cpu overclocking software

[zietbukuel]

Hi, ive overclocked my A64 3200+ from 2.0 to 2.4ghz, and now im looking for some software that will check if my machine is OK, something like these windows programs:

prime95, superpi, cpuburn

[amne]

app-benchmarks/cpuburn is in portage.
_________________
Dinosaur week! (Ok, this thread is so last week)

[zietbukuel]

Thanks a lot

[MrApples]

emerge gimps, its the same as prime95 basically
_________________
http://www.whatsinyourbox.org -- Technology discussion, news, and more.

[bexamous2]

Just so you know, you can boot a knoppix cd and do 'apt-get update' and 'apt-get install gimps' to have mprime installed, this way you don't have to risk your gentoo install before you know its 100% stable. Also when knoppix cd first shows up you can type 'memtest' and run that first if your ram is being oced.

[zietbukuel]

Ive tried to use gimps, but dont understand it to much, ive done the benchmark without problems, but with the other tests I've got segmentation faults, i think my sistem is stable as it was up an running without problems for about 30 hours... I have gnoppix, is it the same as knoppix? can i use the memtest with it?

[MrApples]

segfaults are a bad thing obviously, often times pointing to faulty hardware (or bad overclock)

if you are doing it in gentoo, do /opt/gimps/mprime -t
this will run the torture test, you need to run it for at least 12 hours for certain (well almost certain) stability

also, the act that you were up and running for ~30 hours does not mean you have a stable system by any means

and yes, gnoppix is essentially the same as knoppix
_________________
http://www.whatsinyourbox.org -- Technology discussion, news, and more.

[energyman76b]

Hi,

the best test is:

compile kdelibs&kdebase over and over again.

Or built a douzend of kernels.

If you get segfaults or ICE's, stop overclocking.

OC'ing makes your computer stop being a deterministic maschine and starts introducing randomness. It may work for you for a while, but at some point it will bite you in the ...
_________________
Study finds stunning lack of racial, gender, and economic diversity among middle-class white males

I identify as a dirty penismensch.

[zietbukuel]

[bexamous2]

"OC'ing makes your computer stop being a deterministic maschine and starts introducing randomness. It may work for you for a while, but at some point it will bite you in the ..."

Blah. Alot of certinty in that statement considering how broad your claims are... hell considering how stupid it is.

--edit--

oops forgot to reply to the actual topic...
"Ive done that, ive compiled GNOME and KDE, lots of software (200+ pkgs), recompiled my kernel several times and i had no problems at all, just this things that keeps segfaulting... "

Well none of that really is going to mean your computer is stable, and the fact your getting segfaults would actually more likely mean that it is infact not stable. The quickest way to find the source of the segfault problem is the comptuer back to default speed and try it again and see if it segfaults. If it doesn't then thats pretty much proof your computer was not stable. If it does segfault it doens't really mean much because the package was compiled with a potentially unstable computer... again this is why using knoppix cds (or whatever varient) is nice. Really unless your checking for errors many can occur without you knowing it. Alot of people recommend compiling the kernel alot, however really theres no reason if an error occurs it will necessaly make the process fail and exit... it would be more of a test if it did a md5sum on each resulting kernel and matched it after every time. Also if you watch the cpu usage while compiling it isn't a constant 100%... mprime is, plus the results are checked against something, although I'm not sure to what degree.

Anyways... dinner time YAYYY... I'm rambling too so perfect timing =D

[energyman76b]

[quote="bexamous2"]"OC'ing makes your computer stop being a deterministic maschine and starts introducing randomness. It may work for you for a while, but at some point it will bite you in the ..."

Blah. Alot of certinty in that statement considering how broad your claims are... hell considering how stupid it is.

--edit--

no, stupid is someone who overclocks and shuts his eyes in front of potential problems.

As long, as you are using your computer in the borders of its specs, it is guaranteed, that you'll get correct results.

Lets take a look at the CPUs.
In the factory, they are speed binned. The manufacture checks up until which speed it delivers correct results. When it starts making errors, it is clocked to high, and the last correct result will define the speed which will print on top of it.
Or look at the busses. There are reasons, that they are not clocked any higher by default. The reason: they are save from errors at their specified speeds. If you go higher, you'll introduce errors. Some of them will be caught, some will go unnoticed, and some nice day, your harddisk will be flodded with corrupted blocks.
Harddisks is a nice point, do you know how sensitive this beasts are against overclocking? No? Ever wondered, why they introduced 80wire cables with udma66?
There are a lot of PCI-cards that will throw in some errors once in a while, as soon as the PCI bus is clocked higher than 35Mhz. Especially controller cards are known for that.

But hey, why should you be interessted in data integrity, if you can overclock?
_________________
Study finds stunning lack of racial, gender, and economic diversity among middle-class white males

I identify as a dirty penismensch.

[zietbukuel]

mmhhhhhhh it seems that overclocking its a bad idea after all... so i'll just stop doing it right now this is a new machine and i dont wanna screw it up... thanks a lot for your advices! they are really appreciated!

edit: ive tried that software without overclocking and i dont get any errors... so, no more overcloking!

[bexamous2]

"no, stupid is someone who overclocks and shuts his eyes in front of potential problems. "

Who's doing that? Thats why people test a computers stability when oced, to rule out potential problems.

"As long, as you are using your computer in the borders of its specs, it is guaranteed, that you'll get correct results."

Is it? I bought 1GB of cosair ram at xmas.. when I first got it ran memtest overnight no errors. Gentoo everything setup fine, this is all default speed no ocing involved. After a couple weeks I had something crash, can't remember what it was. Did memtest found nothing, ran prime95 it went 16 hours and then gave an error. Couldn't figure out what was worng, kept doing stuff and all of a sudden the ram failed horribly. Memtest gives 1000s of errors in a few seconds. Find the bad stick and atm I'm running with only 512 until the replacement comes. GUARENTEED means alot heh. Most of my job is actually dealing with components that failed, none of htem have ever been oced. Hell how often does stuff come DOA too for example? Everything can fail no matter what... ocing is not required. Yes I understand what your saying but guarenteed is a word.. it does not make anything more or less likely to happen... running within its specs has a very high probabilitly of running without problem, however running outside of spec does not mean a high probability of failure.

"In the factory, they are speed binned. The manufacture checks up until which speed it delivers correct results. When it starts making errors, it is clocked to high, and the last correct result will define the speed which will print on top of it. "

Well that sounds like a poor way of testing, how much is time worth? Oh yes alot. How long does it take for an error to occur? People can run prime95 for a few hours error free for example, yet overnight an error happens. Probability and statistics come in handy here. They can't spend lots of time testing every single processor, it would drive up the costs to insane levels. We need a way to quickly test. The best woudl be to take a small sample of cpus, do alot of testing on them and use ethe statistical data to determine a 'fast' way of binning. They are not trying to find a cpu's max error-free speed, they are tring to find the max error-free speed they can find very quickly, this is important... bin has NOTHING to do with the max speed. This does not even take into account when you start getting better yields and things bin high, if you need lower bin cpus you take hte higher ones and mark them lower. But lets ignore that.

Now theres alot more about cpu's rated speed. You need to specificy a specific max temp, power requirements alot of things come into play. Lets just look at those two. Max temp. Higher temp more a error to happen, thus you can lower the max temp and raise the speed and keep same probabilities. Power requirements, look at DFI or Asus who market cleaner power for the cpu, this is tighter specs than the cpu was designed for.. or lets just look at the cpu's core voltage.. they rate it at a specific speed and voltage.. higher voltage transistors switch faster and therefore can run faster.

Anyways when overclocking you can spend whatever time you want, there is nothing special about the testing amd/intel does... you can do your own testing. You can do more testing and have a more fine result, they might not spend to time to get a higher rated speed, you have the time. You can also change the parameters of the cpu, you can test it to your own settings. Their worst case might be 80C you can test for the worst case of 60C. For example, me, I have a opteron 170, 2.0ghz default speed. I tested it at 2.5ghz, 1.4v running a copy of mprime on each core for 48 hours... no errors... then I put cpu at 2.4ghz 1.4125v.. what is the probability that my cpu generates an error? For me it is a very acceptable probability. Actually I'm still waiting for my replacement ram before I do any other testing but really I'm not concerned... I feel no less safe at this speed than 2.0ghz, however most likely I will do more testing just because I'll get bored one night. Not that inconvieetn to run mprime iwth 19 niceness if you think about it.

"Or look at the busses. There are reasons, that they are not clocked any higher by default. The reason: they are save from errors at their specified speeds. If you go higher, you'll introduce errors. Some of them will be caught, some will go unnoticed, and some nice day, your harddisk will be flodded with corrupted blocks.
Harddisks is a nice point, do you know how sensitive this beasts are against overclocking? No? Ever wondered, why they introduced 80wire cables with udma66?
There are a lot of PCI-cards that will throw in some errors once in a while, as soon as the PCI bus is clocked higher than 35Mhz. Especially controller cards are known for that. "

You don't even know what your talking about. PCI BUS!? Unless your running a pentium 120 every decent motherboard has a option to lock its pci bus to 33mhz. Honestly I cannot remember my last motherboard that did not have this feature. Because you're trying to make arguemnts that make no sense I'll try to clue you in. Every oc-friendly motherboard- and there are many, hell DFI seems to be building their business on overclocking friendly boards- has lots of nice ocing options including changing every multiplier you could need, or locking speeds. There is no reason to run ANY bus out of spec if you do not want to. Ram for example, ratios to the fbs, alternatively you can buy ram rated at a nigher speed. PCI you can lock. PCIE you can lock. Hypertransport for a64s have a multiplier. Hell now AMD is marking the FX line to overclockers, they're all going to be multiplier unlocked so you don't even have to TOUCH any bus type setting, just raise its multiplier.

OCing is not limited to the cpu of coarse... I mean you can oc any bus you want.. or video card's gpu/ram you can oc. Hell nvidia rates its chips at a specific speed, BFG sells cards OVERCLOCKED running out of nvidia's spec GUARENTEED with a lifetime warrenty. I wonder how that is even possible considering, according to you, "at some point it will bite you in the ..." Wonder how long till it bites them in the ass?

Really any component rated at any speed is based on testing. Nothing stops you from doing your own testing on your own specific compoenent, alot easier to test one cpu at a specific speed than to test 1000s and 1000s... anyways if you do poor testing it might fail, if you do good testing however I think there is no reason you should be any more likey to have a failure than someone else who runs their cpu at hte default speed.

Oh yea I could have made this entire post alot shorter, all I initially said was claiming EVERY overclocked compoent WILL fails is stuipd. You didn't say might, or more likely, you said WILL. And I'll say it again, thats a stupid claim to make.

[MrApples]

zietbukuel, that is probably a good choice then for the time being
overclocking and doing a good job of it isnt a simple task, and it takes a lot of time, patience, and knowledge to do without hindering stability
_________________
http://www.whatsinyourbox.org -- Technology discussion, news, and more.

[Jake]

This thread inspired me to try gimps on my overclocked Athlon64 Newcastle socket 754 2800 (natively 1.8GHz) running at 2.4GHz with stock HS&F.

[AlphaDriven]

There is a program that is called System Stability Tester which has linux x86 and amd64 ports. It calculates the Pi using the Borwein algorithm with two or more threads and it compares the results of all the threads after each step. In the case the results are not equal it warns you. You may have it to run continiously for hours and log the results of each step.

You may download the x86 static binary from here:
http://www.pctechnology.gr/mod/bench/systester-0.6-x86-bin.tar.bz2

The amd64 binary from here:
http://www.pctechnology.gr/mod/bench/systester-0.6-x64-bin.tar.bz2

And the sources from here (GPL v2):
http://www.pctechnology.gr/mod/bench/systester-0.6-src.tar.gz

Two nice screenshots are here:
http://www.pctechnology.gr/vbull/vb/attachment.php?attachmentid=8173
http://www.pctechnology.gr/vbull/vb/attachment.php?attachmentid=8174
_________________
"I know you insist but I won't pollute it with crap!" -- Dave Cutler to Bill Gates during the development of Windows NT