[Uni-verse] Audio demo situation?

[Peter Lundén]

Why do we always do quantum leaps in complexity. First we got a model
with slightly more than 700 polygons, refered to as testscene. I thought
this model was quite nice for demoing the audio stuff, although I dont
know how it looked (I have still not been able to run Quellsolaar or any
other renderer except for the simple one). It works quite good, no
complexity problems.

The new model is almost 70000 polygon, for UVAS this is over 160 times
more complex (n*logn). Now we got serious problems to handle this
complexity in various Uni-Verse components (dont know how the geometry
reduction together with UVAS is handling this yet). I know UVAS cant
cope with the complexity by its own.

My conclusion is that we should stick with the simpler testscene model
and maybe add some furnitures and stuff for the reducer to eat. Its
better to have a simple working model than a complex one that is
problematic. I think complexity do not add so much value to the
demonstration. The main thing is to show that we can do real-time
acoustic simulations which has not been previously possible. It would be
nice to be able to make real-time simulations in complex buildings but
it is obvious that we are not really there yet.

Best regards,
--PLu

Emil Brink wrote:
> [...]
>> 5) Rendering Have you tested the latest FHG renderer on the web site?
>> Have you tested Quel Solaar on your HW?
>>
>> 6) Many different models exists. Quality of the model is unclear.
>> Patching to be watertight may be a problem More complex and visually
>> pleasing models would be needed.
> 
> I've done some testing with the first model from Slas' page, this one:
> <http://www.pdc.kth.se/~slas/universe/models/model-snapshot-061108_0129_r1.vml>.
> 
> 
> I've only run Connector, and so far, results are ... making me rather
> nervous. :/ Performance is rather low, and I don't know what can be done
> to improve it.
> 
> Some concrete issues:
> 
> 1) I noticed there is a lot of material in the scene (949 nodes), perhaps
>    some objects should be joined into larger meshes? I'm not sure (Eskil
>    can correct me, here), but I think that is more efficient.
> 2) What about sharing geometry? Browsing the node list in Connector, I saw
>    at least six identical-looking (to me) chairs. Are they distinct objects
>    in Max too? I would hope they could be expressed by linking to a single
>    chair geometry, from a number of object nodes. Any comment, Slas?
> 3) I never saw the interior of the room get rendered, in color. This
> screen-
>    shot: <http://unwind.se/connector-ist-model.png> took many minutes of
>    processing to generate, but it has blank walls. Is this expected?
> 4) Connector uses a lot of memory, at least in Linux I think it is still
>    leaking. I can look for that, but not very efficiently. When I made the
>    above screenshot, Connector was using over 1.2 GB of virtual memory ...
> 
> Any ideas on what we can do to improve the situation are welcome, of
> course.
> 
> Regards,
> 
> /Emil
> 
> _______________________________________________
> Uni-verse mailing list
> Uni-verse at projects.blender.org
> http://projects.blender.org/mailman/listinfo/uni-verse
>