Stress Test

[CodeGen]

Very interesting !
But wouldn't it be good if the engine could perform necessary optimizations itself ? ( Some scene graphs already can combine compatible objects ( merge nodes containing identical geometry ( as is your case ), transforms, effects, etc. )

i think it would be nice if the engine could give a warning, if it could select, i don't know, LOD optimizations for achieving target frame rate.
( i guess it's almost impossible until some sort of human-like AI is involved. )

Adding a validation pass before scene is rendered would be easy. During that pass the scene graph will be checked ( no cyclic references allowed ), the engine would issue warnings, errors if something's wrong, etc. After validation we can optionally run optimization passes.
Would be interesting to see such an engine in action and what it translates your scene into.
No, of course, that's impossible (?), just crazy stupid ideas...

[bitplane]

Not really something we'd want to add to the engine itself, but a scene optimizer would make a very interesting side-project

[grumpymonkey]

wow I tested it with that exact code and it only used about 1,200,000K of memory and I got around 1-2 FPS which is surprising since at the same time I had 74 processes running at the same time, including a music player plus, I have windows vista (64 bit) with only 2GB of RAM

I think its lagging alot because irrlicht has to render every single node individually. It would probably be a lot faster if you combined all the cubes into one mesh or something, that way irrlicht would only have to send 1 render call for all of them.

[yamashi]

Using culling algorithms, crowded rendering it should be possible to get a decent frame rate...

[sio2]

I just did a very quick and dirty test using hardware instancing (based off a D3D sample - completely non-Irrlicht) and I get:

1 million cubes @ 7.3 fps
100,000 cubes @ 50.9fps

Peak memory usage is 77MB.

[sio2]

FYI: I just tried a DX11 card and got 26fps for one million instanced cubes.