overclocking can cause floating point errors'?


log in

Advanced search

Message boards : Number crunching : overclocking can cause floating point errors'?

Author Message
Alex
Send message
Joined: 26 Sep 01
Posts: 260
Credit: 2,327
RAC: 0
Canada
Message 13557 - Posted: 4 Aug 2004, 1:37:53 UTC

Am watching the Video by Mr. Anderson
http://www.climateprediction.net/project/openday.php

16 minutes (and 50 seconds) into the video, he mentions that Seti sometimes gets bad results from overclockers.


Profile JigPu
Avatar
Send message
Joined: 16 Feb 00
Posts: 98
Credit: 506,133
RAC: 0
Message 13577 - Posted: 4 Aug 2004, 3:45:39 UTC
Last modified: 24 Aug 2004, 4:20:16 UTC

Overclocking can cause any kind of CPU error. Integer, floating point, SIMD, all are susceptable to error. If the CPU is running with insuficient voltage for it's speed, too much heat, or just faster than the chip can handle, errors will inevitably creap in. If you strive for a stable (like Prime95 stable) overclock though, it should be a nonissue.


_______________________________

Alex
Send message
Joined: 26 Sep 01
Posts: 260
Credit: 2,327
RAC: 0
Canada
Message 13579 - Posted: 4 Aug 2004, 5:07:41 UTC - in response to Message 13577.

Is there a benchmark that overclockers can run to verify they're not having floating point errors?


EclipseHA
Send message
Joined: 28 Jul 99
Posts: 1018
Credit: 244,307
RAC: 74
United States
Message 13580 - Posted: 4 Aug 2004, 5:10:08 UTC - in response to Message 13577.

> Overclocking can cause any kind of CPU error. Integer, floating point, SIMD,
> all are susceptable to error. If the CPU is running with insuficient voltage
> for it's speed, too much heat, or just faster than the chip can handle, errors
> will inevitably creap in. If you strive for a stable (like Prime95 stable)
> overclock though, it should be a nonissue.

If you want good data, overclocking is bad, as the manufacturer specs are exceeded, and even if it works clean for a time, it could fail later, and you won't know when...

On Seti/Boinc however, it will make those "No Scheduler Responded" and "no work available" messages display a bit faster!

Seti/BOINC is only crunching data that was already crunched by Classic (Jan04, Aug03, Sept03) so there's no real loss even if you can get work to crunch!

(I feel a cheerleader generating a response!) YES, SETI BOINC IS NOT PROCESSING NEW DATA OR OLD DATA IN A NEW WAY TODAY! (see http://setiathome.ssl.berkeley.edu/calendar/calendar.html)

Message boards : Number crunching : overclocking can cause floating point errors'?

Copyright © 2014 University of California