I appreciate the feedback guys - I'll have a full report after the
weekend.  BTW - I'm taking care to get results on win2k and winXP.

b

On 2/9/06, Victor Lazzarini <victor.lazzarini@nuim.ie> wrote:
> It is easy to see why you get a worse performance with
> kr=sr. It's
> as if all your signals were audio signals. More calculation,
> more
> computing time. If you need kr=sr for parts of your
> instruments,
> try to isolate them as UDOs and run a local ksmps=1.
>
> You can check it out by simply looking at your task manager
> CPU usage graphs, compare the same orc with different values
> of kr. Ksmps=64 seems to be a good choice, but I often use
> 100. PD,
> for instance, uses a default of 64 samples per block.
>
> I'm not really sure why CsoundAV seems to work well, while
> Csound5
> does not, as I said, this could be a winXP thing, whereas in
> my win2k
> box, all's well.
>
> It's fair to say though, that for these low latencies you
> need ASIO, as
> the MME in win2k/XP does not like small buffers, as it puts
> the audio
> to a software mixer rather than directly to the soundcard.
> On win95
> MME worked with much less latency, in fact.
>
> Also make sure that whatever you are doing you're not using
> more
> than 100% CPU (denormals are a tinker for doing that).
>
> Victor
>
> >
> > <snip>
> > > On my win2K box, I run csound5 with ASIO to a m-audio
> > > quattro USB with lowest latency setting and low settings
> > > of -B and -b. The best latency in town as far as
> > > synthesis systems on Windows are concerned.
> > >
> > > Victor
> > </snip>
> >
> > :) I know - as you've said, and I am jealous, because it's
> > not working as well on my machine. There are more than a
> > couple of permutations I have not yet tried, but it really
> > should work with the same buffering settings as I use with
> > CSAV, shouldn't it?
> >
> > I have heeded Istvan's advice, and have tried all
> > combinations of -b and -B one might think of, and
> > monitored the control panel of my soundcard/interface as
> > the case may be, to see that -B (samples per hdwr buffer)
> > settings are reflected, and still no improvement.  I'm a
> > fulltime programmer and have been using csound actively
> > since 97, so I would THINK I'd be able to track it down by
> > now, but I'm at a loss. :)
> >
> > Things I still need to try:
> > 1) I have a suspicion that the ASIO stuff installed with
> > Cubase is conflicting somehow with csound.  I obviously
> > don't know enough about ASIO/VST to articulate that
> > accurately - part of the problem.  I need to take a day,
> > uninstall Cubase and anything associated with it, as well
> > as all csound stuff, and see if performance changes with
> > ONLY cs5 and drivers for my soundcard.
> >
> > 2) I don't understand why changing the ksmps to 1 (kr=sr)
> > makes things so much worse in CS5.  If I have kr=sr, the
> > crackling from latency is really bad.  It improves
> > somewhat with ksmps=10, but not much.  Here's my test
> > orc/sco
> >
> > orc:
> > sr=44100
> > ksmps=10
> > nchnls=1
> >
> >     instr 1    ;untitled
> >
> > ain1    inch    1
> > out    ain1
> >     endin
> >
> > sco:
> >
> > i1 0 3600
> >
> >
> > 3) I may have a stray portaudio dll on my machine that is
> > introducing the latency.
> >
> > =-=-=-=-
> >
> > Also - just to be clear - RT output of _any_ orc/sco I've
> > written is fine.  It's input that is the problem.  I would
> > expect to hear the signal produce by the above orc/sco to
> > hit the output clearly with less than .1 sec of delay.  Am
> > I expecting too much?
> >
> > Viktor:  Are your pieces always processing RT input?  I
> > could see that if your music is not rhythmically active,
> > you might not notice the latency we do.
> >
> > Thanks for the feedback -- and thanks, Bill, for piping
> > up! b
> > --
> > Send bugs reports to this list.
> > To unsubscribe, send email to
> > csound-unsubscribe@lists.bath.ac.uk
> --
> Send bugs reports to this list.
> To unsubscribe, send email to csound-unsubscribe@lists.bath.ac.uk
>