Prime95 and stability

[RGone]

:nod: :nod:

remember times of the old via 4in1 driver

 

Hehehehehehe. Just love that ole 4.17 driver. hEhehehehehe.

 

Sincerely, RGone...:angel: :shake:

[G00s3]

The last few weeks I've been trying to get my system stable for 36h prime. 24h I can pass everytime, but it never made it to 36 so far. But otherwise no crashesn, no blue screens, no nothing... But the stress.txt file that comes with Prime says something along the lines of, if prime fails, you have a problem somewhere, no matter if otherwise it runs for weeks stable without powering down. And that sort of stuck in my mind... I've been wanting to get a toaster but I somehow thought it not fair since I can pass 8h easily, 3dmark all the way, at 250x10 but somewhere after the 24h mark it mysteriously fails with a rounding error. I thought it would be cheating a bit knowing that it will fail eventually...

[THunDA]

Well.. It really depends on what your doing with your computer .. I personally run prime for 8-10 hours and call that stable for "ME" and what "I" do ( gaming ) on my computer.. It has always left me with a stable and crash free computer.. Although I dont remember the last time I ran prime recently.. but everything has been smooth and stable..

 

36 hours is insane man.. dont take offence to that but its just my opinion.. Of course if your somewhat competing to be 36h prime stable as someone else is then go for it.. thats the whole fun of OC'n ... keep making the goals to reach..

 

 

GL

Thunda

[G00s3]

None taken I'm not really competing or anything - it's just that I start prime, leave the house and usually I'm not back within 24 hours (I'm studying at my girlfriends place). Otherwise I'd never have found out, overnight would be plenty then . But it just bugs me, knowing it's not 100% stable but fails in one out of so many billion calculations. After my exams I'm going to work on my cooling a bit and then see how far I can go with 8h prime stability as the benchmark.

[Torzi]

@GOOs3: whoa man....you are using the puter for other things....right?

 

actually...I'm currently not. since I don't have an AGP card yet....and I do all my work on a laptop. But....gainward sent it out yesterday (hooray.....it was 3 months from warranty). I know as soon as I get it though...it'll open a whole new can-o-worms in 3D.

 

As far as prime goes....I don't know what happened, but it didn't error at my sig settings for 16 hrs before I stopped it. I was fiddling with mem timings w/ NVtweaker....windoze crashed (lil too tight/ambitious), rebooted, started prime..and it just ran longer than I've ever got it to run @ 250!??!

 

I'm using custom...1024 to 4096, not in place, 900MB, 30mins. Perhaps this is helping to burn in the ram/board?...still only about a month in use. I bumped tRAS 11 to 10 (best b/w), and raised Vdimm to 3.1, and where it used to crap out pretty early at that voltage...it seems to be humming along. Perhaps these hynix chips just like a slow burn, they weren't responding well to a 72hr 3.3V memtest #5 burn-in.I know prime is stressing them pretty hard though... They get pretty warm even w/ a delta screamer mounted over 'em.

 

I like the idea of SP2004...just stress testing, and good notification when it fails. I wish the author would/could add an auto restart/resume. It would be interesting to be able to compare how many times prime fails in a given amount of time, mean time to failure, deltaMTF etc. This would definitely help rule out/pinpoint certain causes. Perhaps I'll send him an email.

 

I'm just wondering if the community's general consensus is whether or not a system MUST pass that benchmark to be considered "stable", or at least stable enough to be "safe" (as in safe from Windows exploding with no survivors ).

 

IMO:

"stable"....well that depends on what you want to use your computer for. If it'll run everything but prime without errors, then enjoy the speed. I agree fully with RGone though....definitely keep backups, because there IS some instability in there somewhere, and you never know what might set it off without excessive tinkering.

 

however....I wouldn't mind being able to run some matlab sims on the machine, which in this case reqires the best accuracy/precision my system can provide. If prime is failing at all....then I'll definitely lower my clocks until it's stable before I'll run an app like that. (Yet another reason I'm loving this CMOS Reloaded feature).

[lordbane]

I backed down my FSB to 245MHz CPC ON guys, although I CAN do 250MHz CPC OFF and pass Prime95 Blend. I've still got enough MHz to get my Toaster though if I want it (2578MHz).

 

My thanks to everyone that gave insight into this thread:

 

Here is a summary of my experiences:

 

RAM Timings/FSB:

 

Try as I might, I could not get my own OCZ 4200EL to accept anything better than:

 

10*-4-4-3-2.5-11-14 (all rest on auto/aggressive settings) @ 245MHz - CPC ON @3.0v

 

* - I gained a little bit of bandwidth and a few 3DMarks going to 10 for the Tras from 11 and it seems to work ok, even though you guys suggest 11.

 

Even with CPC off, I could not get the memory stable in Prime95 past ~252MHz, but this might be a limitation of my chipset, and not the RAM.

 

The memory simply would not function if I tightened any more beyond that. It has something to do with the way that OCZ has configured the memory to handle high FSBs, and if you read the product page for the memory, it specifically says that. These are high quality chips with tweaked latencies to handle high frequencies, but they need relaxed timings to do so. They are not on par with the legendary BH-5 chips.

 

If I'd known what I know now, I probably would have looked for some good BH-5 based modules and worked with those.

 

C'est la vie...

 

CPU:

 

I can pass Prime95 torture tests at ~2600MHz @ 1.85v for extended periods (which is really lucky and good for air coolings, even though my case sounds like a vacuum cleaner powered up and I live in a cool climate [Canada]). Seeing as my best stable FSB is 245MHz, I chose a multiplier of 10.5x for a final CPU speed of ~2578MHz.

 

I weighed the opinions of you all carefully, and after much testing, i'm settling at the results above, and will not run my main installation of Windows XP beyond that point.

 

RGone made a good point about stability and the PITA reinstalling Windows and all your games can be, even with Ghost, etc. Corruption caused by too much tweaking is not always repairable. Thus, even though I have backups, I've gone down to what I know for sure is a very solid configuration. I will never boot into my main Windows XP installation on anything beyond this setup unless I have tested it extensively.

 

I've set up a second installation of Windows XP on my PATA drive which I can boot into for testing and tweaking. If I corrupt that installation, it's no big deal.

 

This is what I've learned so far (and might be a bit of advice for newbie O/Cers reading this thread):

 

Trying to run your memory too aggressively is asking for a corrupt BIOS or a corrupted Windows installation (I learned this the hard way), so I know better than try anything nutzo.

 

I use ClockGen and lower multipliers to test FSB, then use lower FSB and higher multipliers to test CPU stability.

 

Once you get errors in memtest or fail to POST, you've gone too far.

 

**IF YOU CAN PASS MEMTEST AT SAY 260FSB BUT GET ERRORS AT 262FSB (LIKE ME), IT DOES NOT MEAN WINDOWS WILL RUN PROPERLY, OR THAT YOUR WINDOWS INSTALLATION WILL SURVIVE A BOOT WITHOUT CORRUPTION AT 260FSB!!!!**

 

Booting into Windows at a lower FSB (I would say no more than X-10 FSB where X is the highest FSB that memtest would run stable on) is advisable. Then run Prime95 "Blend" test. If this fails, try more juice to the chipset/RAM or back off the FSB. If it passes, use ClockGen to up the FSB by a notch and rerun. Let it go for a while each time.

 

Prime95 is not the end-all-be-all of system stability tests, but it will give you a reasonable indicator of when you are pushing your FSB too far. You can still run games, boot into Windows and not have corruption but fail in Prime95. The hard part is knowing exactly at what point, you will get corruption/Wierd Things[TM] happening, and you want to avoid that point.

 

For me, 245MHz is the maximum FSB that I can use and not have Prime95 Blend error out (usually inside 15 minutes, randomly). The installation will run stable up to about 253-255MHz, and I don't crash, etc, even though Prime95 Blend fails. If I go past 257MHz FSB, bad things start to happen.

 

Booting at or adjusting my FSB to 258MHz+ FSB will result in disk corruption, even though memtest passes without errors up to 264MHz FSB, and the system will actually boot up at 261MHz FSB (but dies in a blaze of glory).

 

To any newbies: do what I did, and make a spreadsheet with a list of frequencies, timings, scores, cpu speeds and FSBs. It's a lot of work, but at least then you can take a scientific approach to trying to find at what points your system will fall down.

 

I guess the ultimate goal of overclocking is to run your system just inside the fine line between stability and instability (and get bragging rights!).

 

:nod:

[Torzi]

To any newbies: do what I did, and make a spreadsheet with a list of frequencies, timings, scores, cpu speeds and FSBs.

 

Even a pencil and a pad will do (which is what I use).... important thing is to record what you've tried, and what the reslult was. Not only does this keep track of what you've tried....but it also keeps you from repeating your mistakes. Once you get to a point where you're stuck, look back at your notes to make sure you don't repeat your mistakes.

[Mevunky]

Originally posted by G00s3

The last few weeks I've been trying to get my system stable for 36h prime. 24h I can pass everytime, but it never made it to 36 so far. But otherwise no crashesn, no blue screens, no nothing... But the stress.txt file that comes with Prime says something along the lines of, if prime fails, you have a problem somewhere, no matter if otherwise it runs for weeks stable without powering down. And that sort of stuck in my mind... I've been wanting to get a toaster but I somehow thought it not fair since I can pass 8h easily, 3dmark all the way, at 250x10 but somewhere after the 24h mark it mysteriously fails with a rounding error. I thought it would be cheating a bit knowing that it will fail eventually...

 

i fail after 16 or so hours, gonna do a reset soon. but is it prime, is it my setup, is it driver related, so many issues that make prime not a 100% sure test

 

i think my psu is overloaded seeing i see 12v as low as 11.5 and 11.6 all the time...

[lordbane]

Originally posted by Torzi

Even a pencil and a pad will do (which is what I use).... important thing is to record what you've tried, and what the reslult was. Not only does this keep track of what you've tried....but it also keeps you from repeating your mistakes. Once you get to a point where you're stuck, look back at your notes to make sure you don't repeat your mistakes.

 

While I was working I did the same thing, but I'm inputting my findings into a spreadsheet. It would be awesome if we were to all do something like that (i.e. make a spreadsheet with tests and what failed) and compared our work.

 

It might be a way to develop a safer, more empirical approach to tweaking up our DFI boards. It also might make it easier to understand which specific adjustments have the greatest effects as well as contribute greatly to newbies trying this stuff for the first time.

 

Being a newbie overclocker can be as dangerous as running through a minefield; they just don't understand everything before they dive in and judging from the posts here it's usually only the newbies that have a lot of problems with dead boards, etc. because they just pushed them too far (they didn't know any better).

[uwackme]

What is SP2004?

[Torzi]

Stress Prime 2004...

 

Just the 23.8 version of prime with a face.

 

http://www.dblsys.ca/sp2004/

[RGone]

...MR. lordbane!

you heard it first from lordbane> Being a newbie overclocker can be as dangerous as running through a minefield; they just don't understand everything before they dive in and judging from the posts here it's usually only the newbies that have a lot of problems with dead boards, etc. because they just pushed them too far (they didn't know any better).

 

I have been reading forums for the better part of 5 years now and suddenly a statement that so astounds me with insight. I am just blown into oblivion. Holy Happy Holly, you don't say? My gawd! Hehehehehehehehehe. " lordbane " you get 50 lashes with a wet noodle. :nod:

 

I am ever so thankful for my general knowledge of things and that when I first started to try and push my boards some 5 years ago; that my first inclination was to read and read some more. Then I took some very baby steps and began to attend forums and I was not a youth as I only started computering at age 51 and had enough sense to study before practice and then I met a few of the first bunch of overclockers using the KK266R Iwill Rocket boards. They took me under their wings and taught me as they sensed I wanted to know and knew that overclocking was a privilege and not a right and I have used skills they taught me; now these last 3 years and always I think of my mentors.

 

Many of you guys in here are good at mentoring and expend great amounts of energy to do such. For that I extend my thanks. Good day guys and good luck.

 

Sincerely, RGone...:cool: