The eX3 project is a research project to evaluate future exascale technologies which will run over five years, with a probability of extensions and expansions. The project´s aim is to explore emerging technologies and software systems in an exascale setting. In addition to the host institution, Simula Research Laboratory, the project consortium consists of the operational organisation for national supercomputing resources Uninett Sigma2, research groups at NTNU, the University of Tromsø and the University of Bergen, as well as the supercomputing suppliers Dolphin Interconnect Solutions, Numascale, Skala Technologies and Fabriscale Technologies. Simula will build and offer those mentioned above educational and research institutions and industrial partners the latest technologies that are essential for exascale and which we will request for future supercomputer procurements for the national computational centers.

The initial phase of the project starts with the procurement of a very heterogeneous Linux based supercomputing cluster, with the latest general processing elements, with dedicated processing elements for machine learning and artificial intelligence, as well as the latest Infiniband high-performance computing (HPC) network and globally shared file systems. The requested elements are:

— 2 redundant administrative nodes,

— Login node with Grid GPGPU,

— the latest HDR 200Gbps Infiniband based HPC network topology,

— 10Gbps network for the offered nodes for administration, imaging/provisioning, failover for global file systems, etc.,

— 1PB net enterprise level storage system, with an option for expansions,

— metadata servers for the global file systems,

— a scale-out global shared file system for artificial intelligence/machine intelligence machine learning,

— lisences for modern 'state-of-the-art' cluster file systems,

— up to 16 nodes with general processors with different instruction set (ISAs),

— a GPGPU OpenCompute(OCP) based HGX-2 system for machine learning/artificial intelligence/deep neural network,

— an OCP OpenPower GPGPU system for machine learning/artificial intelligence,

— building blocks for scale-out global shared file system,

— a modern Cluster Manager for supercomputing clusters (30 nodes, expandable),

— various software development toolkits for the offered architectures.