Hi!
I made a quick test of the OpenCL SPH solver to check the new optimizations I implemented and this is the result, as always with a Nvidia GTX670:
- 8 million particles
- 450 frames
- 1h 50 minutes
So, well I have to say I'm quite happy now with the speed, and I don't think I can optimize it much more.
Next step: Foam! :O)
Cheers!
5 comments:
impressive speedup. have you compared it with a cpu implementation? guess your using cuda instead of opencl?
Hi Anonymous,
no, it's not CUDA, it's still OpenCL.
I didn't try with a CPU implementation( I wrote one long time ago and it was very slow), but I'm pretty sure it would be much slower. OpenCL is supposed to work with the CPU too, so I'll try that just to see how it works.
would be interesting to see how it compares. currently working on a sph sim myself, but it's only cpu based right now and yea, even the 1M particle example takes hours. if i take it right you're using a time step of around 0.042? so i guess your not enforcing incompressibility?
The timestep is adaptative right now, but usually is around 0.002, it's just the GPU is really powerful to compute this kind of simulations
Hi Tony
this is really cool implement. and the speed is very good.
would you make a public beta of this for Maya community?
Post a Comment