Irrlicht Engine

"5000 randomly positioned sydney models, 3395000 triangles, this produced about 6 fps "

perhaps I am missing something, mostly as am an artist, but in my experience, that is a real load of models(which often can kill more than tris amount) and tris. Really think you may have very good hardware...

Is not only even good for an indy engine...

But ok, if can be improved, even better

I mean, to me is a good news it can handle *all* that at 6 fps. Before reading you , I'd have though that amount would be killing any machine of today...But am also very bad informed on latest hardware...

vermeer wrote:perhaps I am missing something, mostly as am an artist, but in my experience, that is a real load of models(which often can kill more than tris amount) and tris. Really think you may have very good hardware...

Is not only even good for an indy engine...

But ok, if can be improved, even better

I mean, to me is a good news it can handle *all* that at 6 fps. Before reading you , I'd have though that amount would be killing any machine of today...But am also very bad informed on latest hardware...

I have pretty good hardware. After getting the basics done I'll upgrade to a good system

You aren't missing anything, it's shitloads of stuff on the screen. You can only see fraction of the models. The problem is mainly that 5000 nodes directly responds to the ammount of stuff you can put in to your program. Even if you don't see the nodes you'll only get around 30 fps (on my system that is) which is seriously screwed. Add physics/AI/networking/whatever, you get the point

For reference, here's the main specs of my system, results were under winxp
Windows XP/Slackware 10 dualboot
Asus A7N8X Deluxe Gold, Watercooled
Athlon XP 2400+ @ 2400 MHz, Watercooled
1GB 400MHz DDR dual channel mode
ATI Radeon 9700 Pro Oced, Watercooled

About DX7/8 and older machines. I'm sorry to say but from my end you're screwed. We're aiming for next-gen engine so prev-gen hardware ain't gonna cut it We only use OGL since DX aint portable.

Also I thought about the optimizations more and I have a list of stuff to optimize/add (I had some more but forgot to write them down)
Better frustrum culling
kDTree scene handling
Coherent Hierarchical Culling
Texture/Mesh memory optimization (textures keep copy of the image data)
Texture compression (compress the texture data instead of making all images 32 bit, if the card can use compressed textures that is)
Vertex memory optimization (make a flexible vertex class that doesn't store unneeded stuff)
Progressive Mesh
Better LoD system for terrain

I think everyone can guess that was me, forgot to login

the first thing I did was disabling automatic culling because this does NOT work for me (with my own camera), second you should focus on only one rendering API and optimize it (VBO, compiled, etc.), another point is storing your models in only 1 buffer (when you have 1000 objects you have 1000 vbo's, try packing more than 1 into one buffer and merge/generate new texture-skins (this is hard but gives the best speed up)