OpenGL Point Rendering.
This page contains some new timings for the point rendering program used
in the paper
"Hardware Accelerated Point Rendering of Isosurfaces". The new timings
have been
performed on an NVIDIA Quadro 4 graphics card using vertex programs to scale
the point size.
The main reason for repeating the timings is that the old timings were
made using a Geforce2 graphics card without vertex programs. Hence, point
scaling
had to be done differently - in the way explained in the paper. Also, the
primitive
generation timings turned out to be slower when creating (large) display
lists. In
these tests, display lists are turned off.
The vertex program for point scaling is here
Platform
icc 7.0, Linux 7.3, Xeon 2.2 GHz, NVIDIA Quadro 4 900 XGL
Primitive generation timings
Primitive
|
BEAR
|
SKULL
|
ENGINE
|
Points
Triangles
Strips
|
0.9
6.7
37.4 |
3.8
13.5
1355.5
|
3.5
13.8
180.1
|
|
seconds
|
seconds
|
seconds
|
Frames per second
Primitive
|
BEAR
|
SKULL
|
ENGINE
|
Points
Triangles
Strips
|
38.9
14.9
28.6
|
14.9
3.6
6.9
|
38.1
11.9
26.2
|
Conclusions are the same: It is faster to generate points and to render points.
It will always be
faster to generate points, since point generation is a subproblem of
triangle generation. It can be
faster to render triangles, but, in general, point rendering becomes more
and more attractive
the larger the models we have to deal with.