Opengl Performer Scene Graph( FAQ part 1 ( 1 thru 50 )

On Windows run as administrator
On Unix systems such as Irix or Linux run as Root
If you cannot run as root then use account with root privileges for your application
Run Full screen and border less, windowed application are slower than full screen
Run only one Window per pipe, use multiple channels rather than multiple window
On Multiprocessor System lock processes to a process and give them non-degrading priorities
Only run your application
Don't run multiple graphics program on the same pipe graphics context switching is expensive
Run the release build of the application
Close all none essential services or program
Close any clocks, gr_osview, or other tools that may wake up and draw themselves
Close any messenger servers such as MSN or Yahoo
Close your virus checker ( or at least exclude the files in you Sim from the scan)
When multiprocessing, make sure the executable is on a local file system.
Store data on local file system if possible rather than network drives
Convert you models to a fast loading binary format such as pfb, vsb or fst etc.
Convert your textures to DDS available
Preload you texture if possible
Preload your models if possible
Reduce the ratios for the near and far clip plane use
Avoid sub-facing or decals, cut them in polygons are cheaper to draw
Keep textures to a power of 2
Keep texture size as small as possible
Don't exceed 512x512 for texture size currently cards struggle to move larger sizes
Combine textures together if possible 1 512x512 is better than 20 32x32 images
Keep the number LOD nodes to a minimum ( for APP/CPU bound applications )
Keep the number DCS/DOF nodes to minimum ( for APP/CPU bound applications )
Do flatten and clean you models (pre process to a fast binary format)
In your 3d models create a deep hierarchy rather than a flat
Make sure you models contain no no-planar or co-plane polygons
Use textures for detail rather than polygons
Add system memory the more the better 1gb min 2gb better more is great (memory is cheap)
Graphics memory get the best card you can 256mb min
Get a good CPU
Multiples CPU for Vega/Vega Prime and Performer really improves and shows there prowess
Use fast and local hard drive
Defrag your hard drives
Make sure you have the latest graphics driver ( although some times you may need to step back versions)
Turn Sync to Vertical retrace off on your card (if on you can only run as fast as the vertical retrace)
Avoid using glReadPixels as frame buffer reads are typically an expensive operation
Disable C++ exceptions in your compiler this could get your a 15% improvement
Try to keep the FOV small and the far clip as close as possible
Keep the number of Light sources to a minimum
Try to avoid deleting large time many times, if possible remove form the scene and keep in memory if need again or use a deferred
deletion scheme
For very large area databases you as mechanism such as LADBM and Virtual/Clip texturing
Avoid IO ( printf , cout etc ) were possible on windows create a non-console as its expensive
Try to share material and colour state were ever possible ( Pre-process and again save in a fast binary format)
Try to use create geometry that can use triangle strips or meshes ( Pre-process and again save in a fast binary format)
Make sure you are not generating any floating-point exceptions in your code
Put only latency-critical tasks between the applications AppSync() and drawFrame() calls
See you run-time API documentation for specific things you can do to improve performance with that API
Keep to a mimimun the use of Isectors each frame, these are quite expensive