Re: NEC-LIST: Re: NEC 4.1 Anomaly

I think the main reason for using single precision would be to allow more
segments for a given amount of RAM. It allows root 2 times the number of
segments over double precision. On a 4GB PC, SP allows over 22,000 segments
whereas DP can only handle 15,800. Of course one needs to check what
effect, if any, there is on the accuracy. Often, there is negligible
difference.

Regards,

Paul Carlier
FanField Ltd

At 01:04 PM 3/19/2006, Roy Lewallen wrote:
>I see this with NEC4S compiled with CVF. My guess is that it's an
>accumulation of error. I don't see it with NEC4D, or with EZNEC/4 which
>uses its own double precision helix creation routine even when NEC4S is
>selected to run calculations. If you need to use single precision NEC4, it
>would probably be a good idea to rewrite the helix subroutine in double
>precision.

I'm curious. I run only the double precision NEC4, so I don't have a feel
for the speed difference, but on even a 1GHz laptop (which slows down to
around 200-300 MHz running NEC, for thermal reasons), models run reasonably
fast (as in <1 minute for models with a few hundred segments).

The only really slow model is a 3x3 array of patches with 2000+ segments,
but that's a "go get coffee" run time, even on a fast desktop.

Why run single precision at all?

Jim

-- 
The NEC-List mailing list
NEC-List_at_robomod.net
http://www.robomod.net/mailman/listinfo/nec-list