The Open Forcefield Toolkit provides implementations of the SMIRNOFF format, parameterization engine, and other tools. Documentation available at http://open-forcefield-toolkit.readthedocs.io
With my laptop, the simulation wrote 319 frames using a stride of 10 steps and 79 frames using a stride of 100. I don't think there's unique value in frames 80 to 319, so this seems like a pure improvement as it moves a large portion of the CPU time away from writing trajectory frames and back to running the actual simulation. Increasing the stride more might cause too few to be written for lesser hardware or what's provided on cloud instances we don't pay for.
I also added a quick check for if CUDA is available, in which case it's bumped up to 1000. I didn't want to go all the way to detecting if a GPU is properly detected by OpenMM, especially since OpenCL is used for AMD GPUs but typically available on machines with weak or no GPUs.
Closes #1724
With my laptop, the simulation wrote 319 frames using a stride of 10 steps and 79 frames using a stride of 100. I don't think there's unique value in frames 80 to 319, so this seems like a pure improvement as it moves a large portion of the CPU time away from writing trajectory frames and back to running the actual simulation. Increasing the stride more might cause too few to be written for lesser hardware or what's provided on cloud instances we don't pay for.
I also added a quick check for if CUDA is available, in which case it's bumped up to 1000. I didn't want to go all the way to detecting if a GPU is properly detected by OpenMM, especially since OpenCL is used for AMD GPUs but typically available on machines with weak or no GPUs.