I am an OpenGL programmer for so long now. I am using NVIIDA’s Quadro FX500 graphics card. I am looking for profiling tools to actually measure general OpenGL application performance, GLSL shader performance etc… A tool that can if possible identify the bottle neck, show the driver generated assembly code, vertex/pixel shader throughput, gives information about number of cycles, temp registers used etc…
Also any information about CPU - GPU interaction, whether is depends on programming language etc…
Any help pointers of advice would be greatly appreciated. Thank you!
I really would like to see a standardized profiling extension to enlighten the black box GPU a little bit providing a decent standardized tool for profiling/optimization (dynamic load balancing/LOD?).
I’m still dreaming of the possibility to query for some basic GPU performance/stalling informations and accordingly dynamically switch at runtime between Scene, Mesh, Shader and Texture LODs…
A good place to start is to take a look at Ian Buck’s GPUBench .
Although, this is a really nice tool to profile the different GPU operations in a particular GPU/Driver setup - it does not assist you in profiling a particular application.
Originally posted by SirKnight: An OpenGL version of NVPerfHUD will be out early next year. I know this doesn’t help right now but it’s something to look forward to I guess.
Will it be windows only or will be a linux version too?
Good question. Because I’m not sure. I’m going to guess it will support both OSes. Usually NV’s opengl demos work with both so I’d assume this will too.
Originally posted by SirKnight: Good question. Because I’m not sure. I’m going to guess it will support both OSes. Usually NV’s opengl demos work with both so I’d assume this will too.
The last version of the nvidia sdk was 6 and there isn’t no tool support beside the driver. I haven’t figured out how to set GLSL to strict under linux so that it’s doesn’t compiling wrong GLSL sources.