Horde3D

Hello all,

I've downloaded the demo and seem to be getting very low frame rates.

The Chicago demo yields frame rates of ~30 fps while the physics and particle demo each yield ~60fps. My system specs are: AMD 4600x2, 2GB RAM. Video card is Nvidia 7800GT. Are these frame rates typical? They seem very low relative to the geometry being displayed.

I did notice that occlusion culling is yet to be implemented, but frustrum culling and LOD don't seem to affect frame rates either. The only noticeable change in frame rates occurs when I pan the camera to a completely empty (black) area and even then the rates seem to top out at 60.

Any thoughts?

Thanks.,

Hello Hanglide,

you have probably enabled vertical synchronization (V-Sync) for your graphics card which means that the graphics card waits for a signal of your monitor (usually 60 Hz for flat screens). You can disable this option in the driver settings.

On a similar system like yours, I get 250 FPS for the Knight sample. The Chicago sample runs with 50 to 60 if I recall right, I will check that again. The Chicago sample is very demanding because it renders 100 characters with full skeletal animation, ambient pass and two light sources with shadows.

Thanks for replying Marciano,

I had already set Force Sync to off to with no effect but found that I had to explicitly set the checkbox to "use adavanced settings" in the "adjust image settings with preview screen"

I'm now getting 45-90 fps in chicago and 100-300 in Knight depending on view and zoom level.

It may be that I don't understand the intent of frustum culling, but
shouldn't framerates shoot WAY up when the view includes very little polys? I can pan and zoom to display only the stone floor and/or the black "sky" without the frame rates changing significantly in the Chicago sample. Am I missing something?


Thanks again for the help.

The problem with the culling are the shadows. Even if you look into the sky the light sources still calculate the shadow map. If you disable shadows you get your expected boost.

This is a point which still offers room for optimization, I will think about it. I also just managed to speed up the Chicago demo quite a bit with some low level optimizations (aligned GPU memory accesses).

Yes, the framerates should increase when there are less polies on the screen - but only if the vertex pipelines are currently acting as a bottleneck.
In the chicago demo, the main bottleneck is actually the fragment pipelines (i.e. the pixel shader), so in order to increase the FPS, you need to reduce the amount of pixels being rendered. The only surfaces in the demo which do not stress the fragment pipeline are the black backgrounds (so you should notice a slight fps boost when looking straight upwards).
Due to the amount of processing that goes into each pixel, the best way to increase the FPS is to lower the screen resolution.

However, as marciano said, when shadows are enabled, the scene must be rendered from the view-point of each light source - so even if you turn your camera away from an object, if a light is shining on it then it must still be rendered (to an invisible shadow texture).

I have optimized the engine some more. Now the lighting calculations for forward and deferred rendering use a scissor rectangle to save fillrate (quite a plenty when light is small or far away) and there are tests which check whether the shadow map needs to be updated. Now looking into the sky gives you a much higher framerate when not standing inside the light volume. Some other high level optimizations also help to increase performance under certain frequent conditions.