Jump to content
  • Advertisement

Archived

This topic is now archived and is closed to further replies.

NeoGL

How much worth is a BSP Tree now ?!!

This topic is 6226 days old which is more than the 365 day threshold we allow for new replies. Please post a new topic.

If you intended to correct an error in the post then please contact us.

Recommended Posts

I have a model that''s got around 2500 triangles. My BSP tree splits it up into 7500 triangles and then displays on an average 3000 triangles as I move around...sure it is displaying less triangles than it has in the tree but it still is greater than my original number of tris. So, I was thinking if I have a 32MB GeForce and i can transfer the whole geometry to the GPU (Range and Fence ext.) then where does BSP stand ? .I mean even if my BSP was rendering 1000 triangles, I would still get a higher frame rate by transferring my triangles to the video card and displaying all 2500 of them. So the question is (collision detection will need it) for display purposes, are BSP really worth it anymore ? NeoGL

Share this post


Link to post
Share on other sites
Advertisement
One thing you''re forgetting is fill rate. If your BSP tree is being used to eliminate overdraw, then that can be a big difference. I''m not saying the overhead of creating and traversing a BSP tree is worth the savings in fillrate for a 25000 poly model; I''m just not sure. It depends on the target video card and CPU involved, as well as your specific implimentation.

Share this post


Link to post
Share on other sites
I am assuming, the fill rate issue you are referring to is the clearing of the depth buffer. You are right TerranFury, getting rid of the z-buffer and drawing back-to-front with BSP could be a substantial leverage to the frame rate. But I have been playing around with it, and now i am not very sure if i really want to get rid of the z-buffer anymore. Two Reasons

1. BSP Tree has the problem of "polygon cracking" that shows up in places during rendering. Though it''s not a big deal and might not even be noticeable in all cases, but it''s still there.

2. While introducing dynamic geometry in the scene, boy i sure do miss the z-buffer....sure can merge the BSP on the fly as per Carmack''s article on BSP in Quake. But then again , is it all worth it ?

Each face takes about 84bytes of storage (incl. vertex normals and tex coords), on a card with 16MB of video memory, i can safely lock in around 10-12MB without hurting interface/desktop/dynamic geom. perf. So, i can stock away 100K - 150 thousand faces for rendering passes in the silicon. That''s almost a medium sized doom level....!!!

NeoGL

Share this post


Link to post
Share on other sites
I think he was referring to the overdraw of hidden objects. If you draw front to back, tho, that''s not so much an issue. And for the modest polygon reduction you''re getting, it doesn''t really sound so worth-while.

Share this post


Link to post
Share on other sites
Guest Anonymous Poster
Let your GPU do the work. It will be faster to let the dedicated hardware sort out the 2500 polys, then to let your CPU handle the 7500 split-up triangles and then still end up rendering 3000 of them. Even if you didn''t upload the geometry, rendering 3000 tris without depth test could be slower than 2500 with depth test. And you free your CPU for other tasks.

Share this post


Link to post
Share on other sites

  • Advertisement
×

Important Information

By using GameDev.net, you agree to our community Guidelines, Terms of Use, and Privacy Policy.

We are the game development community.

Whether you are an indie, hobbyist, AAA developer, or just trying to learn, GameDev.net is the place for you to learn, share, and connect with the games industry. Learn more About Us or sign up!

Sign me up!