DNR - No wait, R-R-R!!!! Single Core Question....

[Shadow703793]

If your concerned about CPU life don't even bother. I have a Celeron D 315 @3.6+Ghz since 2003! (see my sig) Its still running fine. By the time the CPU dies technology will be way ahead. One of my uncle's has a Am386 thats OCed and still running MS-DOS!. OCing has been aroun a LONG time. See:
http://www.tomshardware.com/forum/204157-29-long-overclocking-around

 

[MU_Engineer]

If your concerned about CPU life don't even bother. I have a Celeron D 315 @3.6+Ghz since 2003! (see my sig) Its still running fine. By the time the CPU dies technology will be way ahead. One of my uncle's has a Am386 thats OCed and still running MS-DOS!. OCing has been aroun a LONG time. See:
http://www.tomshardware.com/forum/204157-29-long-overclocking-around

The OP's situation and what he's said so far puts a limit on OCing:

- He wants the machine to last for several years. This means that an overclock needs to be at a reasonable voltage as excessively high voltage causes electromigration and kills CPUs. Personally, I'd use the maximum value in the typical Vcc_VID_MAX range as my ceiling as the manufacturer has guaranteed that voltage won't kill chips in less than 3 years (they have a 3-year warranty.) A limit to what is essentially stock Vcore may put a damper on overclocking to some extent, but most chips will overclock at least a couple hundred MHz on stock Vcore, some 500+ MHz.

- He is doing actual work with the CPU and not gaming. This means that the results of a program execution matter and the overclock has to be both operating stable and mathematically stable. I'll be willing to bet that most people on here don't test to see if their overclock is mathematically stable by running a program with a known result of computation and then doing a diff between that and what the CPU spat out. This is important as a CPU can have a transistor flip and goof up a mathematical operation and not crash the computer but it will ruin a computation. Running Prime95 is a good way of determining the mathematical stability as it checks your CPU's computed answers to known-good answers. Let it run for many hours to ensure that a power ripple or something won't cause math errors.

So he can overclock, but he needs to be judicious about it and probably won't get one nearly as high as the gamers get due to those criteria. Personally, I'd forget overclocking in this situation as I'd want to just get down to crunching the data and not dick around finding the best OC, and also knowing that the machine is rock-solid-stable. But I don't know, maybe that extra 20% he might get may be more important.

 

[nukemaster]

Prime for 48 hours....and folding all the rest of the time (it will report errors too...or so i hear)

 

[MU_Engineer]

Prime for 48 hours....and folding all the rest of the time (it will report errors too...or so i hear)

Yes, it will. It's quite sensitive to core voltages being too low for the CPU's current speed. I know this from experience in undervolting my CPU using a Cool 'n Quiet FID/VID remapper. I've found my particular X2 4200+ is operating-stable at 1.175 volts at 2.20 GHz, 5-hour mPrime (Linux Prime95) stable at 1.200 volts, but will cause Folding@Home to spit out errors and kill the current WU within 10-20 minutes if the Vcore is below 1.225 volts at full speed. If the Vcore is >= 1.225 volts, it'll run at least 40 days without a single error- that is as long as I have gone between reboots due to security > uptime in importance to me.