I’ve noticed that for the last few openSUSE (13.1 & 13.2 at least) releases the nVidia CUDA core implementation is entirely missing from the repos along with the reason I want to use it: Blender Cycles with CUDA rendering. Can someone help me understand the dependencies and the feasibility of installing CUDA capabilities? I have a GTX560 Ti which has a good 300+ CUDA cores going to waste otherwise… :(.
P.S. Is there a good way I can squawk about the cycles render not being in the repo build of Blender? This is really starting to anger me.
However, Packman’s blender package does support CUDA AFAIK, no idea if that’s the same as (or includes) Cycles though.
Cycles on other builds I’ve used has the ability to use the GPU to compute ray bouncing etc. which really speeds up rendering, as far as I know this is done by using a CUDA implimentation. I just downloaded the tar.bz2 straight bin version from Blender’s official website and it seems to contain GPU compute abilities with no CUDA install required.
Is there a way to officially override the repo version with the tar.bz2 version I got from blender.org? I’d really rather have just one copy of Blender on my machine that is nicely integrated in the system menus.
You don’t need to “override” anything. Just uninstall the blender package and use the downloaded one. You might have to create a menu entry yourself, in KDE you can do that by right-clicking on the menu icon and choose “Edit menu entries…” or similar.
i am using " the hard way" Well i have been using the .run for 10 years so…
install the .run for my card
install the .run 6.5 cuda sdk and NOT replace my driver
install the cuda “examples”
build said examples in nvidia’s rebuild of Eclipse “nsight”
have fun !!!
Then grab the blender 2.73a tarball
install it
set the user preferences to USE your card and cuda
( this is NOT!!! 100% automatic sometimes)
no problems with CUDA in blender – mind you you really will NOT see a difference if you are not using the nodes and BGE
Correct me if I’m wrong.
If you use the tarball, downloadable from blender site
There is no need to install the nvidia cuda.
You might need to ask in the blender forum or search there to
confirm this.
As I already wrote, CUDA works fine with the nvidia driver RPMs too. There is an extra CUDA repo for openSUSE provided by nvidia.
And at least Packman’s blender package does support CUDA, so no need for the tarball either. (I’m not sure whether the standard package included in openSUSE does support CUDA or not).
But the questions here were actually about the “Cycles” engine, which is something different AIUI.
is it possible to install CUDA and Nvidia graphics driver on my OpenSuse 13.2 x64 without turning off the x server?
When I enter init 3 or rcxdm stop the only thing I see after that is a black screen…
If you install the packages from the repo, you shouldn’t need to switch to text mode.
I think the .run installer has an option to switch off that check too and allow installation inside an X session, but I’m not sure.
Try running “NVIDIA-xxx.run --help”.
Maybe it helps if you boot to “recovery mode” though? Or add “nomodeset” to the boot options?