Nvidia Driver and CUDA version? about waifu2x

[nagadomi]

You need CUDA 6.5 or later.
The problem seems to be with the installation of Nvidia Driver.

[shiftySenpai]

I have made progress and fixed CUDA w/ clean 6.5 install, but still errors. I copied the cuDNN lib files like this:
cd cudnn-6.5-linux-x64-v2
sudo cp lib* /usr/local/cuda/lib64/
sudo cp cudnn.h /usr/local/cuda/include/

This time it is cuDNN errors:
/usr/local.share/lua/5.1/cudnn/ffi.lua:376 libcudnn.so : Cannot open shared object file: not found in library path

I checked my .bashrc and I have this at the end of it
export PATH=$PATH:/usr/local/cuda/bin
export LD_LIBRARY_PATH=:/usr/local/cuda/lib64

so close!! Atomousukoshi!

[nagadomi]

Did you try source $HOME/.bashrc ?

[JingleiWang]

I faced the same problem when run cuda samples.
How can I fix it ?

[successtheman]

I cannot get it to work either. Nagadomi is there any way you could put this in the AUR for arch linux users?

[JingleiWang]

CUDA6.5 support 340 or pre version driver.
I re-install cuda6.5 and driver 340 manually.
Then I run the exampl with super identity :sudo ./deviceQuery.It worked.
If without sudo, you may get 30 error.

[successtheman]

@JingleiWang, where did you find the files for CUDA 6.5?

Thank you,

On Sun, Jun 21, 2015 at 8:58 PM, JingleiWang [email protected]
wrote:

CUDA6.5 support 340 or pre version driver.
I re-install cuda6.5 and driver 340 manually.
Then I run the exampl with super identity :sudo ./deviceQuery.It worked.
If without sudo, you may get 30 error.

—
Reply to this email directly or view it on GitHub
#22 (comment).

[JingleiWang]

http://developer.download.nvidia.com/compute/cuda/6_5/rel/installers/cuda_6.5.14_linux_64.run

[ctrlcctrlv]

@rjmxtechnology

I created an easy Arch Linux PKGBUILD. I appreciate your comments if you get the NVIDIA GPU support working.

https://aur.archlinux.org/packages/waifu2x-git/

Note that this is for tanakamura's build, not nagadomi's.

Tanakamura's build supports AMD GPU or even CPU only version, at the cost of speed. However, it still completes in under ten seconds for a 1024x1024 image for me with cheap Intel Atom CPU. Also note that the Tanakamura build is written in C++ and to my knowledge has no web interface, if you want to run a web interface you would need to make one.

For the best chance of getting the NVIDIA support working, install cuda and opencl-nvidia before the build.

[successtheman]

I never got it to work @ctrlcctrlv