With the new akmod-nvidia-331.20-3.fc19 released today comes an option to build the nvidia kernel modules several times. This is useful as advertised in the nvidia FAQ to "Minimize the software overhead when driving many GPUs in a single system" . That, specially when each GPU is processing different workload such as with CUDA applications.

You can use it by setting the following RPM macro, with up to 8 different kernel modules:
#File /etc/rpm/macros.nvidia - Set the number of nvidia kernel modules to build (up to 8)
%_nv_build_module_instances 2

Then you can use akmod-nvidia to automatically rebuild the kernel modules with the above option:
$ sudo yum install akmod-nvidia kernel-devel (or kernel-PAE-devel when using kernel-PAE)

If you previously had the package installed, and want to rebuild with the new option, you can use:
$ sudo akmods --force akmod nvidia

This will build the specified number of kernel modules along with the nvidia-frontend.ko
$ rpm -qpl ~builder/rpmbuild/RPMS/x86_64/kmod-nvidia-3.11.8-200.fc19.x86_64-331.20-3.fc19.x86_64.rpm

Then, you can modprobe the appropriate kernel modules, as mentioned in the FAQ, this command will assign two cards to the nvidia0 kernel module instance:
$ sudo modprobe nvidia0 NVreg_AssignGpus="0:01:00.0,0:02:00.0"

The Official documentation for the RPM Fusion packaged nvidia driver is still at nVidia How on the RPM Fusion Wiki.