> Most of the machines today will be able to draw on the order of 1000 to
> 10000 polygons per second (That's only 100 to 1000 per frame!!). It
> doesn't take very many trees to use that up.
> I agree that bandwidth can be decreased substantially. However, I don't
> think that most of us will be able to display models like Barcelona in real
> time (remember - only 3300 polygons). At most, Barcelona could use
> inheritance for only a few chairs.
> I'm not against adding caching, but I wouldn't want to spend too much time
> on it up front because I think that the models we will be able to display
> and work with interactively will not be all that big for the next year or
The point regarding display capabilities a really good one; most interesting.
I think that concidering caching and perhaps inheritance/coagulation now is
probably fairly wise because it affect the way in which we write the
object-references; an intrinsic part of the system. The barcelona example (re:
inheritance) is a good one for the envions which model the real world exactly,
but am not so sure that it holds true for an entirely constructed world in the
moo/mud style. I'm also not sure that we actually need to be able to do
realtime "full-rendered" walk throughs ala doom on low capacity machines; is
realtime walkthrough vs. scene realism is a user and platform made
tradeoff ? Many CD games which provide pre-rendered single viewpoint
scenes done in 3DS, etc, provide a high degree of immersion, even when
they take 5 or even more seconds to change between scenes. I can see a
machine overloaded by a scene rendering an initial scene to a degree of
realism in 5 secs, using wireframe animation for movement, them during
periods of inactivity, "filling in" detail it didn't render while moving. I'm even
quite happy to see my trees as sticks while walking about, only to have them
turn into real trees if I hang about and look at them for long enough. It would
be possible to have several different models of stuff hanging about, for
different rendering scenarios.
In the end however, I beleive it is more important to pick off a techinically
dooable goal; if that means canning fancy stuff like caching, inheritance, and
co-agulation, in favor of getting something working, so be it.