Can applications to use CUDA technology just installing nVidia driver on Linux?

Hello.
I have an nVidia GT 430, using Kubuntu 12.04 64 bits.
As you know, Blender can render using CUDA with Cycles.
Windows people can use Cycles with CUDA just instaling nVidia driver. So, I want to know if in Linux further the driver also CUDA Toolkit must be installed.
Well, I’ve tried 313 and 319 drivers, and I have no available CUDA on Blender. After several failed attempts trying to install CUDA Toolkit, I have installed version 4.2 and I have finally CUDA in Blender.
From what I’ve searched on Google, there are many people like me who have lot of trouble wanting to install CUDA Toolkit and wanting to run CUDA on Blender in GNU/Linux.
I know that in Ubuntu 13.04 cuda toolkit is in the repositories, but you have to download about 600MB of dependencies for developers that I think are useless (unnecessary) for users who only want to use Blender with CUDA.

So, Could you make CUDA can be used with just installing the driver, like you do with the driver for Windows?
Thank you very much.

You shouldn’t need the whole CUDA toolkit. You just need the CUDA runtime. On Kubuntu, install: the libcudart5.0 package.

Hi.
Do you say that because you’ve tried and it works for you by installing only that package?
You talk about libcudart5.0, so I assume you mean to Kubuntu 13.04. In Kubuntu 13.04 64 bit I could not even install any nVidia driver because this bug:
https://devtalk.nvidia.com/default/topic/542937/linux/nvidia-319-17-bug-on-ubuntu-13-04
So I am using this PPA for Kubuntu 12.04:
https://launchpad.net/~otto-kesselgulasch/+archive/nvidia-cuda
Then, I installed only the libcudart4 package and I have no CUDA available in Blender. Install only this package is not enough for me.

http://wiki.blender.org/index.php/Dev:2.6/Source/Render/Cycles/Building

It’s a packaging issue. I didn’t realize Ubuntu didn’t package the precompiled CUDA kernels.

I remind you that I can have CUDA available on Blender 2.66.1 in Kubuntu 12.04 with 3.2.0-39-generic Ubuntu Kernel, of course installing cuda-tolkit from the PPA.
So let me see if I understand you. You say that if Ubuntu had provided a precompiled CUDA kernel, then I would need to install only “libcudart” package instead of everything related to cuda-toolkit?
If correct what I say, then I will make a request in Launchpad for they provide a precompiled CUDA kernel.

Still in this way, the common Ubuntu user need to know that an extra package need be installed to have the CUDA available on Blender, which adds a complication. So, could nVidia people add the content of “libcudart” into the nVidia driver to solve this complication?

Of course if nVidia people could add all the necessary into the driver for CUDA work in Blender even without a CUDA precompiled kernel and only installing nvidia driver, it would be much better :)

By precompiled CUDA kernels, I mean the CUDA programs that ship with blender. This is unrelated to the Linux kernel.

This is a packaging issue on the Ubuntu side. They should create a package recommends dependency on the cuda runtime.

The software necessary to compile the CUDA kernels is the CUDA toolkit. The Windows package of Blender compiles these and ships them with blender. In principle, Ubuntu could also compile these beforehand and include them with Blender.

@YAFU, You don’t mention if you have an Optimus laptop or if this is a desktop machine. If it’s a laptop, you’ll need to use Bumblebee http://www.bumblebee-project.org in order to get the NVIDIA drivers to install on a separate X screen and use the optirun syntax to run any GPU-related apps.

Sorry for forgetting to give that information, I have a desktop computer.
As I said before, I can run CUDA in Blender after overcoming several problems.
The intention of the thread was to know who could make life easier for the Ubuntu Linux common user to can work with CUDA in Blender: nVidia, Blender or Ubuntu
Something like this:
Install nVidia driver > Install Bldener > Be happy with running CUDA on Blender.