World Community Grid/F@H: Crunching for the Cure... or are we?
 

A lot of us here run either World Community Grid, F@H, or both, myself included, which is a great thing, and I admire our efforts. Similarly, a lot of us overclock our rigs. Processors, graphics cards, memory, all... presonally, I find myself tweaking whatever can be tweaked to get every ounce of performance out of it... not because I have to, but because I can, and because the faster my components are, the more work I do.

Now, this too is a great thing, but I have seen the topic of 'old school' and 'new school' overclocking argued countless times. Myself, coming from the old school, am sure to say that the new school method is wrong, because the new school meathod seems to be strictly trial and error—set something and roll with it. If something errors, change it. Now, that's fine if that's how you roll, but consider this: if you run your system this way, not knowing whether it's truly stable or not, how can you be sure you're not sending in bad results to the WCG/F@H servers? Sure, they send the same work unit out and compare the results for differences, but there's two problems with this. The first problem is it's possible for something to slip through, as with any system. The second problem is that if you're sending in work units that are getting thrown out in the end, you would be doing much more work running at stock than overclocked.

Anyways, my point being made, I encourage each and every one of you, if you havn't already, to thoroughly test your overclocks. Run LinX overnight, and if it errors, do something to correct it—back down the clocks, change voltages, whatever. Same with your GPU... run the OCCT GPU test for a while, like set it to run before you go take a shower, and check it when you get back. I usually take about 20 minutes once everything's said and done, this should be enough time to expose any errors. If it errors, back down your clocks.

Links to some stability tests:
tests provided by Stanford: http://folding.stanford.edu/English/DownloadUtils
OCCT: http://www.ocbase.com/perestroika_en/index.php?Download
LinX: http://www.xtremesystems.org/forums/...d.php?t=201670

monitor temps with realtemp and gpu-z
http://www.techpowerup.com/realtemp/
http://www.techpowerup.com/gpuz/

Might want to change the title, I thought your thread was going to be about if Folding & crunching are worth while.

That's kind of my point. It's not worthwhile if the work is being done by not-so-stable rigs.

I think you bring a very good point to the surface.
I stress and stress every clock and at times even the ones that pass 24 test still fail on a few WCG units or my rig overheats. Some like to avoid work units that give errors but to me that just tells me my rig just isn't 100%.
I am doing this now. The 1055t and the I7930 haven't done a full 24h test the lower clocks have but the new ones no not at this time.

My rule for OCs is that it must pass 8 hours of LinX, 8 hours each of Prime 95 Small FFTs, Large FFTs, and blend. If it passes this, it's fully ready to go for F@H/WCG

Just got home and the I7 failed Prime 95 blend. clocking her down to 4.0
Thanks for the reminder Hat. Also I realy need to find a new home away from the main rig that vents strait at her.

I've always firmly believed overclocking and science don't mix. Overclocking increases the likelihood of errors and good science doesn't tolerate errors. The 25% performance gain by overclocking is easily lost on wasting 4 hours of CPU time with a completed task being invalidated.

There has to be a fine line, just how much is too much. This is T POWERUP we will push.

how do they know if a work unit is invalid?

They send the same work unit out to a bunch of machines and compare results.

2 or 3, typically, although I think it's more for HCMD2 and was more for RICE as well

hat, how about some links in the above? Also, F@H has 2 tests available to test your system. A memory test for Nvidia cards and a CPU stress test, if anyone's looking for another way to test. Link

When memtestg80 asks for your memory clock... do you put the in the speed before or after DDR?

Just a guess, but I'd say before. http://fah-web.stanford.edu/MemtestG...G80-readme.txt

gah, now I can't figure out how to make it use more than 128mb vram...

occt has a similar test that works fine, though

For example, to run MemtestG80 over 256 megabytes of RAM, with 100 test
iterations, execute the following command:

MemtestG80 256 100

If I do that, the program just crashes when it starts the test...

I usually test my overclocks with WCG. Never had any rig return errors, if it crashed I adjust it and that's it. You spend 8 hours stress testing or more you are not returning any results. You had a rig crash one or twice in 8 hours, at least you got something done. But of course that's my two cents, In no way do I want to start a arguement, I truly respect hats opinion. However, I would just consider myself new school I guess.

I'm the same way... no need to prime for 24 hrs if it is crunching without errors or crashing. I just watch temps!

I have never Prime'd for 24hrs, however I will not run WCG without some sort of stability testing done first.

All I do is OCCT the CPU for 90-120 minutes and if it passes I let it rip on WCG 24/7 and watch my results. I have yet to return an invalid WU due to OC:D

You are doing it wrong. I very strongly argue against this. I don't care what you do with your own data, but if you're handling work units on a possibly unstable computer, you're possibly playing with lives because you were too lazy to run a good test for a while. One day max of time spent testing and not computing work units is nothing compared to a lifetime of corrupt results generated from a bad overclock.

Did you really jst call me lazy... do you know the amount of hours i put into oc'ing my computer? Do you know the hours of research i have done. I respect your opinion but it's your and no need to call me or anyone else names... especially when you don't know what your talking about!:cry::cry::cry::cry::cry:

your all folding and crunching to help north korea build nuclear weapons you just dont know it yet

Rock on teabag...new info everyday...i love my life.. :banghead:

The fact that you are taking someone's good intentions and equating it to "Playing with people's lives" is shameful. :slap: You have no right to call someone else lazy, you are not them. Lighten up.

Not really appropriate. :shadedshu