Using Intel® SHMEM in Intel® Tiber™ AI Cloud#

Intel® SHMEM is now enabled in Intel® Tiber™ AI Cloud for oneAPI. To get access or sign in, please refer to the Getting Started Guide:. When requesting hardware to launch an instance, please choose the Intel® Max Series GPU from the hardware catalog list. This will ensure the current supported GPUs for Intel® SHMEM are added to the requested instance. Intel® SHMEM has been tested and validated on instances with Ubuntu 22.04 LTS operating system (labeled as ubuntu-2204-jammy-v20230122).

After logging into the allocated instance, the Intel® oneAPI DPC++/C++ Compiler needs to be added to the environment:

source /opt/intel/oneapi/setvars.sh

By default, cmake is not installed in the instance which is required for building Intel® SHMEM. It can be added with the following command:

sudo apt install cmake

We also recommend to install a specific version of GNU autoconf that works well with Sandia [1] OpenSHMEM installation. During testing, we occasionally found issues with the default autoconf version of 2.71 . Users can follow the instructions below to download and install autoconf 2.69 in such a case:

wget https://ftp.gnu.org/gnu/autoconf/autoconf-2.69.tar.gz
tar -zxvf autoconf-2.69.tar.gz
cd autoconf-2.69
./configure
make
sudo make install

With the above, users can directly follow the installation instructions provided in README.md to install Intel® SHMEM with OpenSHMEM back-end.