Questions on Prime95

Quote:

Originally Posted by Belial 

Hi. So I'm running prime95 blend to test cpu overclock. I try to do maybe 12 hours, if I have to use the pc i just set priority down from 10 to 1 (better to do 8 hours of priority 10 than 2 more of p1 than just 8 hours, right?).
2 questions:
1. How to understand the results log. My PC will sometimes crash from an unstable overclock, but I can't tell where it crashed at. Like take these results:

Code:

.............not including 6 hours previous of the test
[Wed Nov 07 00:56:43 2012]
Self-test 1920K passed!
[Wed Nov 07 01:04:19 2012]
Self-test 192K passed!
[Wed Nov 07 01:09:30 2012]
Self-test 192K passed!
Self-test 192K passed!
Self-test 192K passed!
[Wed Nov 07 01:20:57 2012]
Self-test 2240K passed!
Self-test 2240K passed!
[Wed Nov 07 01:26:14 2012]
Self-test 2240K passed!
Self-test 2240K passed!
[Wed Nov 07 01:37:16 2012]
Self-test 240K passed!
Self-test 240K passed!
Self-test 240K passed!
[Wed Nov 07 01:43:37 2012]
Self-test 240K passed!
[Wed Nov 07 01:52:46 2012]
Self-test 2400K passed!
Self-test 2400K passed!
Self-test 2400K passed!
[Wed Nov 07 01:59:19 2012]
Self-test 2400K passed!

So did p95 fail on FFT 2400k, or the 'next' value? How do I figure out what value it failed at?

I can't tell, but when you're looking at the log, look for the first occurrence where there are less than 8 in a row (for me, it would be 4 due to 1 thread per core).  So let's say you're looking over the log to see where it failed and all you see are 8 a piece and then suddenly you see a point where it's 7 a piece.  The first time you see 7 a piece is the FFT it failed on because the previous one passed just fine due to all 8 threads passing.

Does that make sense?  I'm a little bit scatterbrained due to being way out of my routine.

Quote:

Originally Posted by Belial 

2. Is it possible to pause/resume testing? I tried that - on the tray icon, I hit 'stop' to pause it, and then I hit 'continue", and I get this:

Code:

November 8 22:26 Worker Stopped
November 8 22:26 Waiting 5 seconds to stagger worker starts.
November 8 22:26 Setting affinity to run worker on logical CPU #X
November 8 22:26 No work to do at the present time. Waiting.

What is this?
Thanks.

That's because the Stop option stops it.  That's why it's called "Stop" instead of "Pause".  :)

There's no need to understand the rest of Prime95's log entries here.

Quote:

Originally Posted by Belial 

H
On a side note, I always try to set priority to 10, except when i am using the computer for light browsing. It seems to me it detects instability better and quicker that way, I just set it to 1 when i must use the computer and i feel its better to keep a test going at p1 instead of stopping it.

Weird.  I've been told the opposite:  never use the computer while testing its stability because everything you do steals CPU cycles.  So, I take it a bunch of steps further and I disable everything that would happen automatically, including the screensaver.

Based on your results, I no longer know which way is best.  I mean, the person I learned this from said he could get Prime95 to last longer by using the computer while running Prime95.

Quote:

Originally Posted by Belial 


No I have absolutely no idea what you are trying to say here. My sig rig is correct, I am running a quadcore phenom.
I have noticed that in the log, sometimes the 4 cores/workers dont finish the tasks at the same time and get a bit out of sync. Is that a problem? I imagine each core might be working slightly differently and after enough tests they get slightly out of sync.

Oh, that's normal.

Quote:

Originally Posted by Belial 

less than 8 in a row... less than 8 what? What row? I dont see any rows. I truly have no idea what you are trying to say, and im trying very hard here.

Edit:  Oh, we're on the same page after all.  My bad.  The only way the log will reflect a failure is if Prime95 itself has a failure, such as one of the workers failing.

The phrase "in a row" means "consecutive", or "one right after another".

Look at the log and count how many matching FFTs you see.  You'll see 8 that are together, and then the next FFT with 8 more entries.  Why are there 8?  Also, why are there time stamps in between some of these groups of 8?

Here's a section from one of my logs that shows the point of failure (and as I'm opening it, I'm reminded of what you would see when a worker fails - I completely forgot about this):

[Tue Jul 12 18:07:15 2011]
Self-test 1120K passed!
Self-test 1120K passed!
Self-test 1120K passed!
Self-test 1120K passed!
[Tue Jul 12 18:22:22 2011]
Self-test 32K passed!
Self-test 32K passed!
Self-test 32K passed!
Self-test 32K passed!
[Tue Jul 12 18:37:55 2011]
Self-test 1200K passed!
Self-test 1200K passed!
Self-test 1200K passed!
Self-test 1200K passed!
[Tue Jul 12 18:53:39 2011]
Self-test 48K passed!
Self-test 48K passed!
Self-test 48K passed!
Self-test 48K passed!
FATAL ERROR: Rounding was 0.5, expected less than 0.4
Hardware failure detected, consult stress.txt file.
[Tue Jul 12 19:08:39 2011]
Self-test 1344K passed!
Self-test 1344K passed!
Self-test 1344K passed!
[Tue Jul 12 19:24:11 2011]
Self-test 64K passed!
Self-test 64K passed!
Self-test 64K passed!
[Tue Jul 12 19:39:31 2011]
Self-test 1536K passed!
Self-test 1536K passed!
Self-test 1536K passed!

etc.

Judging from this, my CPU failed the 1344K FFT because at that point there are only 3 while the previous FFT (48K) has all 4.

So, this is how you'd know based on what the log says.

Quote:

Originally Posted by Belial 

There's nothing conflicting with what you are saying, and me. I totally agree, yes, it's true, you shouldn't be doing anything else while stress testing. Theoretically, you should never have to set priority 10, because you just never do anything else and disable stuff like screen saver (which by default has a higher priority than prime95).
But what I'm doing, is that I think running 16 hours of prime95, where maybe 4 of those hours you were also browsing the internet, is better than doing just 12 hours of testing. That's what I'm doing here.

All I can say is, do what seems to work best.  I have to say here that I learned this in the days when the Intel Core2 CPUs were still the hottest thing on the market. lol  So, maybe it's ok to do other things while Prime95 is running - I'm not sure.

Quote:

Originally Posted by Belial 

No it's not really okay to run other things when doing prime95. It's just my opinion that's it's better to do 12 hours dedicated + 4 hours with light browsing = 16 hours of prime95, than just doing 12 hours dedicated of prime95. I suppose ideally, you just leave your computer the hell alone for 24 hours, but... you know, internet and stuff.

Yeah, 24 hours without my computer would be difficult for me.

Quote:

Originally Posted by Belial 

I was hoping there was a way to figure it out when your computer just crashes, since that seems to be a more common way for instability to manifest, at least in my experience. So i guess there's no way to tell, you just got to figure out what FFT comes next in the cycle and suppose that's what you failed on.

I never thought about this until now:  use Event Viewer.  Then you can try to match up the Kernel power error time with the Prime95 log.  I don't know if this will work out to give an idea of which FFT was the culprit, so it's really just a theory right now.

Quote:

Originally Posted by Belial 

i wish the results log was a bit more user-friendly... tells you when each test starts and ends, maybe told you about unexpected crash, a different log for each instance. This problem is over a decade old with constant updates and still has the same log has 10 years ago.
Hey I got a quick question - if you run 2 molex 12v fans off the same PSU supply connector, like daisychain them, will that lower their output? or will they still blow at full speed?

I don't know, but I think they would still run at full speed if they are 12V fans because they'd be using the +12V rail.  If they're 5V fans (unlikely), then there'd have to be enough power on the +5V rail (and there's usually plenty, such as 100W for your PSU).

Here's an article that assisted me a little bit with this answer because I hadn't thought before about where fans get their power:  http://www.overclock.net/t/722251/on-psu-voltages  Thanks to this, now I remember that all of my fans say "DC12V" somewhere on them, and now I know what it means.  :)

Edit:  Oh, I don't know much about fan controllers, so I don't know if having two fans on one channel would result in a reduced maximum fan speed.