| Subject: Re: Very Very short completion time??? |
| From: "Stratcat" <none@no.org> |
| Date: 22/07/2004, 19:00 |
"Martin 53N 1W" <ml_news@ddnospamddml1dd.co.uk.dd> wrote in message
news:b6SLc.642$eV.42@newsfe1-gui.ntli.net...
<snip>
You don't know if the FPU is good or just producing fast garbage.
A CPU can be stressed such that you do not see any system crashes whilst
the FPU timings fail and produce 'unseen' numerical errors.
Run the GIMPS program, select Torture Test, and you are then running
some very intensive /calculations/ with *checksums* that test if your
FPU is good under those conditions.
To not do so, you might be just calculating expensive wothless garbage
and you will not know.
Diagnostics are useful.
Agreed.
My team had a very interesting discussion on using Prime torture and GIMPS
doublechecks for machine validation, before putting into svc, particuliarly
when running GIMPS as a DC project.
The discussion was reasonably intelligent, many of the contributers where
pretty seasoned working pros, and other stability related topics such power
conditioning (and even memory bit cosmic ray errors!) were discussed.
If you have broadband, you may be interested in checking it out.
(caveat: contains many GIMPS terms and concepts)
ARS Forums via TinyURL redirect:
http://tinyurl.com/3t8z6
--
Strat