Intel positions the High Performance Reference Stack (HPCRS) as a simple option to use a range of preinstalled software with a focus on high performance computing and AI. The included compilers, program libraries and frameworks are optimized for the use of Intel hardware. The software stack appears in the initial release with version number 2.0.

Instead of the in-house Clear Linux, Intel CentOS 8 as a base image. The AI ​​and deep learning frameworks PyTorch and Horovod are included as additional components. For hardware acceleration of tensor calculations, PyTorch can fall back on the oneAPI Math Kernel Library (Intel MKL), which is also contained in the image. In addition, with the oneCCL, oneTBB and MPI libraries, other Intel oneAPI components are also on board. Spack is used as a multi-version package manager. The SYCL (OpenCL) based programming language Intel Data Parallel C ++ (DPC ++) is provided for programming parallelized workloads.

Schematic representation of the High Performance Computing Reference Stack.

(Image: Intel)

Try Intel-HPCRS The ready-made Container image can, provided Docker engine, with the command docker pull sysstacks / hpcrs-centos from the Docker hub registry. Interested parties can alternatively clone the code from Intel’s GitHub repository, make adjustments and then build the image themselves.

In addition to the installed Docker engine, Intel names a Linux host system that supports the AVX 512 – Can use the instruction set as a requirement.

(ndi)