You can leverage the most advanced GPU on IBM Power machine.  The Power System S822LC for High-Performance Computing (code-named “Minsky”) contains two Power8 CPUs and  two Nvidia Tesla P100 GPUs. It is available at Minicloud 🙂

This post is an updated version of the steps described here.

Verify that the GPU cards are detected

sudo lspci | grep -i nvidia

Install basic development packages (CUDA toolkit dependencies)

$ sudo apt-get install build-essential

Download the DEB package of the CUDA repository for Ubuntu 16.04 on POWER8

wget https://developer.nvidia.com/compute/cuda/8.0/prod/local_installers/cuda-repo-ubuntu1604-8-0-local-ga2_8.0.54-1_ppc64el-deb

Install the DEB package

sudo dpkg -i ./cuda-repo-ubuntu1604-8-0-local-ga2_8.0.54-1_ppc64el-deb
sudo dpkg -f install

Update the package manager’s definitions

sudo apt-get update

Install the CUDA toolkit

sudo apt-get install cuda

Configure the search paths for the CUDA commands and libraries, making the setting available to all users and persistent across reboots

echo 'export PATH=$PATH:/usr/local/cuda-8.0/bin' | sudo tee

echo /usr/local/cuda-8.0/lib64 | sudo tee /etc/ld.so.conf.d/cuda-8-0.conf

sudo ldconfig

Reboot the system

sudo reboot

After rebooting, verify that the nvidia module is loaded and the GPU devices are available

sudo dmesg | grep -i nvidia

sudo lsmod | grep nvidia

nvidia-smi --list-gpus

Perform a simple test with the CUDA sample

mkdir ~/samples

cp -r /usr/local/cuda-8.0/samples/ ~/samples/

cd ~/samples/7_CUDALibraries/simpleCUFFT

make

./simpleCUFFT

References

http://www.redbooks.ibm.com/redpapers/pdfs/redp5169.pdf

https://developer.nvidia.com/cuda-downloads-power8

http://www.tomshardware.com/news/ibm-power8-nvidia-tesla-p100-minsky,32661.html

https://www.ibm.com/ms-en/marketplace/high-performance-computing