### Better RPMs

Since Nvidia still hasn’t provided RPMs for Fedora 22 (which was launched 4 weeks ago as-of this post date, having been in Alpha for 3 months prior), we need to use the Repo created by the most excellent ‘negativo’ (which Nvidia would probably characterise as some amateur hobbiest, rather than the actual hero that he is) :

Unfortunately, the directory structure for the cuda files is not what Theano expects, so, as root do ::

which rigs up a more standard tree of folders, which is what Theano expects when executing your models :: /usr/local/cuda/{include,bin}.

Now, as root, fix up Nvidia disallowing gcc greater than v4.9

In file /usr/local/cuda/include/host_config.h, look to make the following replacement :

Now, to set the right path for nvcc, in the user’s ~/.bash_profile add ::

#### Test the installation

Check that the kernel modules are there :

Looking good:

Hmpf…

### Installation of libgpuarray

Install the bleeding edge libgpuarray into your virtualenv - first compile the .so and .a libraries, and put them in a sensible place :

This will likely complain about not finding clBLAS, which isn’t a problem here. Although, if you know you will require clBLAS in the future (and this is for advanced/experimental users only), see my OpenCL post, since you need to install this before running cmake above).

Next, install the Python component (after going into the same virtualenv) :

And then test it from within a regular user directory (using the same virtualenv) :

A good result is something along the lines of :

### Theano stuff

Store the following to a file gpu_check.py :

And then run, successively :

and

but