Installing Nvidia CUDA on Ubuntu 14.04 for Linux GPU Computing

Installing Nvidia CUDA on Ubuntu 14.04 for Linux GPU Computing

In this article I am going to discuss how to install the Nvidia CUDA toolkit for carrying out high-performance computing (HPC) with an Nvidia Graphics Processing Unit (GPU). CUDA is the industry standard for working with GPU-HPC. In a previous article Valerio Restocchi showed us how to install Nvidia CUDA on a Mac OS X system. In this article I am going to describe the same procedure but carry it out under the latest version of Ubuntu, namely 14.04.

Installation and Testing

The first task is to make sure that you have the GNU compiler collection (GCC) tools installed. This is carried out by installing the build-essential package:

sudo apt-get install build-essential

I'll assume that you have a 64-bit system for the remainder of the article. The next step is to download the specific DEB package for the 64-bit version of CUDA for Ubuntu 14.04. I placed this in my home Downloads directory:

cd ~/Downloads
wget http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1404/x86_64/cuda-repo-ubuntu1404_6.5-14_amd64.deb

The following commands will install CUDA 6.5:

sudo dpkg -i cuda-repo-ubuntu1404_6.5-14_amd64.deb
sudo apt-get update
sudo apt-get install cuda

We also need to add the following lines to our .bash_profile file in our home directory, in order to obtain the required compilation tools on our PATH:

export PATH=/usr/local/cuda-6.5/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-6.5/lib64:$LD_LIBRARY_PATH

Remember to make sure that the terminal has access to these variables:

source ~/.bash_profile

Before proceeding to test the GPU cards we will ensure that the drivers are correctly installed. The following line will provide us with the driver version:

cat /proc/driver/nvidia/version

The output on my system is as follows

NVRM version: NVIDIA UNIX x86_64 Kernel Module  331.89  Tue Jul  1 13:30:18 PDT 2014
GCC version:  gcc version 4.8.2 (Ubuntu 4.8.2-19ubuntu1) 

Check the version of the Nvidia CUDA compiler:

nvcc -V

The output on my system is as follows

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2014 NVIDIA Corporation
Built on Thu_Jul_17_21:41:27_CDT_2014
Cuda compilation tools, release 6.5, V6.5.12

In order to check that the installation was successful we are going to compile the CUDA samples, test that we can query the GPU device and ascertain its bandwidth. In the following code sample below, change <target_directory> to your preferred installation location for the sample scripts:

cuda-install-samples-6.5.sh <target_directory>

Change directory to the <target_directory>/NVIDIA_CUDA-6.5_Samples and run the make command:

cd <target_directory>/NVIDIA_CUDA-6.5_Samples
make

This will take some time. Once complete we can run the deviceQuery script to test if we can communicate with the GPU:

cd bin/x86_64/linux/release
./deviceQuery

I have two GPU cards in SLI configuration on my system and so I've only shown the output for the first device:

./deviceQuery Starting...

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

Detected 2 CUDA Capable device(s)

Device 0: "GeForce GTX 780 Ti"
  CUDA Driver Version / Runtime Version          6.5 / 6.5
  CUDA Capability Major/Minor version number:    3.5
  Total amount of global memory:                 3072 MBytes (3220897792 bytes)
  (15) Multiprocessors, (192) CUDA Cores/MP:     2880 CUDA Cores
  GPU Clock rate:                                1084 MHz (1.08 GHz)
  Memory Clock rate:                             3500 Mhz
  Memory Bus Width:                              384-bit
  L2 Cache Size:                                 1572864 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:                     Yes
  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 Bus ID / PCI location ID:           1 / 0
  Compute Mode:
     < Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >

..
..

> Peer access from GeForce GTX 780 Ti (GPU0) -> GeForce GTX 780 Ti (GPU1) : Yes
> Peer access from GeForce GTX 780 Ti (GPU1) -> GeForce GTX 780 Ti (GPU0) : Yes

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 6.5, CUDA Runtime Version = 6.5, NumDevs = 2, Device0 = GeForce GTX 780 Ti, Device1 = GeForce GTX 780 Ti
Result = PASS

The final line is the most important. It states that the test was successful as we received a "PASS". We also want to check the bandwidth to our GPU. We can run the bandwidthTest command:

./bandwidthTest

The output on my system is as follows:

[CUDA Bandwidth Test] - Starting...
Running on...

 Device 0: GeForce GTX 780 Ti
 Quick Mode

 Host to Device Bandwidth, 1 Device(s)
 PINNED Memory Transfers
   Transfer Size (Bytes)    Bandwidth(MB/s)
   33554432         6308.7

 Device to Host Bandwidth, 1 Device(s)
 PINNED Memory Transfers
   Transfer Size (Bytes)    Bandwidth(MB/s)
   33554432         6464.2

 Device to Device Bandwidth, 1 Device(s)
 PINNED Memory Transfers
   Transfer Size (Bytes)    Bandwidth(MB/s)
   33554432         264346.8

Result = PASS

As before, the final line is the most important. It states that the test was successful as we received a "PASS".

That concludes the installation and testing of the Nvidia CUDA toolkit! You should now be able to follow Valerio's second tutorial on creating a "Hello World!" for CUDA.

I found the following articles helpful when installing CUDA on my system as I initially had issues with my Nvidia driver: