The Extreme-scale Scientific Software Stack (E4S) undertaking has launched model 22.02 of its assortment of software program packages for creating, deploying and working scientific functions on high-performance compute (HPC) platforms.
E4S is supported by the US Department of Energy (DoE) by way of the Exascale Computing Project, and is successfully a curated ecosystem of numerical libraries, runtime programs, and instruments which can be meant to make it simpler for the HPC and AI/ML developer communities to deploy them.
With model 22.02, E4S now contains 100 full-release merchandise, in contrast with simply 24 full and 24 partial-release merchandise for the primary model of the gathering delivered in October 2018.
The record consists of instruments comparable to Catalyst, ParaView, OpenMP, Kokkos, and the Flang Fortran compiler, plus improvement instruments comparable to HPCToolkit, TAU and PAPI, and mathematical libraries together with PETSc and Trilinos.
According to the E4S launch notes [PDF], v22.02 features a subset of the complete portfolio of Exascale Computing Project (ECP) Software Technologies (ST) merchandise, however demonstrates a goal strategy for future supply of your entire ECP ST software program stack.
E4S makes use of the Spack multi-platform package deal supervisor as a meta-build device to determine and monitor the software program dependencies of all of the packages, and merchandise have been apparently focused for inclusion primarily based on Spack package deal maturity in addition to location throughout the scientific software program stack, so not all ECP ST packages have been included on this launch.
According to E4S undertaking chief Mike Heroux, the undertaking finally ends up constructing tons of of dependencies so that there’s a full software program stack out there to the tip customers. Heroux can be a senior scientist at Sandia National Laboratories.
“One type of minimalist method to consider E4S is that it is a Spack construct script, which in reality, it’s. Spack is a device that lets you outline the way to construct your product, after which it lets you determine the opposite merchandise that your product relies upon upon,” Heroux defined within the E4S announcement on the ECP web site.
“You can consider E4S as being the ultimate supply automobile of hardened and really strong capabilities which can be offered by ECP as reusable libraries and instruments.”
In future, E4S will more and more embody AI instruments and libraries which can be wanted particularly to use to scientific downside units, in line with the undertaking crew.
The E4S assortment is accessible on GitHub and might be downloaded as a container picture in Docker, Singularity, or CharlieCloud format, or by way of a Spack manifest to put in from supply.
It can be out there as an AWS EC2 picture. It helps Arm, x86, and PowerPC processors, plus Nvidia, Intel, and AMD GPUs, and therefore consists of GPU runtimes together with CUDA, NVHPC, ROCm, and oneAPI in addition to AI/ML packages comparable to TensorFlow and PyTorch.
While E4S is open supply software program printed underneath the MIT License, the packages throughout the assortment every have their very own open-source license. ®
https://www.theregister.com/2022/04/01/e4s_22_02/
