simtk.org

Hello
I have installed the gromacs from openMM gromacs source code I have also compiled openMM libraries ,I have added path of openMM library through LD_LIBRARAY_PATh and also kept openmm lib file in gromacs/lib/openmm folder even though the mdrun is not using GPUs.
since it not printing openmm:cuda or openmm:refrence on console when i am running.
where i am going wrong

Hi Vinod,

We're just about to release a new version of OpenMM, so rather than spending time figuring out what's happening with the old version, it would probably make more sense to wait for the new version and try it out. If you still have problems then, we can help you debug them. The new release should be posted either later today or tomorrow.

Peter

I do have the same problem: I installed openmm preview 3 and openmm-gromacs preview 3 according to the directions on my macbook pro, but get the message "OpenMM Platform: Reference". Do I have to invoke mdrun-openmm with a special cuda option?

Christian

I do have the same problem: I installed openmm preview 3 and openmm-gromacs preview 3 according to the directions on my macbook pro, but get the message "OpenMM Platform: Reference". Do I have to invoke mdrun-openmm with a special cuda option?

Christian

Do you have CUDA in your library path? Before running Gromacs, you probably want to type something like

export DYLD_LIBRARY_PATH=/usr/local/openmm/lib:/usr/local/cuda/lib

Peter

Also make sure that Gromacs knows where to find the plugin libraries. If you've installed OpenMM somewhere other than /usr/local/openmm, you need to set the OPENMM_PLUGIN_DIR environment variable to point to (openmm)/lib/plugins.

Peter

Hello peter,
I have added the library path and now its showing
OpenMM Platform: Reference this means its running on CPU.So is their any other way of submitting on GPUs.
Secondly now when running on CPU its not generating
any output.What could be reason for this?
I have compiled both openMM and gromacs new version.

Thank you Peter for this tip. The CUDA lin path was missing. Unfortunately, now I get a memory allocation error:
cudaMalloc CUDAStream::Allocate failed initialization error
Maybe, the Geforce 9600M GT in my Macbook Pro is the problem?

Christian

Just a follow up. During a fresh Install of Cuda Toolkit, I found that the kext is not installed by default. Now I installed it, but the new error message says: Bus error
Maybe I try it on a Linux machine.

Christian

Two questions. First, are you certain you are using preview release 3? The error message you quoted looks like it was generated by PR2, not PR3.

Second, how large is the system you are trying to simulate? The stream allocation failure may simply indicate you don't have enough video memory. (And how much video memory do you have?)

Peter