Faster card, slower graphics!

Here you go. Note that the "discontinuity" error continues until I kill the xbmc process.

http://pastebin.ca/628699

Regards.

Both your glxinfo and xbmc.log look fine to me. I'm not sure what's causing the slowdown. Make sure that the .so files are up to date and that you're running the latest svn. If that doesn't fix it then maybe this needs to be a bug report. At what fps does it play when you 'Tab out' of fullscreen mode?

I tried this again with the latest SVN and still get very pour FPS during video playback. Sound appears to be ok, but frames are less than 1 FPS.

Here is my xbmc.log:
http://pastebin.ca/633402

I have the same problem...

/frogge

Hi, I have the same problem on my laptop (Nvidia Geforce FX5300 Go, with a 2.8 Ghz Pentium 4) and have been trying to figure out whats wrong. It seems to me that the problem is caused by the FX series lack of OpenGL features.

This a small part of my laptops glewinfo output:

GL_ARB_texture_non_power_of_two: MISSING
GL_ARB_texture_rectangle: OK

This is a very annoying problem, since in this case nvidias OpenGL driver is reporting the existence of the texture rectangle extension (responsible for drawing textures where height and width are "not the power of two"), but its missing the hardware acceleration of non_power_of_two rendering. These issues are further explained in this paper:

http://download.nvidia.com/developer/Pap...upport.pdf

The highlights of this paper are (page 9 - 10):

1. GeForce FX and NV3xGL-based Quadro FX GPUs lack hardware support for non-power-of-two textures (excepting the texture rectangle functionality provided by the ARB_texture_rectangle extension). If any texture unit could sample a bound 1D, 2D,3D, or cube map texture object with non-power-of-two size, the driver will automatically render with software rendering, which is correct but extremely slow.

2. Unlike NVIDIA’s GeForce 6 Series and NV4xGL-based Quadro FX GPUs, GeForce FX and NV3xGL-based Quadro FX GPUs do not have hardware support for fragment-level branching.

3. GeForce FX and NV3xGL-based Quadro FX GPUs lack hardware support for separate RGB and alpha blend equations. If the RGB and alpha blend equations are different, the driver will automatically render with full software rasterization, which is correct but extremely slow.

4. NVIDIA’s GeForce FX and NV3xGL-based Quadro FX GPUs lack hardware support for vertex fetching. GLSL vertex shaders that perform vertex texture fetches will fail to compile.

and

Future driver revisions may successfully compile GLSL vertex shaders with texture fetches but perform the vertex shader completely or partially with the CPU. In this case, the GPU can still accelerate rasterization and fragment processing.

From reading the source of LinuxRendererGL, I'm not quite sure how much of this affects the current renderer, but from my own test the problem is not solved by reverting to Power of Two textures and I believe the reliance on vetrex, fragment and pixel shaders in the rendering pipeline. Since I'm not really familiar with OpenGL programming I'm currently unable to build an alternative rendering pipeline, and don't quite understand the purpose of the shader code

Just thought Id share my findings with everyone else having the same problems.

Glewinfo: http://pastebin.ca/638426

GLXinfo: http://pastebin.ca/638429

Thanks HenrikDK.

That definitely sounds like the cause of this problem. I have updated the renderer to fall back to use the texture rectangle extension and power of two textures if hardware support is not detected.

@d4rk, wow that was fast looking forward to test the new improved rendere when i get back from work