Hi all, I'm still pretty new to Project Chrono and am really impressed by 
its capabilities! I was looking through both the SCM and DEM methods for 
deformable terrain simulation, and the GPU-based parallelization of DEM 
simulations seem incredibly useful. However, when running SCM based 
parallelization, I noticed that it still uses multi-CPU thread 
parallelization instead of GPU-based parallelization for the same 
operations. While this allows a single agent to run in close to real time, 
it makes data collection over multiple runs very slow; makes running faster 
than real time difficult, while more threads could conceivably speed up the 
process; and running multiple agents and simulations in parallel difficult 
without the use of university-level supercomputers. I'm particularly 
interested in learning-based control strategies, which would require *many* 
runs for data collection across multiple environments; parallelizing the 
SCM process would potentially allow for much faster simulation.

Are there any plans ongoing to integrate GPU support for SCM simulation? If 
not, are there any significant barriers to implementing it (beyond the 
[significant] hurdle of re-writing everything to interface with GPUs for 
tensor support)? I'm wondering if this hasn't been done because it's hard 
and nobody has had a particular use case for it, or if nobody's done it 
because the underlying mathematics behind SCM make it even more challenging 
in ways not immediately apparent.

Thanks!

-- 
You received this message because you are subscribed to the Google Groups 
"ProjectChrono" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion visit 
https://groups.google.com/d/msgid/projectchrono/372003ae-f733-4ed5-a1ae-023632a14b13n%40googlegroups.com.

Reply via email to