simtk.org

I am wondering if future releases of OpenSim will be coded to take advantage of GPU processing?

Thanks,

Eric

Hi, Eric. Computational performance is always at the top of our things-to-do list so we're always looking for ways to speed things up. The numerical systems we solve tend to be dense and highly coupled, so do not appear to be well-suited to acceleration with GPUs. Our group here does make extensive use of GPUs for molecular dynamics, where the presence of thousands of particles makes the computation ideal.

At the moment it appears that we can make significant performance gains via (a) algorithmic changes, (b) threading, and (c) use of the relatively-narrow SIMD instructions of the CPU. We don't know how to map our computations onto the wide SIMD required for effective use of GPUs. Please let us know if you have any ideas!

Regards,
Sherm

P.S. Of course we do use the GPUs to accelerate our graphics ...