Bullet 2.73 has been released, download from http://code.google.com/p/bullet/downloads/list
For 8192 fast moving objects on NVidia 8800 GTX, Intel Core 2 3Ghz: CUDA (btCudaBroadphase) 6ms, OPCODE Array SAP 37ms, Bullet dynamic BVH (AABB tree, btDbvtBroadphase): 12ms. When nothing moves, the CUDA broadphase still takes 4ms, whereas SAP and dynamic BVH (btDbvtBroadphase) are practically 0ms. For even larger amounts of moving objects (64k) the CUDA still performs fine, whereas SAP/dynamic BVH grind to a halt. This makes it a good candidate broadphase for environments with huge destruction.
- Check this forum topic to see what’s new
Thanks to everyone for feedback, bug fixes and improvements!