What the best for nvidia ?
I read some days ago that nvidia cards prefer this for a best rate:
1. allocation in the vram (with *gl*AllocateMemoryNV)
2. use strip or fan triangles or quads
3. call glDrawRangeElements()
Okey, but:
. i don''t use strip still i can''t.
So, that''s my question:
Have i to use interleaved arrays, or stay with my separate arrays ?
I think glDrawRangeElementExt would be better (that sounds normal, isn''t it ?)
does using glDrawArray is a good idea ?
And last: We said me not allocating in vram the indexes and the materials: why ??
Fratt
The performance FAQ is unfortunately outdated with new drivers. Most hardware dependent figures are still correct (eg. fastest primitive, texture format, cost of texGen, etc), but the extremely restrictive number of supported ''high performance'' vertex array formats has been extensively relaxed in current driver revisions. Also note, that the hardware performance data is *only* valid for GeForce2. GF3 & 4 show totally different behaviour.
Despite of the numerous times I asked nVidia to update the performance FAQ, and the numerous answers like ''sure, the updated performance FAQ is ready, we''ll put it online next week'', there is still no such thing available.
Perhaps someone should ask Carmack to ask them, this would definitely work better
> And last: We said me not allocating in vram the indexes and the materials: why ??
This one is not answered in the FAQ. Textures (if you mean that by materials) are *always* allocated in vram, but this is entirely handled by OpenGL, you shouldn''t interfere with it. Just use the standard GL texture functions.
For index arrays, current GeForce chipsets do not have the capability to pull them from vram/AGP, this only works with vertex/vertex-attribute arrays. Since index arrays are ''normally'' read and transfered, storing them in AGP ram would slow you down, since it is uncached, and should *never* be read by the application.
A.H aka Blueshift
Despite of the numerous times I asked nVidia to update the performance FAQ, and the numerous answers like ''sure, the updated performance FAQ is ready, we''ll put it online next week'', there is still no such thing available.
Perhaps someone should ask Carmack to ask them, this would definitely work better
> And last: We said me not allocating in vram the indexes and the materials: why ??
This one is not answered in the FAQ. Textures (if you mean that by materials) are *always* allocated in vram, but this is entirely handled by OpenGL, you shouldn''t interfere with it. Just use the standard GL texture functions.
For index arrays, current GeForce chipsets do not have the capability to pull them from vram/AGP, this only works with vertex/vertex-attribute arrays. Since index arrays are ''normally'' read and transfered, storing them in AGP ram would slow you down, since it is uncached, and should *never* be read by the application.
A.H aka Blueshift
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement