Is it really stable?

katelin · May 31, 2003

OK, I hope I won't get lectured at for posting this here, but it is SETI related since all my PCs are running SETI, and I do want them to be 100% stable. So be quiet Greg.

On my 2100+ PC that I run at 186x12, it runs SETI for about a month now without crashing. And I checked SETISpy logs, all of the WUs so far ran to completion (>99%). But when I run Prime 95 (I read somewhere that it's the best for testing stability), Prime 95 crashes always after 6 or 7 hours. Everything I run on this PC runs fine. So my question is, should I really care that Prime 95 crashes after running 6 hours on my PC. I usually see this error:

FATAL ERROR: Rounding was 0.4999959469, expected less than 0.4
Hardware failure detected, consult stress.txt file.

Thanks!

TC · May 31, 2003

The way I look at this is if 99% of the programs you normally use don't have a problem, but this particular program you don't normally run at all has a problem - why worry about it? In fact Prime is one of those programs I have tossed in there with the likes of 3D Mark. I've seen 3D mark crash stock machines from the likes of Dell and IBM. Makes me wonder if it's not so much the hardware as the software.

DeltaSierra · May 31, 2003

I don't know diddly-squat, but I sure liked TC's answer!

aftermath · May 31, 2003

iv had 3d mark crash too at stock too. realy through me that did.

rogerdugans · May 31, 2003

I had one pc that ran fine for weeks at a given overclock but would crash every time I ran 3Dmark on it; I ran prime and some other hardware stress programs on it with no problems.

My theory is that if any one stress program causes a crash and yet 2 or 3 others don't then it is probably something in the software.

I say try a few more stress apps and double check.
It is probably fine but that way you can be worry free

psyshack · May 31, 2003

Intresting..... prime can give me fits running by itself at times,,, then it will run for days.... I like to take a machine and just hog it down for afew days.... fire seti up along with sandra burn, cpu burn,,, then throw 3dmark at it and maybe even prime... if a machine can do all that and keep chuging along... its good to go!!

psy

katelin · Jun 1, 2003

Thanks for the replys. Those are just the reassurances that I needed. I'll just leave it alone then since SETI and everything else seems to run just fine on it. I do notice that the CPU temp runs about 3-4 C hotter when I run Prime 95 as oppose to SETI. Which lead me to think that SETI wasn't stressing my system hard enough. I'm going to try Prime 95 on my Dell and see what happens. I might also lower my overclock on the 2100+ and see if Prime 95 still crashes.

LandShark · Jun 1, 2003

TC said:
The way I look at this is if 99% of the programs you normally use don't have a problem, but this particular program you don't normally run at all has a problem - why worry about it? In fact Prime is one of those programs I have tossed in there with the likes of 3D Mark. I've seen 3D mark crash stock machines from the likes of Dell and IBM. Makes me wonder if it's not so much the hardware as the software.

same here. i've seen machine runs 3DMark/SETI/CPUBurn-In/etc. for days w/out crashing, but would crash on Prime95..... therefore, if i could throw anything in it w/out crashing (either real world game or stressing program) but just 1 or even 2 cpu stressing programs, i wouldn't worry too much.....

'cos at the end, we should all take note that it's not 100% be the hardware problem!! it could also be software problem too!!

Greg M · Jun 1, 2003

katelin said:
So be quiet Greg.

<Eeyore Voice On>
Thanks for noticing me.
<Eeyore Voice Off>

Zzzzzip.

katelin · Jun 1, 2003

Re: Re: Is it really stable?

Greg M said:
<Eeyore Voice On>
Thanks for noticing me.
<Eeyore Voice Off>

Zzzzzip.

LOL.

Greg, don't take that the wrong way. I do value your opinion very much. I remembered when you lectured me once for posting some non-SETI related post in here. I still mad at you for that

Hey, can you crunch under my name for a week so I can catch up to Cuper? Hehe.

Mictlan · Jun 2, 2003

If I recall, there was an article in the main page where they were looking at this problem in particular. Seems that Prime95 is really being stretched by the nowadays hardware. Some people were complaining thatt heir stock PIV Northwood wasn't able to finish any Prime95 test, so someone came with a fix (really don't remember if it was the software developers or a third party). SO why don't you check in the main page or drop an e-mail to Ed to see if he can dig this one for you?

Cy · Jun 2, 2003

I'm late to this party but I'd agree with what has been said. If only one prog has a problem then it is prolly software related.

Cy

Prime95 · Jun 2, 2003

I hate to be a party pooper but....

If you cannot pass prime95 torture test, then you should reduce your overclock. Seti@Home performs calculations very similar to prime95 (floating point Fast Fourier Transforms). There is a chance that your Seti calculations are incorrect and that is not helpful to the Seti@Home project.

If you weren't running a distributed computing project, then I'd say it is OK to ignore the occasional prime95 error.

katelin · Jun 2, 2003

May I ask who you are and what your background is with Prime 95?

Prime95 · Jun 2, 2003

I wrote prime95 and run the GIMPS distributed computing project.

FWIW, GIMPS has an error rate of about 2 to 3%. We do double and triple-checking of all submitted results so we have a fairly accurate idea of how many bad results we get.

Seti's error rate will be less than this as the WU is of much shorter duration. All it takes is one error during a 2-week run to produce a dud GIMPS result. One error in a 2-week seti run would produce 13 good results and 1 bad result assuming one WU per day.

There are some differences between seti and prime95 which may mean seti will run fine even if prime95 generates errors, but there is no way to know for sure. For example, prime95 runs larger-sized FFTs, so it probably puts more stress on your memory and caches than seti, but less stress on the FPU. If you're on a P4, then prime95 is all-around more stressful than seti as prime95 is heavily SSE2 optimized (or maybe seti is too - it wasn't last I heard)

Prime95 · Jun 2, 2003

One more thing, don't be lulled by the 2-3% error rate (thinking I'll leave the overclock as is if I produce 97 good results out of a 100). That 2-3% is over all participating computers. The error rate for overclocked machines is higher although I cannot quantify it.

GIMPS has participating machines that never produce good results. My personal overclocking results are mixed. I had a Celeron 300@450 and Celeron 633@950 and never had an error in 4 years. My P4 1.6 was in the 2300-2400 MHz range for 6 months, but I kept getting occasional errors (and bad results). I'd reduce the OC 1MHz at a time without improvement. Finally, I just whacked it down to 2133 and haven't any trouble since.

katelin · Jun 2, 2003

Prime95 said:
I wrote prime95

Not to doubt you, but if you would point me to a website that contains your name and an email address that I can send some questions to I would appreciate it.

LandShark · Jun 2, 2003

hi Prime95, glad to have the creator of the program in here w/ us.

but, i do have a questions for you, i understand and 100% support that IF a machine is turning in bad WU back to Berkeley, it's totally a waste and useless, and what's the point?! however, in certain add-on program, like SETIQueue, it has some sort of result validating built-in that check bad WU result and reject them. you can see that in my SETIQueue's history. usually it could be hardware (o/c), or software (program/client corrupted). people either downclock the o/c or reinstall the program (client or addon) normally fix the problem.

thing is, i've seen machine turn in fine result (didn't pick up by SETIQueue) and fail Prime. yes, it could be SETIQueue isn't good enough to detect error.

however, i DID see/test in person stock name brand machine (not gonna name names but it's a household brand!) that fail nothing but Prime....?! that's the main reason i'm not fully support 100% software prefect (non error/failure/crash) on any machine 'cos there is a certain degree (even 0.1%) of software related failure instead of hardware related.

just my 0.02

TC · Jun 2, 2003

I also played around with Prime about 2 years ago. I was getting errors on a plain old Dell laptop. That's when I decided perhaps it was likely a software issue.

Prime95 · Jun 2, 2003

katelin said:
Not to doubt you, but if you would point me to a website that contains your name and an email address that I can send some questions to I would appreciate it.

Go to http://www.mersenne.org , there are email links there to woltman at alum.mit.edu

Is it really stable?

Member

Senior Seti Addict

Member

Member

Linux challenged Senior, not that it stops me...

Member

Member

Super Shark Moderator

Member

Member

Senior OC BOINC User

Senior Member

Registered

Member

Registered

Registered

Member

Super Shark Moderator

Senior Seti Addict

Registered

Similar threads