What is really "stable" - a newbie perspective

[twilyth]

I'm not really a newb in general, but I am regarding overclocking. I've done some oc'ing in the past, but only in the 10-15% range I'd say.

I got more excited about it recently with a 1090T and xhair IV. I let the Asus EVO utility do the first pass. this got me up over 3.8ghz with a 16 multi and an NB speed of something in the low 240's.

I then lowered the NB to 235 and raised the multi to 17. CPU voltage was about 1.38 more or less (although I had thought it was lower for some reason). I can't remember the NB volts - probably 1.35 but I'm not sure.

I ran OCCT linpack for 3 hours with no errors. So I set the CPU loose on WCG. At first I got a few errors but not a lot and they didn't occur all of the time. Recently though it's gotten a lot worse.

I checked to see if other crunchers were getting errors on these work units and a few had but mostly, I was the only one. So I figured that despite my stress testing, the oc must not be stable.

That brings us up to last night. I figured I should really beat the piss out of the CPU, so I ran linpack PLUS prime 95 on the high-heat setting. I set it to run for 8 hours. I'm about half way through and there aren't any problems.

I did have a couple of crashes initially so I raised the CPU volts to about 1.39 (a shade under that right now). I set everything else to default. then, beside the cpu volts, I let the m/b set the ram volts to 1.65. I set the multi at 20 and didn't touch anything else. Final oc was 4013mhz.

So, my question is, how do you know when you have a truly stable oc. According to most of what I've read, the 3 hour linpack test should have done the job, but obviously it didn't.

Now, running for 8 hrs with both linpac and p95, if this sucker isn't stable now, I'm gonna be rma'ing some shit.

 

[niko084]

I generally run linpack for 36 hours and have my own limits of what kinda temps I'll accept.

For instance, nothing over 80c on my i7 860, that means in that 36 hours of linpack if my cpu breaks 80c, the clock gets dropped.

 

[erocker]

Playing games, doing the stuff you normally do on your PC without issue for a couple weeks. That's stable to me. Generally I start with more voltage then I need and slowly back down.

 

[cadaveca]

Hey twilith! I see you are slowly drifting over to my side of the "stability camp".


I've mentioned this elsewhere, but I'll pop this out here for you too.

I start with stock, and OS installed. First I test ram...that has it's own routine.

Then CPU.

First, I clock the cpu up by multi.

First is Linpack, standard test.

Second up up Maximum settings. I run this for about 2 minutes, then end it. This will free up more ram, so that the test is more accurate. I let this run for 10 passes(about an hour).

Then, I will run Prime95, and this takes about 9.5hours @ 3.6ghz, blend test, for it to go through all FFT sizes and start over again.

Usually by this point, I can call my system stable...usually ram/cpu, anyway. Failures along the way dictate what I do to bring back stability.

WHen it comes to crunching...I won't load any crunching app until I feel sure that the results I return are not invalid. The apps don't have to throw errors for every little thing...

Oh, and by the way...if you can pass Linpack standard, but not maximum, look at cpu/NB volts!

 

[ERazer]

i mainly use prime95 for 10hrs and anything over 70c i back down

 

[twilyth]

OK. Thanks for the input guys. I get the impression that linpack, NOT OCCT's version of linpack is the way to go. But since I'm in the middle of the second 8 hour test (with p95 having run continuously from the first test), what do you think about OCCT linpack with P95 at the same time?

The last test crapped out at about 7 hours, then there was probably 1-2 hour delay before I noticed. I kept P95 running though. I bumped up the cpu volts 2 notches so now I'm at 1.39375 I think - something like that. I did that through the EVO utility and saved to bios (which won't take if it crashes completely).

If I get through this 8 hour test now with the higher voltage, can I call that stable and go back to WCG? You know how much it grieves a cruncher not to be generating points.

Or . . . should I scrap the current test, get the Intel (???) version of linpack and do that on max for 3, 6, x hours?

edit - oh yes, forgot - temps are never over 47 - 46.5 last run was the max. (how's it hangin' Cadaveca?)

 

[cadaveca]

I use IBT myself. I have used OCCT in the past, but find it doesn't meet my personal needs. Good free tool though.

I would NOT run 2 stress tests at once. Only say, like linpack, and then a 3d test, with some audio, but never two tests that test the same part(multiple tests used to verify PSU integrity/case cooling, for me).


Oh, and things are good. Kids will be back in school soon, wedding anniversary next week(looking forward to going out with my wife...we haven't gone on a "date" in like 2 years, if not more. 4 kids, an no help here for us).

If your cooling is up to it, don't be afraid to give 1.45v. should be fine.

 

[twilyth]

I use IBT myself. I have used OCCT in the past, but find it doesn't meet my personal needs. Good free tool though.

I would NOT run 2 stress tests at once. Only say, like linpack, and then a 3d test, with some audio, but never two tests that test the same part(multiple tests used to verify PSU integrity/case cooling, for me).


Oh, and things are good. Kids will be back in school soon, wedding anniversary next week(looking forward to going out with my wife...we haven't gone on a "date" in like 2 years, if not more. 4 kids, an no help here for us).

If your cooling is up to it, don't be afraid to give 1.45v. should be fine.

What don't you like about running 2 tests at once? I thought that would be much more taxing for the CPU and therefore a better test - no?

I'm not familiar with IBT. Is it free? doesn't Intel have a stress tester? What about that one?

Thanks again. Glad to hear things are mellowing out - well, you know, at least in relative terms.

 

[newconroer]

Just wanted to add or reiterate that while an overclock may be 'stable' for 85%+ of the time, it's that 15% when the system crashes that can be detrimental. A normal system crash is no big deal(usually), but under more stressful conditions(such as an unhealthy overclock), a crash could easily result in RAM damage. Even today's DDR3 memory is not impervious to errors.

 

[cadaveca]

IBT is IntelBurnTest...a Linpack test program.

when it comes to cpu, two tests will then contend for cpu resources, leading to a lesser load than a single test. The tests we have at our disposal are far more work than most of us will ever need...so one is perfectly fine.

The summer has been fantastic...I love being a parent, and getting that chance to enjoy the small things with my kids. With them being 2-4-7-9(soon 3-5-7-9), we are kind of limited as to the activites we can do, but making them happy and them having fun is the most rewarding thing ever.

Even as I'm typing, my kids are all within about 10 feet of me. And it's like that every day, all day long...and I wouldn't have it any other way. For me, life can't get much better than it already is...but of course, perspective is everything.

 

[DanTheMan]

Not to get off topic but Cadaveca, you sound like a very classy guy, the world needs more parent like you. Someone who enjoys being around their kids, not only does it enrich your life but it mostly enriches theirs. When they grow up they become great parent themselves remembering what they learned and enjoyed doing with you. I have surely enjoyed my two (boy age 12 and girl - 18 months) with my boy I do boy scout stuff and he is well on his way of being an Eagle Scout. And I really enjoy my baby girl's smile saying daddy daddy when I walk in the door. I also would not have it any other way!

 

[twilyth]

OK, so I'll terminate occt and p95 and run linpack max for 2 hours and see what happens.

I wasn't sure about the contention issue but I figured at least it would keep the temps up rather than cycling them like occt does.

I'll report back later today.

Thanks again everyone.

 

[cadaveca]

Not to get off topic but Cadaveca, you sound like a very classy guy, the world needs more parent like you. Someone who enjoys being around their kids, not only does it enrich your life but it mostly enriches theirs. When they grow up they become great parent themselves remembering what they learned and enjoyed doing with you. I have surely enjoyed my two (boy age 12 and girl - 18 months) with my boy I do boy scout stuff and he is well on his way of being an Eagle Scout. And I really enjoy my baby girl's smile saying daddy daddy when I walk in the door. I also would not have it any other way!

My parents weren't a large part of my life. But they had me pretty much right out of high school. so I now kinda understand how things were for them in that situation. So it's very important to me that I give my kids everything they need...more so than the average parent. Everything I do ultimately has them in mind, as now that I am much older I can see how my own parents choices affected my own childhood. I have almost literally been to hell and back, so I think I have a good perspective as to what's truly important in life.

I get tonnes of comments from people when out in public with my kids, such that it's hard to not let it go to my head, so I do get regular reminders that I am doing a good job when it comes to my kids. My grandmother says "Sorry that your parents weren't there for you when you were younger. They weren't there to teach you much." My response was "Sure they did..they taught me what NOT to do".

But htat's enough O/T talk from me!

OK, so I'll terminate occt and p95 and run linpack max for 2 hours and see what happens.

I wasn't sure about the contention issue but I figured at least it would keep the temps up rather than cycling them like occt does.

I'll report back later today.

Thanks again everyone.

That cycling behavior is a big part of why I don't like OCCT. I find that I can get more load on the cpu using IBT or prime95...prime95 requires some custom settings, but that's no big deal. I merely change it from blend settings to the same FFT size, but more ram usage.

IBt is good becuase you can get it to load up all the ram you got...and with me having 8GB, that can be pretty hard.

 

[twilyth]

I couldn't figure out how to use linpack. I found executable for intel chips but that was it. So I fired up P95 again on the 2nd setting (mixed, max heat).

the machine had crashed in the meantime but kept the bios changes. i bumped the voltage to a little over 1.4 and restarted p95.

I'll let it run for 9.5 hours at least and see what happens.

edit: I'm not worried about memory since I rolled the nb back to 200 and am working solely on the multi and voltage right now.

 

[cadaveca]

I've attached the IntelBurnTest here for you.

It's so easy to use...one of my favorite tools over the years. I think ATITool is the only other I've used as much.

 

[twilyth]

I guess I should read more carefully. I thought that IBT and linpack were the same thing. Bad brain cells! Bad!!!

Thanks again.

 

[cadaveca]

IBT is just the interface to run the Linpack tests.

 

[freaksavior]

Playing games, doing the stuff you normally do on your PC without issue for a couple weeks. That's stable to me. Generally I start with more voltage then I need and slowly back down.

I do the same. I oc then let me bro play with it while i'm at work, if it doesn't work when I get home, i fix it and rinse, lather, repeat

 

[cadaveca]

I'll take stock voltages, and then scale clocks until stability is lost.

I then take stock clocks, and lower voltage until stability is lost.

This gives me the real voltage needed for stock speeds, as well as a higher speed, at higher volts, creating a range to play within.

I then play within that range, to find how the clocks scale with voltage. I'm looking for the place where the clocks stop scaling incrementally with like voltage increases. So say .25v gives 200mhz each time...when I need to boost increases to .30v or so, I stop.

then I change to better cooling, and repeat the same process, looking to see if that spot where I had to increase the voltage boost has changed. If it has, I then look for that same behavior again.

I'll do this for everything....bus speed, multi, ram dividers, ram timings, etc...usually takes a few motnhs before I've got it all done. Then I sell it, and buy another. I might buy the same chip, or one sold as "better", looking to see how those characteristics change.


But in the end...I don't do it for any real reason, you know? I'll have various stable configs for each part, but if I end up using the parts for a cruncher...it runs stock or less...looking for the best performance/dollar in power consumed. But it's not often I find a chip that I'll do it with...really gotta budget power usage based on how the house I'm in is wired.

 

[twilyth]

You know, I'm pretty sure I've used that before. God getting old sucks. It must have been on one of the other rigs.

anyway, 10 passes took no where near an hour, so i set it for 100 passes and set my stop watch. What do ya think? 2 hours on max for the win?

I know you said it doesn't cycle, but my temps go up to 52C and then down to 46C on each pass. there might be some other variations in there. I didn't watch speedfan every second, but there is definitely a range it is hitting.

However, it is still a better test than p95 since it maxed out at less than 47C.

I like your methodical approach but I just wouldn't ever have the patience. I suspect the bad WU's were coming from memory errors with the NB set so far above the std 200mhz. Although the crashes since going back to stock on the NB would indicate that maybe it was a little of both.

I'm happy with a 20 multi and stock everything else as long as I can run at 4ghz or close to it.

 

[FordGT90Concept]

Stable is simple: the system never BSODs, locks up, no computational errors, restarts itself, or powers itself off without being told to by the user. If it does any of the above, it is no longer stable--there's something wrong. Computers are stable until proven otherwise.

 

[twilyth]

After 113 minutes, it has completed 27 runs on maximum. I my guess sense of time was off since that works out to be 10 runs in 42 minutes. I thought it was more like half an hour.

Temps have ranged from 46C to 52C, that seems to be the highest. No errors. No crashes.

So, does y'all think I can get back to crunchin' ?

 

[cadaveca]

Should be good.

 

[t77snapshot]

i mainly use prime95 for 10hrs and anything over 70c i back down

Same here.

 

[Bundy]

I understand you don't feel like doing your OC the long slow way but I agree with the direction Cadaveca put forward. It is better to break your tests down to individual components, speeds and voltages and then put it all together later.

A stress test can only give you an answer that tells you whether the test passed or not, it doesn't ever confirm that your system is stable. All stress tests are the same. To have a better idea of stability (but still not 100%), I prefer to push each component individually to the point of giving errors or excessive heat, and then backing off the freq until it passes.

I never change two values in bios at once, always run a stress test after each change and keep a log book. The log bok is especially handy because it is easier to see where you have been, what you should do next etc. with a summary in front of you.

I doubt you will find much difference between IBT and OCCT but each to their own. I also consider WCG as a stability test. In the end, your final set up should be decided upon your full knowledge of heat, errors, tests, voltages and frequencies, not whether it passes a particular test or not.