PDA

View Full Version : Better performance with indexed arrays



BlackIceCube
01-29-2003, 04:00 AM
Hi !

Do you know how to get better performance rendering a big mesh?
Right now I'm using indexes arrays (the problem only needs points and fog)
I've split the mesh in n sub-meshes (because only small regions inside are updated periodically) but without so much speedup.
some hints? ideas?

Thanks !

knackered
01-29-2003, 04:19 AM
You're rendering just points in fog?
Don't use indexed vertex arrays then...you're not sharing any vertices.
Use glDrawArrays instead of glDrawElements.

harsman
01-29-2003, 06:20 AM
Putting geometry in displaylists will probably help or you could use either nvidias vertex array range extension or ATI's vertex array object extension.

jwatte
01-29-2003, 07:50 AM
Are you sure that geometry transform is your bottleneck? Do you get a higher frame rate if you make your window smaller? If so, you're fill rate limited.

BlackIceCube
01-30-2003, 08:22 AM
Hi again !

I tested glDrawArrays without results..
The situation is:
about 3M points in a shell, but only a small % of them moving periodically (really updating their position), at 3Hz or less.
The first idea I got was to split the geom in several zones to improve performance.
Would I keep some of those zones in the (AGP Card) Memory instead of do a glDrawArrays ?

Thanks !!

zed
01-30-2003, 08:51 AM
do not draw 3million vertices in a sinlge call split it up into groups of about 4000 vertices.
prolly stick an octree inside your cloud and subdevide until u have about 4000 points in each node.
once u have this u can test what nodes are on screen + only draw those which will save a lot of performqnce