i run every .1 ghz for 1~2 hours until i reach a point where it wont be stable with a voltage im comfortable with (anything more than 1.5 volts for a 24/7 overclock, 1.7 if im running it under water for a benching overclock) then i tone it down to the last highest thing i got stable with that voltage, then i run prime 95 for another 12 hours (unless its a benching run then i just boot and run the benchmarks and pray to god that they complete and i get screenshots before my cpu takes a dump ). yeah, when im overclocking its a multiple day ordeal

[ibleet]

5 days of Prime95 is not a little overkill, its downright INSANE!

[Sovereign]

Whoever says they run it for 8 hours or longer is lying.

Fail.

SpeedStep was on, so the CPU clocked down when the system became unloaded after the Prime run.

[ibleet]

You just ran it for 8 hours because I said that.

[Zelig]

8 hours (overnight) while testing overclocks, minimum 24 hours when I think I've settled on a clock for 24/7 use.

I generally reboot my PC monthly, so it needs to be damn stable for that entire time, regardless of what I'm doing with it.

No screenies, but I've forgotten about prime running in the past, and had it >100 hours.

[martini161]

5 days of Prime95 is not a little overkill, its downright INSANE!

not really, ive been pwned all to much by latent failures

[geokilla]

Waa, I'm the only one that voted for 20 to 24 hours....Maybe I'm insane..

Off Topic:

Is there a special setting I have to check when using Prime so that it'll show errors and stop Priming if it does encounter an error?

[Alathald]

Is there a special setting I have to check when using Prime so that it'll show errors and stop Priming if it does encounter an error?

Nope, if it errors it says it errored and quits...

I usually set mine to Prime about an hour before bed, check right before bed to see if it's errored (if so, I bump to volts or dropped the freq) and leave it to Prime overnight so that's what, 8-10hrs?

[Sovereign]

Is there a special setting I have to check when using Prime so that it'll show errors and stop Priming if it does encounter an error?

Nope, if it errors it says it errored and quits...

I usually set mine to Prime about an hour before bed, check right before bed to see if it's errored (if so, I bump to volts or dropped the freq) and leave it to Prime overnight so that's what, 8-10hrs?

Actually, I was told under Options to specifically set "Round-Off checking" or else some errors may be missed.

~ibleet: I can show you the file creation date to prove I ran the test before you said that