MATLAB Central Newsreader - Computer precision?

Computer precision?

onofftopic@gmail.com — Sat, 03 May 2008 15:03:35 -0400

Hi,

according to the Matlab Help double computer precision (64-bit) is
default in Matlab for 32-bit computers. Maybe is this question
offtopic, but i would like to know why is there no higher precision,
e.g. 128-bit precision, for 32-bit computers?

What is the limitation for the computer precision?

Thank you in advance.

Re: Computer precision?

carlos lopez — Sun, 04 May 2008 20:21:03 -0400

onofftopic@gmail.com wrote in message
<d039b0df-5f51-43a0-9b4b-e3127d398868@27g2000hsf.googlegroups.com>...
> Hi,
>
> according to the Matlab Help double computer precision
(64-bit) is
> default in Matlab for 32-bit computers. Maybe is this question
> offtopic, but i would like to know why is there no higher
precision,
> e.g. 128-bit precision, for 32-bit computers?
>
> What is the limitation for the computer precision?
Hello onofftopic@gmail.com:
On principle there is no fundamental limitation... aside
from the degradation in performance.
In order to use arbitrary precision you might want to test
the Multiple PRecision toolbox by Ben Barrowes, available on
the FEX. It allows to simulate any precision.
"Double precision" is defined in the IEEE 754 standard; 32
bit matlab implements it on all the architectures, so you
can achieve identical results irrespective of the platform.
If you need just a bit more precision than double AND you
are using INTEL processors, you might want to look at the
Extended Precision class by Marcel Leutenegger, available
somewhere.
Regards
Carlos

Re: Computer precision?

Peter Boettcher — Mon, 05 May 2008 13:06:09 -0400

onofftopic@gmail.com writes:

> Hi,
>
> according to the Matlab Help double computer precision (64-bit) is
> default in Matlab for 32-bit computers. Maybe is this question
> offtopic, but i would like to know why is there no higher precision,
> e.g. 128-bit precision, for 32-bit computers?
>
> What is the limitation for the computer precision?

Modern computer hardware implements 64-bit floating point very
efficiently. Anything larger than that requires software
implementation. It is possible to implement 128-bit floating point
using pairs of 64-bit FP values, with software glue, but any approach
like this will be much slower than the hardware-supported 64-bit
computations.

-Peter

Re: Computer precision?

roberson@ibd.nrc-cnrc.gc.ca (Walter Roberson) — Mon, 05 May 2008 16:59:00 -0400

In article <muyskwwj3y6.fsf@G99-Boettcher.llan.ll.mit.edu>,
Peter Boettcher <boettcher@ll.mit.edu> wrote:
>onofftopic@gmail.com writes:

>> according to the Matlab Help double computer precision (64-bit) is
>> default in Matlab for 32-bit computers. Maybe is this question
>> offtopic, but i would like to know why is there no higher precision,
>> e.g. 128-bit precision, for 32-bit computers?

>Modern computer hardware implements 64-bit floating point very
>efficiently. Anything larger than that requires software
>implementation.

most x86 systems have 80 bit floating point internally (though the
values are usually rounded to 64 bits when stored to memory.)

I do not know if it is possible to turn on 80 bit mode in Matlab.
I have never been able to keep track of the system dependant
functions such as system_dependant(). I didn't see anything likely
when googling, but I didn't spend particularily long at it.l

--
"Is there any thing whereof it may be said, See, this is new? It hath
been already of old time, which was before us." -- Ecclesiastes