Setting up openCL

I am running Scientific Linux on a Scientific Linux 6.2 system, Intel P4 3GHz 2CPU 12000 bogomips. My graphics card is an nVidia G92 9800GTX.

See https://developer.nvidia.com/opencl

Setting up the nVidia driver

• 0) back up /etc/r

• 1) download the NVIDIA driver installer from http://www.nvidia.com/object/linux-display-ia32-310.44-driver.html

• 2) Turn off the nouveau driver.
  • in /etc/grub.conf, add rdblacklist=nouveau to the end of each "kernel" line
• 3) reboot to single user, telinit 3
• 4) run the NVIDIA driver installer. Among other things, this will add /etc/modprobe.d/nvidia-installer-disable-nouveau.conf

This adds a whole bunch of files and modules, including /etc/OpenCL with an nVidia vendor file.

Note - I did this before setting up the SDK. The SDK may also have the driver!

Setting up the SDK

• also see http://docs.nvidia.com/cuda/cuda-getting-started-guide-for-linux/index.html

• download CUDA from https://developer.nvidia.com/cuda-downloads

• Yikes! there is no entry for RHEL6 32 bit! Let's try RHEL5 32 bit.
  • Click that button, download the 570MB cuda_5.0.35_linux_32_rhel5.x-1.run
  • run that as root
• modify the PATH to include /usr/local/cuda-5.0/bin . I put this in /etc/profiles.d/cuda.sh:
  • PATH=/usr/local/cuda-5.0/bin:${PATH}
  • export PATH
• I put this in /etc/ld.so.conf.d/cuda.conf:
  • /usr/local/cuda-5.0/lib
• then ran ldconfig as root
• samples:
  • yum install freeglut
  • ln -s /usr/lib/libglut.so.3 /usr/lib/libglut.so
  • sh cuda*run -silent -driver -samples
  • cd /usr/local/cuda-5.0/samples . . . . make

Testing

• /usr/bin/nvidia-settings . . . nvcc -V . . .