I have a project with a huge cube, each side if it consists of panels.

Some of them are on the other side of the cube, so they are not visible( they are rendered and eat FPS, but not visible for the camera).

First, i thought about a trace from every panel to camera to know, if it is visible... But it isnt good idea, as i'd need as many traces, as many panels i have.

Second idea - it to calculate camera's angle... With the help of it, i increase FPS from 80 to 120-140, that depends of the visible blocks (as other sides of the cube are invisible)...

So, the question: is there a build-in(or is it possible) function for culling models, that invisble for camera?

Something like this: http://www.codesampler.com/dx9src/dx9src_7.htm#dx9_occlusion_query

the only culling i see is engine doing is the basic view cull that removes backfacing polygons... i wont ave much to say on this topic tho )

The only built in occlusion function is the BSP tree. DirectX Queries, as in the sample code, require the object to be drawn and thus can't be used for occlusion tests.

In your case the obvious criterion is not the occlusion, but the orientation of the panels. Surfaces facing away from the camera are not drawn by DirectX. Make sure that all panel surfaces face outwards.

I also met with frustrum colling, but that was bad experience: i have had a pyramide of boxes with shadows. So, if moved camera close to the pyramide and watch on the shadows (pyramide wasn at my back), some shadows disappeared, because entities, that droped shadow were culled

In your case you could only have one trace per side of the cube, right?
Since all the panels of that side would be visible or invisible anyway...
That's only six traces and wouldn't take up that much calculation time and always deactivate at least 3 sides of the cube.

Just a thought.

jcl, but i'd needed to run through all panels each frame(compare their orientation)... I'm afraid, it would be little bit slow.

I solved this, just checking position of camera (looking like its the fastest method), and hiding some panels, if camera.pan<a1 and camera.pan>a2
I was just curious if there are(or, probably, "would be") some kinds of culling in nearest time? I think that it could be useful in large numbers of another cases, where camera position would'nt be enough.

Hitsch, yes, its a good idea, to trace from the middle of the every side

hm.... this is a mix of 2 methods: to check the rotation of middle cell

Cude is a single model?

no, it consists of a "cells"
BTW, number of "cells" can be different