The importance of a stable oc.

Just wanted to do the reminder thing for thost of us who fold with O/C'd rigs.
it should be a no-brainer, but ive learned that for my rig anyway, "folding stable" and "stable overclock" as outlined in the stickies-are two very different things.

ive recently gotten my main production rig (in sig) back folding after a brief hiatus, and the results speak for themselves.
have a look at the rebound: http://folding.extremeoverclocking.com/user_summary.php?s=&u=95559

i an currently overclocked by only 6 MHZ , but the rig is NOT screaming, im NOT losing points from corruption, my fans dont keep me up at night,my load temps are below 40C, and my system uptime is allready around 4 days without a single crash -and in all honesty, ive never seen my PPD this high, ever.

if you fold and youre having a tough time, crank it down a bit, see how it goes.

ISTM. orionlion82, that is important as we are sending in important parts of research that become whole projects. They do run redundant sets to check accuracy but to many errors just slows down the research and confidence in the grid. I've fine tuned my rigs with generaly boring OCs also just to be well within the safe zone. I've also started running Prime when I make changes to these rigs. I also believe orionlion82 hit on something else. If you notice that little benchmark thingy that runs first, IMO that is a check to see what your system is running at (speed,stock/OC etc) and a quick sample to check for errors. I've noticed as a borg keeps running and producing results it starts getting the 400-600 pointers. If you are heavily OCed or your system shows signs of instability, it may be the reason you're in tinkerville.

The benchmark is useless since it is drastically affected by anything starting up at the same time it's run. FAH doesn't use it. Instead, the performance fraction is used by the assignment server along with cpuid, connection speed, and available memory. The primary reason people wind up in Tinkerville is due to a lack of big WUs, Aside from QMDs, there are currently only about 400 big WUs available for assignment with many thousand cpus set to receive them. It's luck of the draw if your performance fraction meets the minimum requirement and you have enough memory.

orionlion82 said:

if you fold and youre having a tough time, crank it down a bit, see how it goes.

Click to expand...

What do you mean by "rough time"? I have prime95 48hours clean and 3DMark Nature test 100 Loops clean. Is this enough? How can I tell if my folding is clean?

R

ropey said:

What do you mean by "rough time"? I have prime95 48hours clean and 3DMark Nature test 100 Loops clean. Is this enough? How can I tell if my folding is clean?

R

Click to expand...

You are good, by rough time he means work units crashing due to instability. If your sytem is majorly unstable WU's won't finish and will crash partly though. If you can prime for 48 hours and 100 loops thats stable enough for FAH.

It will early end if simulation instability, is encountered. You may get proportional credit but most times if it's your rig thats unstable you'll get little or nothing because the data is trashed. If it completes the WU and the server accepts it, you don't have to worry about the integrity of the data.
FAH is much better than Prime95 at finding instabilities.

Awe, ChasR, you spoiled the mood I thought the reason I kept getting big WUs was because I had such great rigs. I guess I'm just lucky

ChasR said:

It will early end if simulation instability, is encountered. You may get proportional credit but most times if it's your rig thats unstable you'll get little or nothing because the data is trashed. If it completes the WU and the server accepts it, you don't have to worry about the integrity of the data.
FAH is much better than Prime95 at finding instabilities.

Click to expand...

Please don't use F@H as test for rig stabilization, save that for HL2, DOOM3 or RCT3, this work is too important. And yes sometimes the units do reach the univeristy with bad data since in some cases the bad numbers get crunched along with the good and the app doesn't know it. Yes I work with (not write) engineering software that does goof and send bad results that are not detected until someone looks at the data. You never know where the memory will frag, it can be in the code or data. Once in the data core the is a FP unit get trashed by one transistor locking up due to leakage from overvoltage, the number justs gets passed into the mix. If it's one of the energy-field testing areas of code. it could be just the sign on the fraction that determines whether a chemical is attracted or repeled throwing of the results.
Memtest.
Prime
Fold.

AlabamaCajun said:

Memtest.
Prime
Fold.

Click to expand...

Sounds like a plan to me

thanks,

R

don't forget, in my tests... aquamark is a good indicator of a good OC. If its bad, aquamark is VERY buggy and nasty and locks up. If the OC is good, it runs smoothly (how i found out that anything over 221 FSB on my board is unstable)

Sometimes better lucky than good
There have been so many times my FX-53 has gotten Tinkers while a P3 800 at my office is cruntching a 600 pointer, it's a bummer, but it's plain ol' luck.

AlabamaCajun said:

Please don't use F@H as test for rig stabilization, save that for HL2, DOOM3 or RCT3, this work is too important. And yes sometimes the units do reach the univeristy with bad data since in some cases the bad numbers get crunched along with the good and the app doesn't know it. Yes I work with (not write) engineering software that does goof and send bad results that are not detected until someone looks at the data. You never know where the memory will frag, it can be in the code or data. Once in the data core the is a FP unit get trashed by one transistor locking up due to leakage from overvoltage, the number justs gets passed into the mix. If it's one of the energy-field testing areas of code. it could be just the sign on the fraction that determines whether a chemical is attracted or repeled throwing of the results.
Memtest.
Prime
Fold.

Click to expand...

I've never used FAH to do stability testing in a manner detrimental to the project. What I meant is that you may well find out the rig isn't stable when folding even though it passes Prime95 and memtest. On the Stanford site and a few times here, folks have blamed the core, the WU, Stanford, when they get a bunch of early ends. All claim to have passed Prime95, memtest, StressCPU and therefore it must be Stanfords problem. ATM, even the most complex WUs have an early end percentage far less than one percent. If you EE two in a row, FAH has in all likelyhood found an instability in your system.

You can however, use FAH to test stability without adversely affecting the project. After the download of a WU, disconnect from the internet. Stop the client and make a backup copy of the Work directory and queue.dat. OC and fold away. When you cause an early end, resore the backup and back off or tweak till you get a stable OC. When you've got it stable, reconnect to the internet and turn in the work unit. If you do this on a 600 pointer or a p147x, you can be pretty sure you've got a folding stable OC and benefit the project by reducing the chances of EEing Work Units.

TollhouseFrank said:

don't forget, in my tests... aquamark is a good indicator of a good OC. If its bad, aquamark is VERY buggy and nasty and locks up. If the OC is good, it runs smoothly (how i found out that anything over 221 FSB on my board is unstable)

Click to expand...

i agree, aquamark is awsome for quick testing, it always picks off unstable oc's for me

well, up untill about last month i thought early unit end was a Good thing, because it allways happened - i just figured that meant i finished before the deadline.
i was then miscorrected once and for a while thought it was Stanford pulling a WU "off the shelves" and figured i was unlucky..

flash forward to now and i memtest-prime-game-fold every time i tick up my FSB.
ive found my new harddisk is more sensitive even if i lock down the bus.
3400 used to be memtest-prime-game stable... but after i replaced my harddisk from corruption, i get Bsods...3200 is the highest tested "game stable" so far (please no one say GNDS, i deny all knowledge of its existance. i believe in wishful thinking)


but yeah, i know i cant be the only one in the community struggling with folding now and then, so i just figured i would play captian obvious and maybe help out a few folks if i could but putting this stuff out there.
Happy folding!