[linux-audio-dev] Vector/DSP processor

[Andreas Roedl]

Hello!

Am Freitag, 21. Januar 2005 21:50 schrieb David Olofson:
> > i don't think you meet the synthesis capabilities they described
> > even on a bleeding edge athlon. the voice counts for just the
> > wavetable synthesis along are amazing, and you can run KARMA,
> > wavetable, subtractive, organ and one other synthesis method all at
> > the same time.
>
> That might be a problem, even with heavily optimized SIMD code. :-)
> After all, general purpose CPUs are optimized for dealing with the
> "unpredictable" control flow of normal applications rather than the
> kind of calculation/bandwidth ratios high end DSPs are designed for.
>
> Not much to do with operating systems either way, although the only
> potential solution - massive SMP - depends very much on IPC latency.
> That said, I think you could get away with some CPUs in parallel for
> the synthesis and some more CPUs in parallel for master effects and
> mixdown. Shouldn't be much of a problem, since you can actually
> process one sample at a time reliably (at least at 48 kHz and lower)
> with RTAI or RTL on reasonably good hardware.

Slightly off-topic, but imagine what you can do with a processor like this:

  http://www.blachford.info/computer/Cells/Cell0.html

128 GFlops (douple precision), 20 times faster than todays processors, 8 
vector units, available at the end of the year.



Andi