Apply for access to the eX3 infrastructure
Written by Are Magnus Bruaset, Simula Research Laboratory
Photo: Henrique Kugler
The Experimental Infrastructure for Exploration of Exascale Computing (eX3) is a national research infrastructure open to HPC researchers at Norwegian institutions and their international collaborators. It is specifically designed for experimental research on HPC technologies that are expected to be vital for the next generation of supercomputers, cloud services, and data centers.
The eX3 infrastructure is hosted by Simula Research Laboratory with UNINETT Sigma2, NTNU, UiB, UiT, OsloMet, Dolphin Interconnect Solutions, Graphcore, and Numascale as partners. The facility is on the Research Council of Norway’s Roadmap of National Research Infrastructures and is funded through the INFRASTRUKTUR program.
Favouring explorative use
Production-oriented HPC infrastructures concentrate on bringing high volumes of computing power, based on well-proven technologies and with the highest possible level of reliability, to a wide array of different users. The eX3 infrastructure is at the other end of the spectrum, favouring explorative use over continuous stability. It is not a supercomputer per se, but a carefully curated collection of hardware and software components that are regarded as important for coming HPC environments, and which can be combined and configured in different ways.
In particular, eX3 aims specifically at offering researchers working on HPC technologies, methods and algorithms the necessary experimental testbed for exploration of highly heterogeneous computing environments. On the compute side, this heterogeneity ranges from different CPU architectures to accelerators like GPUs and special processors designed for high-volume machine learning and artificial intelligence workloads. The same thinking also concerns interconnects, memory, and storage.
Due to its specific nature, eX3 is continuously evolving by adding new components to the environment. Currently, interested users can access various x86 architectures from AMD and Intel, as well as ARM architectures from Cavium and HiSilicon.
In addition, a fully equipped 16-way DGX-2 system from NVIDIA supports research on (multi-)GPU computing, while a soon-to-be-revealed system will provide access to bleeding-edge technology constructed specifically for machine learning. Researchers interested in interconnect technologies can experiment with 200Gbps Infiniband HDR, PCIe interconnects from Dolphin, as well as high-performance Ethernet networking.
If eX3 sounds like the type of infrastructure that could support your research, feel free to apply for access