PocketMatrix

i wrote a little performance test for the X50v and i'm a little bit disappointed of its rendering performance.

here's my benchmark:
i render 8 teapots with 1024 triangles each. the mesh is an indexed triangle-strip with 81 strips - so those are quite long strips on average. i activated two lights and no texturing. all mesh data is passed as fixed-point and the indices are passed as unsigned short using glDrawElements(). from my point of view there is nothing that can be optimized here since the 2700G does neither support glMultiDrawElements nor vertex buffers.

what i achieve is this:
rendering the teapots with two activated lights i get 17fps which results in 140k triangles per second. turning of the lights i get 21fps which means 180k tris/sec. both is far away from the 800k tris/sec that intels docs tell...

i attached the rendered image...

any ideas on this?
thx,
Daniel

What FPS are you seeing rendering only wireframe?

Dan East

the fps for lines are a lot lower, since my OpenGL wrapper library emulates glPolygonMode() by rearanging the indices and drawing GL_LINE_LOOPs.

in this case it's about ~50% of the filled speed.

Daniel

If you want to build your test program with STANDARDSDK I'll test it on the ATI 2300 running on an XScale 400MHz.

I don't think I can do it today, but I should be able to this week.

Bandwidth is almost certainly an issue, because we ran into that with the 2300. Geometry caching is critical.

Dan East

Dan: Which library does the ATI 3200 provide for 3D graphics? OpenGL|ES CL 1.0 ?

Daniel