Performance Analysis
Mindaugas Macernis, Vaidotas Mickus, Janne Ahonen, Laurynas Diska, Jonas Franukevicius, Juozas Sulskus "Still Unsolved High-Performance Computing Challenges for up to Pre-Petascale Homogeneous Supercomputers" arXiv preprint arXiv:2210.00934, 2022
Distributed Memory Cluster
Total maximum utilization of 143 compute nodes per task (one node has 64 compute cores with AVX-512, AVX2 AVX, SSE4.2 processor instructions, 376 GB RAM). Expected physical availability of compute nodes: 80%. Nodes are connected to other nodes in the cluster and clusters via a 100 Gbps IB network. The entire system uses high-performance file systems: 46TB (in the 100 Gbps IB network) and 289 TB (in the 40 Gbps IB network). Maximum file size for a task is 200TB. Task queues: SLURM Operating system: Red Hat Enterprise Linux release 8.4 Architecture: Sequana X1000. Usage options: OpenMP and MPI type calculations by selecting the number of nodes, cores and memory required for the task, in the Linux RHEL8 environment. Virtual machine with Windows environment according to special needs.
GPU computing nodes
One node (2 pcs) contains 64 computing cores with AVX-512, AVX2 AVX, SSE4.2 processor instructions; 4 Tesla V100 GPU cards, 376 GB RAM. Nodes are connected by an IB network. Expected physical availability of computing nodes: 50%. Nodes are connected by a 100 Gbps IB network with other nodes in the cluster and clusters. The entire system uses high-performance file systems: 46TB (in a 100 Gbps IB network) and 289 TB (in a 40 Gbps IB network). Maximum file size for a task is 200TB. Task queues: SLURM Operating system: Red Hat Enterprise Linux release 8.4 Architecture: Sequana X1000. Usage options: OpenMP and MPI type calculations with NVidia GPU technology by selecting the number of nodes, cores and memory required for the task, in the Linux RHEL8 environment. Supercomputer "VU HPC Saulėtekis" cluster BullSequana X1000
Installation. News published 2020-06-25: https://naujienos.vu.lt/vu-atsivers-iki-siol-dar-nematytos-galimybes-pasirasytos-sutartys-del-naujo-superkompiuterio-irengimo/
Publication: https://www.vu.lt/projektai/lietuvos-grid-nasiu-skaiciavimu-tinklas-litgrid-hpc/
Open access high-performance computing pools (until 2020)
from supercomputer from "Fizika 100" to "Fizika 2000"
“HPC Saulėtekis” is intended for the work of scientists from VU, the Lithuanian National Center for Physical Technology Sciences, and other Lithuanian research institutions.
The goal of the center is to encourage students and scientists to use high-performance computing in scientific work, to participate in joint European research projects using even more efficient infrastructure, e.g. the PRACE project.
HPC Saulėtekis was created to create conditions for solving relatively resource-intensive problems in materials science, quantum chemistry, astrophysics, and other areas. The advantage of this infrastructure is that almost all of the supercomputer's resources can be used for a single problem, the solution of which takes several weeks or more.
The supercomputer's resources can be used by all users, not only Lithuanians, on a competitive basis in accordance with the rules of the VU open access centers. Currently, the largest possible task is 25.7 TFlop with 15 TB RAM.
The High Performance Open Access Computing Center "HPC Saulėtekis" is part of the infrastructure of the National Center for Physical and Technological Sciences. The supercomputer equipment was purchased with European Union funds in 2006 and 2011.