Issue running OpenMM in the tutorial about seekr2 HOT 5 CLOSED

Hi Lane,
thank you for your detailed reply. I was installing on a supercomputer and yes we have a version of CUDA installed. Since we didn't have the newest version of CUDA I tried to install openmm from source in the last days, but was still struggling with the install.

In the end, I figured out I can install openmm in conda linked to any CUDA-verison by specifying the requested CUDA-version:

conda install -c conda-forge openmm cudatoolkit==10.0
#for CUDA10.0, according to the manual any version of CUDA>=9.2 is supported

Also, a good way to install ccmake without admin-privileges is using conda:
conda install cmake

Perhaps, those tips might be useful for others. Anyway, thank you very much for your support!

Best
Jens

from seekr2.

Ah, this is probably because OpenMM didn't find CUDA when it was installed.

Inside the python terminal, can you type:

from simtk import openmm
openmm.Platform_getPlatformByName("CUDA")

and let me know the output. If this command throws an error, you should make sure that CUDA_HOME is defined in a linux terminal:

echo $CUDA_HOME

from seekr2.

Yes you are right CUDA_HOME is not set and is not found in python.
Do I need to install CUDA and set CUDA_HOME before installing openmm by conda?
Or do I have to set the $CUDA_PATH to the toolkit that ships with the openmm conda-package?

from seekr2.

Yes, you will need to install CUDA. What version of Linux are you using? Is it on a university network, a supercomputer, or a personal machine? Do you have sudo privileges?

If you have a system administrator who can install CUDA for you, that would probably be the best option. For instance, I would expect most supercomputers and clusters to have CUDA easily available to all users.

However, if you are stuck doing it yourself on your own machine, you can follow the instructions from NVIDIA: https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html. Some Linux distributions have automatic installers that can install a version of CUDA (for instance, Ubuntu has apt-get), but I've had bad luck with these, and better luck installing CUDA from source using NVIDIA's guide. Fair warning: tread carefully with this option - CUDA is very difficult to install yourself.

from seekr2.

And once you install CUDA, OpenMM should automatically detect it. Although if not, you'll have to reinstall OpenMM. You will also need to reinstall the Seekr2 OpenMM Plugin by deleting the build/ directory inside of seekr2plugin/ and start over from the "mkdir build" step.

from seekr2.