How to use the mic and GPU in High-performance computing (HPC)

1. Installation of GPU Dirver

Dirver Name: Nvidia-linux-x86_64-310.40.run

Before installation, you need to change the operating system mode to text mode, and modify the/etc/inittab run level to 3.

Under the appropriate directory, run./nvidia-linux-x86_64-310.40.run, start installation driver

After the installation is complete, run Nvidia-smi–l,nvidia-smi–a and nvidia-smi-l can view the information on the GPU.

2. Install cuda5.5

2.1 Basic requirements

1, before the installation of the operating system needs to be changed to text mode, and modify the/etc/inittab run level to 3

2, can not open the server Remote Desktop VNC function, if it has been enabled, please shut down first

Chkconfig Vncserver off

Service Vncserver Stop

3, for the operating system for redhat6.0 and subsequent systems need to manually install the Freeglut package if you do not install this package, the SDK will not be installed and take an example when the demo will be reported missing libglut.so library files, Redhat 6.0 before the version of the system CD will carry a bag, Manual installation, redhat6.0 after the system CD no longer with this package, need to download after the installation

2.2 Installation process

2.2.1 Installation Freeglut

Download freeglut-2.8.1.tar.gz from the Internet,

TAR–ZXVF freeglut-2.8.1.tar.gz

CD freeglut-2.8.1

AutoUpdate

./autogen.sh

./configure

AutoUpdate

Make

Make install

Ln-s/usr/local/lib/libglut.so.3/usr/lib64/libglut.so (Specific path depending on installation)

Ln-s/usr/local/lib/libglut.so.3/usr/lib64/libglut.so.3 (Specific path depending on installation)

2.2.2 Installation cuda5.5

If the Cuda version is 5.0 prior to the version required to manually install the GPU driver, toolkit and SDK. If the Cuda version is 5.0 or 5.5, the GPU driver, the Toolkit tool and the SDK are already in one installation package to download an installation package and install a package. CUDA5.5 's package name is Cuda_5.5.22_linux_64.run

Upload the installation package to the server

And then execute./cuda_5.0.35_linux_64_rhel6.x-1.run, note that if the operating system is RHEL6.0 and does not install after Freeglut, the component installation fails. Follow the prompts to install it.