I think it *is* also worth mentioning that different CPU 'monitoring' tools provide different results: e.g. here I checked qjackctl's dsp load, top (and variants, atop, htop), and included xfce task manager and the difference especially what top and xfce task manager reported for cpu usage is pretty dramatic.

I do have a laptop with Intel's so-called 'hyperthreading' and was reading that readings about CPU usage etc. can be quite weird in this case..

Lorenzo

On 26/11/19 11:20, Will Godfrey wrote:
On Tue, 26 Nov 2019 00:07:50 -0500
Ted Felix <t...@tedfelix.com> wrote:

On 11/22/19 6:39 PM, Will Godfrey wrote:
Exactly what setup have you got there if you don't mind me asking?

   1st gen Core i3-370M.  I'm using "top" to measure CPU usage.

   I go with low spec hardware to make performance issues like this
easier to see.  This is how I was able to improve rg performance
significantly.

   To me the important point is that while idling, fluidsynth uses a
small fraction of the CPU that zyn and yosh use.  That means there is
definitely room for improvement.

   It's entirely possible that the issue is related to the fact that I'm
using whatever version Ubuntu 18.04 shipped with.  Maybe they built this
with optimizations off?  Maybe these are older versions?

Ted.

Hi Ted,

That explains a lot. It would indeed be compiled with no options at all. This
is a debian requirement to maximise the compatibility across a wide range of
devices, and from their point of view makes perfect sense.

If I install such a .deb here, I get around 20% and this is almost entirely the
low priority stuff (using htop), so your old setup is actually doing rather 
well.

It is also somewhat unfair to compare with a more-or-less straight sample player
like fluidsynth which has practically no overhead. While there is always room
for improvement, Yoshi/Zyn have to do a lot of work just to stand still, because
absolutely everything is generated in real time. This is why we try to push
as much as possible out of the RT thread. The more cores you have, the better
this works.

Paul Nasca's original design is quite unique. Lots of people have said that the
'problem' is all the FFT stuff and it should use a wavetable design... until
they look at exactly what it can do and the amount of real time variation there
is in actually sounding notes.

Will.

P.S.
Sorry. That's turned into a bit of a sermon :*)




_______________________________________________
Rosegarden-user mailing list
Rosegarden-user@lists.sourceforge.net - use the link below to unsubscribe
https://lists.sourceforge.net/lists/listinfo/rosegarden-user

Reply via email to