ChaiRuiPeng wrote:i am guessing you are manually limiting fps at 60?
Yes, but with 2 million polygons, I do not exceed 70/75 fps without sync (NVidia 8800GTX).
I can not rise above 2.3 million polygons, which is already well I think.
Radikalizm wrote: but maybe some shadows and some brighter lighting would make it even better
Yes, but there, it will be more complicated in terms of technical and performance ...
ChaiRuiPeng wrote:i am guessing you are manually limiting fps at 60?
Yes, but with 2 million polygons, I do not exceed 70/75 fps without sync (NVidia 8800GTX).
I can not rise above 2.3 million polygons, which is already well I think.
Radikalizm wrote: but maybe some shadows and some brighter lighting would make it even better
Yes, but there, it will be more complicated in terms of technical and performance ...
2.3 million? Is that because of the grass or because of terrain?
If it's the grass you should really do some culling there
for grass, I use hardware instancing and frustum culling mechanisms.
For trees, it is true that do not mesh 'low polygon', it's just to test the charge, and i use StaticArea class with LOD technique (instancing software somehow).
tmyke wrote:for grass, I use hardware instancing and frustum culling mechanisms.
For trees, it is true that do not mesh 'low polygon', it's just to test the charge, and i use StaticArea class with LOD technique (instancing software somehow).
oh so like stress test?
what about testing the software on slower hardware?
i always test my hacky optimizations on my netbook and if it runs at least on 20/25 fps there i know it is pretty good
ent1ty wrote:
success is a matter of concentration and desire
Butler Lampson wrote: all problems in Computer Science can be solved by another level of indirection
For many entities low polygon, then it seems that the hardware instancing is very best. Good management of the VRAM can be important in this case.
For heavier entities and more dispersed on the ground, the second technique seems to work well (with using VBO recess, and LOD + frustum culling).
After all depends on the power of the video card for your netbook.
I think your trees would be good candidates for an impostor system. Basically it switches the models to billboards that are updated only every so often, but at a distance you can hardly tell. There was a paper on using impostors for trees somewhere, but I no longer have the link
I do not know if it comes together, but I use a LOD system, or the entity's most remote
groups are represented by a sprite.I'm still working on it to improve overall.
Impostors only come in handy for objects at great distance. As for trees I'd go with the replace-with-billboards-at-some-distance approach. Cause still in the case of a forest the impostor-way would mean a lot of re-rendering of objects.
It's not "only handy for objects at a great distance," it's actually handy for when you have a lot of objects at almost any distance (save very close up). The advantage to impostors is that it's extremely hard to tell that they're impostors. Replacing with just a billboard is very obvious, and the overhead for rendering to textures with impostors is more than worth it for how good it looks compared to simple billboards.