CUDA-Instructions

Steps to install CUDA on Ubuntu 16.04 (LTS) and configuring NVIDIA Graphics Card

Steps

(Keep these instructions handy in another machine as later on you'll be typing commands in tty mode. Also, take backup of your system before initiating.)

1. Install build essentials.

sudo apt-get install build-essential

2. Download the CUDA toolkit.

wget https://developer.nvidia.com/compute/cuda/8.0/prod/local_installers/cuda_8.0.44_linux-run

3. Extract installers (path in extract argument must be absolute).

./cuda_8.0.44_linux.run -extract=<absolute_path>

4. Remove all traces of nvidia in ubuntu repositories.

sudo apt-get --purge remove nvidia-*

5. Remove xorg config file if present.

sudo rm /etc/X11/xorg.conf

6. Stop the default nouveau driver.

sudo nano /etc/modprobe.d/blacklist-nouveau.conf

• Add the following lines:

  blacklist nouveau
  options nouveau modeset=0
  

7. Update the initramfs image and reboot.

sudo update-initramfs -u
sudo reboot

8. Switch to tty with CTRL+ALT+F1 and login to your user. You can exit the tty mode with ALT+F7
9. Kill the x-server. (This is necessary !)

sudo service lightdm stop

10. Run the following commands. (Flag "--no-opengl-files" is necessary otherwise you will get stuck in a login loop forever !). If you get this error ERROR: Unable to load the 'nvidia-drm' kernel module., try turning secure boot off in the BIOS settings.

sudo ./NVIDIA-Linux-x86_64-367.48.run --no-opengl-files
sudo ./cuda-linux64-rel-8.0.44-21122537.run
sudo ./cuda-samples-linux-8.0.44-21122537.run

11. Set the environment variables.

export PATH=/usr/local/cuda-8.0/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-8.0/lib64:$LD_LIBRARY_PATH

12. Verify the install.

cat /proc/driver/nvidia/version
nvcc -V

If you get error on 'nvcc -V', run sudo apt-get install nvidia-cuda-toolkit

13. Now you can switch back to normal mode. Run sudo service lightdm start

14. Check your gcc version gcc --version. CUDA doesn't support gcc version > 5. So either you can install a lower version of gcc or do the following(works!):

• sudo nano /usr/local/cuda/include/host_config.h
• Comment out the line 119 (//#error -- unsupported GNU version! gcc versions later than 5 are not supported!)

15. Run the following commands.

cd /usr/local/cuda/samples/1_Utilities/deviceQuery
sudo make
./deviceQuery

You Should get the following output:

./deviceQuery Starting...

 CUDA Device Query (Runtime API) version (CUDART static linking)

  Detected 1 CUDA Capable device(s)

  Device 0: "GeForce 940MX"
    CUDA Driver Version / Runtime Version          8.0 / 8.0
    CUDA Capability Major/Minor version number:    5.0
    Total amount of global memory:                 2002 MBytes (2099642368 bytes)
    ( 3) Multiprocessors, (128) CUDA Cores/MP:     384 CUDA Cores
    GPU Max Clock rate:                            1242 MHz (1.24 GHz)
    Memory Clock rate:                             1001 Mhz
    Memory Bus Width:                              64-bit
    L2 Cache Size:                                 1048576 bytes
    Maximum Texture Dimension Size (x,y,z)         1D=(65536), 2D=(65536, 65536), 3D=(4096, 4096, 4096)
    Maximum Layered 1D Texture Size, (num) layers  1D=(16384), 2048 layers
    Maximum Layered 2D Texture Size, (num) layers  2D=(16384, 16384), 2048 layers
    Total amount of constant memory:               65536 bytes
    Total amount of shared memory per block:       49152 bytes
    Total number of registers available per block: 65536
    Warp size:                                     32
    Maximum number of threads per multiprocessor:  2048
    Maximum number of threads per block:           1024
    Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
    Max dimension size of a grid size    (x,y,z): (2147483647, 65535, 65535)
    Maximum memory pitch:                          2147483647 bytes
    Texture alignment:                             512 bytes
    Concurrent copy and kernel execution:          Yes with 1 copy engine(s)
    Run time limit on kernels:                     No
    Integrated GPU sharing Host Memory:            No
    Support host page-locked memory mapping:       Yes
    Alignment requirement for Surfaces:            Yes
    Device has ECC support:                        Disabled
    Device supports Unified Addressing (UVA):      Yes
    Device PCI Domain ID / Bus ID / location ID:   0 / 1 / 0
    Compute Mode:
       < Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >

  deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 8.0, CUDA Runtime Version = 8.0, NumDevs = 1, Device0 = GeForce 940MX
  Result = PASS

Install CUDNN (from https://developer.nvidia.com/rdp/cudnn-download)

Download cuDNN v5.1 Library for Linux, extract it, and navigate to the location in terminal

sudo cp -P cuda/include/cudnn.h /usr/local/cuda/include
sudo cp -P cuda/lib64/libcudnn* /usr/local/cuda/lib64
sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*