zzzzrrr
Joined: 17 Feb 2007 Posts: 139 Location: Washington, DC
|
Posted: Tue Feb 26, 2008 9:36 pm Post subject: Simulation performance |
|
|
I've already received feedback on the demo performance, particularly in regard to how the Blaze 0.6 demo seems slower than Blaze 0.5.
This is in fact not really the case. The Blaze 0.6 main program loop includes 3 iteration steps per frame; Blaze 0.5 had one. This allows the sequential impulse solver to return more "accurate" simulation results. The greater the number of steps, the higher the accuracy.
If you're looking for raw speed, lower the number of iterations steps per frame in your main program loop. However, be advised that this condition will not be ideal for situations that involve many simultaneous collisions, i.e. stacking. If you have a simple need for single collision response, such as collision in zero gravity between dynamic objects with no resting contact, you should be able to get away with only one step per frame just fine...
Try adjusting the demo time stepping scheme and notice the changes in execution speed and stacking stability. It's an interesting trade-off... There are also a number of internal simulation variables that should be user adjusted... such as gravity and a few other engine specific performance variables.
One day I'll have to sit down and write a user manual if I ever have the time... Volunteers
Mason |
|