The new Bullet 2.77 Physics SDK features OpenCL cloth simulation, contributed by AMD under the permissive ZLib license. The OpenCL implementation has been tested on AMD and NVIDIA GPUs for Windows and Linux as well as the Apple OpenCL implementation for GPUs on Mac OSX Snow Leopard. The full source code and precompiled Windows executable demos are available for download.
opengl.org
I’m keeping track of Bullet physics, over the social media sphere or whadd-ya-call-it , twitter and whatnut.
This one got my attention OpenCL acc. Cloth simulation from AMD _
Why not offload almost all physic sims onto bullet devs instead? and just keep an active role in it and work tightly to integrate it into blender.
I think blender is on a good start with the GSoC project implementing bullet physics more into blender.
Truly interesting. IIRC Blender has its own cloth solver. And it is maintained/tweaked when needed. I think this code will become obsolete sooner or later. The new hair sym uses cloth code, I think. One more reason to implement Bullet’s approach: both cloth and hair syms will benefit from OpenCL!
doesn’t bullet also have sph fluid as well thats another thing Blender could benefit from, I know there is already sph fluid sim in Blender but this could be another approach to handling sims
IMO they /blender devs doing cloth and physic sim/ can never catch up with bullet. Bullet is being used in Sony movies, and PS3 games … it’s implemented in almost all autodesk suites already.
Sometimes I think it’s a shame that it takes to long for blender THE open source 3d suite to adapt.
but then it’s the question who’s gonna do it?
Maybe if blenders internal dev for physics is scrapped and Bullet is adapted 100% for that, blender can more and more be just about the GUI development.
bullet is mainly meant for games and realtime simulation though. i guess blender’s fluid, cloth, softbody,… systems have advantages like more exact solvers.
yes, of course it makes sense that the game engine uses bullet (also because of erwin couman’s history with blender) but it doesn’t necessarily make sense to use it for everything else too. maybe as a second option but not as a replacement of the current systems.
naaah, the only simulation in blender that is more precise and not there in Bullet. is the elbeem lib. the fluid solver.
Bullet Physics has a lot of shapes for rigid bodies, that blender doesn’t have. Also the cloth sim, I think is almost 1000x times faster than blenders.
here’s a video of AMD’s havok cloth sim
I highly suspect it’s the one that AMD “donated” to Bullet Physics.
It’s prolly just a question of time until nvidia joins in and donates their fluid sim.
I think, that these tech AMD and Nvidia intially coded in their hardware language has been scrapped and redone in OpenCL to become hw independent. that’s the main reason to give it away to bullet.
bullet is quickly becoming THE physic sim for vfx and games. it was used in mass scale in the pipeline when the film 2012 was done.
these kind of tasks are perfect for GPU parallelization instead of CPU. These multitude of small collision calculations.
I don’t see the issue here, adapt to what? Blender is already using bullet for Rigid Body Simulation and Aligorith is working on better integration. Yes, it would be nice to have that new openCL cloth and SPH thing, but it’s brand new so it’s not surprising that it haven’t made it’s way into blender yet.
As kakapo mentions bullet is developed for real-time simulation and speed usually means less accuracy (less quality). There are a lot of blender physics features that bullet can handle atm (particles, fluid, smoke…). Blender will need it own physic features for some time…
Btw. Havok is own by Intel so don’t think AMD has donated it to bullet…
wow just built it and ran the demos impressive. sph with opencl is nice. Hopefully parts of this will find it into blender. Good people are hopefully working on it. Maybe the physics sprint at the conference if there is one will produce some good results. There are so many projects going on I dont know how they could all go into the trunk with out messing each other up
bullet takes a lot of shortcuts to achieve a solver that scales linearly. this results in a much less realistic simulation. you can’t have mass ratios much bigger than 1:10, friction is not realistic (why can you pull out objects in the lower parts of a huge stack as easily as upper objects?), it is impulse based instead of using forces, there is jittering, the behaviour of the supposedly rigid objects is very spongy instead of rigid, stacks aren’t able to stand on their own for very long without sleeping mode,…
the physics in 2012 didn’t look very convincing to me. i guess there bullet was enough.
i am not saying bullet is bad though. the other game physics engines (except maybe newton game dynamics) have the same problems. i also think that bullet is the most promising one, since it’s open source and physx/havok only get abused as vehicles to sell hardware. there is no money to be made with game physics anymore.
it would be great if erwin included the more exact big matrix solver of ODE as an option. ODE is a bit crappy (buggy collision system, missing features,…) but its big matrix solver is nice. slow but nice.
Disney has also used it for the movie Bolt if I’m not mistaken.
bullet takes a lot of shortcuts to achieve a solver that scales linearly. this results in a much less realistic simulation.
Aren’t movie physics unrealistic from nature ?
