Running SETI@home enough to describe a CPU as 'under load'?

[snow patrol]

I'll be writing a comparative heatsink review in a little while, and I was wondering if having SETi running would be ok for my 'under load' temperatures. Comments?

Thanks

 

[Mikewarrior2]

With socket-thermsitors, usually 10-15 minutes so that the "backside-temp" stabilizes.

With a p3/internal diode, 10 minutes should be mroe than enough to peak cpu temp.


Mike

 

[snow patrol]

Thanks mike. This is with a FCPGA/slocket2/P3v4x combo. Should that give reasonably accurate temps (I know you're the forum guru for this, so don't go being modest )

 

[Mikewarrior2]

Even if the temps aren't reasonably accurate(and that's if), they are completely relative. So you'll get full temp changes between heatsinks. ... So Thanks for doing something a bit more than a lot of other people are doing..

One Suggestion I would make is to wait for the temps to stabilize...



Mike

 

[snow patrol]

Yes, that's one thing that was bothering me slightly. The temp stablizes fine when idle (goes down to around 26C and stays there), but when I'm running SETI, it seems to fluctuate between 35 and 40. Generally it stays on 35 or 36, but it does occasionally jump up to 40...43 even.

Here, I set my MBM interval log to 0.1 minutes and let it record the data for a minute or two. Here's the CPU temp results:

35.00
37.00
37.00
35.00
38.00
35.00
40.00
41.00
35.00
38.00
38.00
35.00

Is that normal for running SETI? It just makes it difficult for me to give a true 'under load' temperature. I'm not sure what I should put..

 

[Mikewarrior2]

That's the same type of thing I got with the p3v4x when I used it too... Hmm... potentially, since MBM uses some cpu time, when it kicks in to record/log temps it could be stealing cpu cycles from SETI.

Have you tried something like prime95 as your "cpu-loader"?


Mike

 

[snow patrol]

Nope, but I suppose it's worth a shot

 

[Mikewarrior2]

Ideally, a cpu-load will keep the same "load" on the cpu over a given time... giving you nice, level readings.


Mike

 

[Poof]

snow patrol - I attached a thermistor to my CPU substrate with the tip as close to the core as I could get it, all for monitoring the OC of a duron 700 I did over the weekend. When running seti, I noticed the temp fluctuation too. I think it may be due to the different types of calculations that seti does as it sweeps through the WU (some stressing the CPU more than others, like the pulse sweep).

I would expect that for the type of work/comparisons you want to do, something other than seti may be a better program to use...

 

[RoadRuner]

better off running seti + quake 3 at the same time to really generate some heat

 

[RabeaticSquirrel]

I would suggest, seti, along with burnbx, cpu torture, prime95, and maybe an mp3 or video stream (stored locally). This would definately give you ideal, constant, 100% load temps. I had fluctuations the first time I attached my thermistor to my CPU substrate like Poof did. After multiple benchmarks, and stress test, I evaluated that the programs were running the same cycle each time, so it must be a hardware problem. After I removed the Hedgehog, I cleaned and reapplied my artic silver (had to do this twice) but in the end, I now have constant 100% load temps ranging from 26C idle to 38.9 full load. VERY rarely does it ever get to 38.9.

 

[Poof]

You guys are nuts....!

Burn up that CPU, why don't you?



[EDIT: Oh and Rabeatic Squirrel, my temp readings don't fluctuate when I'm running Folding@home... Was gonna try RC5 next just to see though...]

 

[snow patrol]

might try RC5, see what that does. That should gobble up any spare cycles I guess, and perhaps along with seti will produce a nice rounded temperature.

 

[snow patrol]

Well, RC5 produces a constant temp of 36degrees. This is only at 1.6volts on a p3 500e overclocked to 667 bare in mind..

 

[Mikewarrior2]

Eeek... that's only 16W of CPU power at peek...



Mike

 

[snow patrol]

<< Eeek... that's only 16W of CPU power at peek... >>



Um, is that good or bad?

 

[Mikewarrior2]

good for power usage, not so good for testing.. .. Should still be good/relative results, though...


MIke

 

[snow patrol]

Yeah, I guess that's the main thing. As long as the conditions are the same for the tests, seeing relative temperatures should be fine.

 

[Mikewarrior2]

Yeah, Just measure ambient case temps for each test.


Mike

 

[snow patrol]

Thanks for the replies and tips Mike. You da man