Re: [Visualpython-users] speed

[Bruce S. <Bru...@nc...>]

I'm pretty much out of my depth here, but Guy Kloss's nearly simultaneous and 
very interesting post about speed suggests that there is no royal road to speed 
+ rapid creation. For that matter, there's only so much you can get out of 
today's graphics cards.

For example, the NVIDIA Quadro FX 4600 (about $1400) is rated at 250 billion 
triangles per second (http://www.deskeng.com/articles/aaajxm.htm). Suppose a 
small sphere is approximated by 4 triangles facing you. Then the card would take 
a second to display 60 billion small spheres. To have good mouse navigation 
capabilities you would need to be able to redraw the scene in about a thirtieth 
of a second in response to a mouse move, which means that you could rotate and 
zoom only about 2 billion spheres. And that's assuming that no computation is 
being done: you generate the 2 billion spheres once, and then wait to react 
quickly to a change in camera position.

A midrange card like the $600 Quadro FX 1700 does only 190 million triangles per 
second, which implies about 2 million navigable spheres.

Bruce Sherwood

Tim Smith (25121) wrote:
> Hmm. Would I benefit from writing the application purely in c++ (not my
> preferred option)?
> I'm treading a line between speed and flexibility of pumping out various
> models, and speed of interpreting the model itself. If the model is
> frighteningly slow to interact with then I'm sunk, but if it takes me
> too long to produce a new model I'm also sunk.
>