Installing CUDA-8.0 on Ubuntu 16.04 with Nvidia GeForce 845m

After having installed nvidia driver 367 by typing:

sudo add-apt-repository ppa:graphics-drivers/ppa -y

sudo apt-get update
sudo apt-get install nvidia-367 -y

and disabled Secure Boot in UEFI (BIOS) settings, I'm trying to install properly nvidia-cuda 8.0.

These were my steps (following this guide):

1. from https://developer.nvidia.com/cuda-downloads I downloaded cuda_8.0.44_linux.run



2. sudo mkdir /usr/local/cuda-8.0


3. sudo ./cuda_8.0.44_linux.run --override


4. in /etc/profile.d/cuda.sh: export PATH=$PATH:/usr/local/cuda/bin


5. in /etc/ld.so.conf.d/cuda.conf: /usr/local/cuda/lib64


6. sudo ldconfig


7. Force cuda to work with gcc 5, commenting line 119 out in usr/local/cuda/include/host_config.h: //#error -- unsupported GNU version! gcc versions later than 5 are not supported!


8. rsync -av /usr/local/cuda/samples in my cuda_samples/


9. sudo GLPATH=/usr/lib make

However, when i type:

./nbody -benchmark -numbodies=256000

in samples/bin/x86_64/linux/release, I get this message:

Error: only 0 Devices available, 1 requested. Exiting.

Any help?

I partially solved this way:

1. sudo apt-get --purge remove nvidia-*


2. sudo service lightdm stop


3. download nvidia driver NVIDIA-Linux-x86_64-367.57.run from the website


4. sudo ./NVIDIA-Linux-x86_64-367.57.run --no-opengl-files


5. sudo ./cuda_8.0.44_linux.run (you should not install its own nvidia-driver 367.48, as you have 367.57 already installed)


6. in /usr/local/cuda-8.0/include/host_config.h, comment this line out: #error -- unsupported GNU version! gcc versions later than 5 are not supported!



7. mkdir cuda_samples


8. rsync -av /usr/local/cuda/samples .


9. GLPATH=/usr/lib make


10. reboot

And you get your cuda-8.0 enabled!

CAVEAT: As you got nvidia driver from the website, nvidia-settings doesn't work. I am trying now to figure out how to make nvidia-settings properly work.