RangerXLT8 said:
This is kind of off-topic, but what programs do you use to determine 724 stable for folding? I'm interested because folding strains a CPU more then any other task\stability test.
I use folding itself ... but you MUST take precautions so you don't hurt the science.
I have seen a few who start folding and produce reams of errors since their rig is not at all stable due to OC or hardware error problmes.
It hurts the science by slowing down the projects ... a work unit (WU) that ends in an error will be assigned to 3 or 4 other computers ... since the error could be the correct result of the simulation stanford has to assign it to several other rigs to see if they get the same error. Subsequent wu's in the sequence cannot be started until the previous one is completed since they depend on the results of the previous wu. So errors caused by a bad oc or hardware errors delay the whole folding sequence simulation as well as wasting cycles on other rigs that need to run the same wu to verify a bad result.
As a result of the above, people do not generally recommend folding as a stress test since reckless use of it hurts the effort.
Since my main purpose for me to have lots of oc'ed rigs is to produce as much good folding as possible, after doing all the normal stress testing, folding itself is my final verifier.
One approach I sometimes use is to run a few saved but already completed wu's (work units) which i can run to benchmark. I have serveral types which stress different parts of the system. eg. I have some of the old qmd's which are heavy on mem bandwidth testing but have been out of circulation since jan.
A second which I normally use, is to fold regular current wus. After i download them, i back them up, and then disable the lan interface so they can't turn in any errored work units due to a bad OC. Then i let them run. If they error out i just stop them, restore the backup, reduce the oc, and start them up again from the last good backup. If they error again at the same point, then the error is not due to my OC but is in fact a valid result.
If they complete ok, then i just enable the lan so they can turn in their results, download another wu and keep folding. Disable the lan again if i don't yet feel confident with the oc, but usually 1 or 2 WU's per core will prove it.
I tend to have more than 1 of the same cpu/mobo combo, so once i have OC'ed 1, it is pretty easy to get the others to a good folding stable OC.
I have run dual p95's for long periods of time only to get errors while folding.
Depending on the wu's you get, they can take from 6-8 hours up to 2-3 days or even more complete 1 wu.
The second method (folding current WU's) is the greedy one since it will usually produce points during a stability test. And does not hurt the science effort by turning in bad results.
Folding will produce higher load temps than p95 so your complete cooling capability will be tested. Almost all my rigs are caseless.
Folding will not test all your memory tho ... the biggest wu's atm use about 100MB ... the qmd's i mentioned above use about 350 mb each.
So p95 (or memtest) still needs to be run first ... but i only run 1 instance and let it run through the the first 2 fft sizes in the blended torture test - an hour or 2 depending on the rig. Then it is onto folding.
Finding the OC initially i use superpi and sandra bandwidth untill i get near what i think the max will be, then i also run pcmark04 to get a score. Once at max, then p95 as above, then folding as the final verification.
This is the routine i have developed from running a folding farm with 15 or so OC'ed rigs.