Subject:
|
Re: review: Radeon 7000 for BrickDraw3D, low-end Mac
|
Newsgroups:
|
lugnet.off-topic.geek
|
Date:
|
Thu, 11 Apr 2002 14:18:35 GMT
|
Viewed:
|
211 times
|
| |
|
|
In lugnet.off-topic.geek, Andrew Allan writes:
> At present, I am writing my own OpenGL renderer for LdGLite (Mac). In order
> to spped things up I am defining the parts within OpenGL and then calling
> then by their definitions. Theoretically this places the entire model within
> the Graphics Card memory.
>
> I started the process in the main memory by fully parsing the files into
> thier primitives, including final vertex locations, vertex normals and color
> as well (very memory inefficient - very small files require large bits of
> memory). However on simple models I could achieve redraw rates of 60 fps on
> my G4/400 - ATI Rage128(PCI) graphic system.
>
> By then loading the entire model into OpenGL and calling it by definition,
> the same system achieves 62 fps. Therefore in highly parsed models, my
> conclusion is the system bus on the G4 is not interfering with card by too much.
Cool! If you can clearly show where you make the changes, I might be
able to port this back into the other versions. I've never been all
that happy with the rendering speed myself. You can see in the code
where I tinkered a bit with display lists, but on my junky old hardware
with software opengl it didn't make much of a difference. Also
display lists seem to be on the way out in the discussions about
OpenGL 2 so I put it on the backburner.
> However, the system is designed to work as follows, each individual brick in
> the model will be saved to the graphics card by name (including color and
> face normals). Then when displaying the model, the system will send the the
> card inly information about the bricks orientation and spatial position. I
> am hoping that this will improve the rendering speeds over those in Don's
> original code used in LdGLite as to date this has been my dissapointment in
> my Mac implimentation of his program.
You'll have to watch out for the optional lines with this approach. To
calculate when to draw them you need to convert the points to screen
coords to do the test. Look at render_five() in stub.c to see what I
mean.
> In conclusion an approach like this should also work wonders on any graphics
> card as it keeps all the rendering and model information in the graphic card
> this minimising traffic to the host computer. I will keep you posted as to
> progress.
I'm looking forward to it.
Don
|
|
Message is in Reply To:
26 Messages in This Thread:
   
- Entire Thread on One Page:
- Nested:
All | Brief | Compact | Dots
Linear:
All | Brief | Compact
|
|
|
|
