OpenMM 6.0

[Peter Eastman]

We are very pleased to announce the release of OpenMM 6.0. The main new feature in this version is a new Platform optimized for running on conventional CPUs. In many cases, such as on laptops or other computers with low end GPUs, this is faster than any other platform. Of course, it still can't approach the speed of a high end GPU, so you'll want to continue using CUDA or OpenCL when you have a good GPU, but we hope this will greatly expand the range of hardware OpenMM can be used on.

There are a few requirements to be aware of. First, the new platform requires a CPU that supports SSE 4.1. That includes nearly all x86 CPUs made in the last several years, but older computers will not be able to run it. On those computers, the CPU platform will simply not be available.

If you plan to use PME, we also strongly recommend installing FFTW (single precision, multithreaded). The CPU platform can still work without it, but PME performance will be much slower.

Aside from the new Platform, there are several minor new features. The more notable ones include:

* AmberPrmtopFile now supports the GBn2 implicit solvent model.
* When using any implicit solvent model with AmberPrmtopFile, there's now an option to model the screening effects of a non-zero salt concentration.
* RPMDIntegrator has an option to omit the thermostat, allowing it to be used for PIMD simulations.
* StateDataReporter has gained several new options to output additional information.
* There is now a reader for Desmond DMS files.

One important note about running on Windows. Starting with this release, we are building the precompiled binaries with Visual Studio 10 instead of Visual Studio 9. Different versions of Visual Studio are not binary compatible with each other, so the OpenMM binaries must be compiled with the same version that your Python executable was compiled with. We chose VS10 because that's what Python 3.3 is compiled with (in contrast to 2.6/2.7, which are compiled with VS9). So unless you compile OpenMM or Python yourself, or get a version of Python from somewhere other than python.org, you need to use Python 3.3 on Windows.

Peter

[Lee-Ping Wang]

Great! Props for all of the good code and hard work!

[Charles Brooks]

Wipee, for whatever reason CUDA platform is now recognized on Mac Osx 10.9, which is wasn't in 6.0 beta. This makes me happy because I can continue developing on my laptop!

Peter, sorry I didn't see your message under the Mavericks post and hence didn't try it for you. As noted above, things now work. Thank you.

[Maxim Imakaev]

Hi Peter!

Is OpenMM 6.0 compatible with CUDA 6.0 and/or optimized for it?

As far as I remember, OpenMM version nubmers were usually following CUDA version numbers; but this one lists cuda 5.5 as a requirement.

[Peter Eastman]

Any similarity between version numbers for OpenMM and CUDA is purely coincidental. OpenMM 6.0 was tested with CUDA 5.5, and that's what the precompiled binaries were compiled against. If you recompile against CUDA 6.0 there's a very good chance everything will work correctly. But that isn't the version we did all our testing with.

Peter