Subject: Re: [linux-audio-dev] float a/d (was Linux support for IEEE1394 mLAN?)
From: Benno Senoner (sbenno_AT_gardena.net)
Date: Wed Oct 04 2000 - 01:32:40 EEST
On Mon, 02 Oct 2000, Jussi Laako wrote:
> Paul Barton-Davis wrote:
> >
> > i don't believe that 64 bit floats are necessary, but i also don't
> > believe that they are a problem either. i think that karl did a pretty
> > good job of explaining why this stuff about "resolution loss" is
> > relatively bogus.
>
> The biggest problem I've been dealing with when using generic computers for
> signal processing is memory bandwidth. It's the major limiting factor. DSP
> speed halves when moving from single to double precision FP...
That's not always true, unless you access large arrays of FP data very
frequently.
Plus think about the fact that if you use small fragment sizes, a big part of
the involved FP variables will be chaches, thus the single-precison FP advantage
partially vanishes in this case.
I have no numbers yet, but EVO will be a good app to test this.
(in fact I have a pure sampleplayback engine (resampling and mixing), and
when switching from doubles to floats the CPU usage drops only minimally,
but I assure you that heavy duty stress tests (float vs double) will follow
within the next few months).
Anyway I can't want to see it run on such an Alpha or PPC monster.
And the next step will be of course a clustered version which runs on beowulfs
interconnected by 100Mbit/1Gbit ethernet / Myrinet cards.
:-)
cheers,
Benno.
>
> I can get around 400 MB/sec from PC and 2.5 GB/sec from Alpha (DS20 & ES40).
>
> - Jussi Laako
>
This archive was generated by hypermail 2b28 : Wed Oct 04 2000 - 00:59:30 EEST