Re: [Linuxsampler-devel] LSCP Protocols - Some confusion about mixing vs. channels

[Mark K. <mar...@gm...>]

On Tue, 8 Feb 2005 21:18:49 -0500, Vladimir Senkov <ha...@gm...> wrote:
> Hi guys,
> 
> > > c) Separate outputs for each gig come out but attach to separate
> > > alsa_pcm:playback_XX channels. The audio is mixed in hardware with no
> > > CPU overhead.
> >
> > Well, just mixing does almost mean no cpu load. Memory bandwith will come more
> > into weight.
> 
> In case of jack, yes.
> But in case of multichannel card . . . there will also be a hit on the
> PCI throughput.
> I'm assuming that DMA is transferring the data.
> In some cases PCI could be a significant bottleneck too . . .
> But . . . if it is functionally required to have multiple HW channels
> (if mixing is done "outside" the sampler machine on some other system)
> then this is what we have to do, but for purely functional reasons,
> not for performance reasons.
> 
> Regards,
> Vladimir.

I'll give some silly background on me and mixing paranoia. In the last
few years it seems that every too I use I eventually find that the
submix buses are the weakest part of the structure. The submix buses
are always causing the most 'hidden' distortions. I don't know why. I
read in Bob Katz's book that he thought both hardware and software
developers didn't pay enough attention to all the stuff that can
happen there in terms of signal overloads, etc. I don't know, but I've
seen it to be true in higher end tools like Pro Tools and more
moderate tools also.

With that in mind I know get good metering in the HDSP 9652, bth under
Windows with TotalMix and under Linux with hdspmixer. If a submix bus
overloads I can see it. With the current state of LS I don't believe I
could.

As for PCI bus bandwidth, you're correct that this method uses more
bandwidth, but the card is designed to handle 26 in and 26 out so I
figure I might as well try. Besides I'm only sending PCM data out in
this case. (LS as a stand alone sampler.) I'm not receiving any data
so the PCI requirements are cut in half just due to that reason.

Just my random thoughts,
Mark