Računalniški vid in umetna inteligenca na slovenski superračunalniški infrastrukturi

Perš, Janez (author)

Prispevek podaja pregled možnosti za izvajanje računsko zahtevnih nalog računalniškega vida in umetne inteligence v Sloveniji na infrastrukturi SLING. Osredotočimo se na dve ključni okolji: petaflopni superračunalnik Vega (240 GPU Nvidia A100 in 960 vozlišč CPU) ter Arnesovo računsko gručo s starejšimi (V100) in novejšimi (H100) GPU za interaktivno in izobraževalno rabo. Opišemo postopke pridobivanja dostopa (testni dostop, razvojni dostop, redni/veliki raziskovalni dostop) in značilne kvote, ki vplivajo na razpoložljivost GPU virov. Pri tem povzamemo tudi vlogo Nacionalnega kompetenčnega centra (EuroCC 2) pri izobraževanjih in podpori uporabnikom. Posebno pozornost namenimo programski združljivosti za računalniški vid (knjižnice PyTorch/TensorFlow, knjižnice CUDA, kontejnerji) ter razlikam med klasičnim paketnim režimom (Slurm) in potrebami sodobnih potekov dela. V zaključku stanje povežemo s projektom Slovenske tovarne umetne inteligence (SLAIF), ki načrtuje nov, za umetno inteligenco optimiziran superračunalnik ter hibridno HPCoblačno okolje z orkestracijo Kubernetes.

• Language: Slovenian
• Keywords: računalniški vid, umetna inteligenca, superračunalništvo, SLING, Vega, Arnes, SLAIF, Slurm, Kubernetes
• Typology: 1.07 - Published Professional Conference Contribution (invited lecture)
• Organization: FE - Faculty of Electrical Engineering
• Publication status: Published
• Publication version: Version of Record
• Year: 2026
• Number of pages: Str. 5-15
• PID: 20.500.12556/RUL-180965
• UDC: 004.93
• COBISS.SI-ID: 272476931

Copyright

Podatek o licenci se nahaja na pristajalni strani.

• Publication date in RUL: 20.03.2026

Record is a part of a monograph

• Title: ROSUS 2026 : računalniška obdelava slik in njena uporaba v Sloveniji 2026
• Editors: Božidar Potočnik
• Place of publishing: Maribor
• Publisher: Univerza v Mariboru, Univerzitetna založba, Fakulteta za elektrotehniko, računalništvo in informatiko
• Year: 2026
• ISBN: 978-961-299-111-1
• COBISS.SI-ID: 270650883

Licences

• License: CC BY-SA 4.0, Creative Commons Attribution-ShareAlike 4.0 International
• Link: http://creativecommons.org/licenses/by-sa/4.0/
• Description: This Creative Commons license is very similar to the regular Attribution license, but requires the release of all derivative works under this same license.

Secondary language

• Language: English
• Title: Computer vision and AI on Slovenian supercomputing infrastructure

Abstract

This paper reviews how computer vision and AI workloads can be executed in Slovenia on the national SLING infrastructure. We focus on two key environments: the EuroHPC petascale system Vega (240 Nvidia A100 GPUs and 960 CPU nodes) and the Arnes HPC cluster, which includes V100 and some H100-based GPU nodes and serves both education and research. We summarise access paths (Test Access, Development Access, Regular/Large Research Access) and the allocation quotas that affect GPU availability for deep learning training. We also note the role of the Slovenian National Competence Centre (EuroCC 2) in training and user support. We discuss software compatibility for CV (PyTorch/TensorFlow, CUDA libraries, containers) and highlight the gap between batch scheduling (Slurm) and modern AI workflows that require interactive development and longrunning inference services. Finally, we outline the perspective of the Slovenian AI Factory (SLAIF), which plans an AI-optimised supercomputer and a hybrid HPC-cloud environment with Kubernetes-based orchestration.

• Keywords: computer vision, artificial intelligence, supercomputing, SLING, Vega, Arnes, SLAIF, Slurm, Kubernetes

Projects

• Funder: EC - European Commission
• Project number: 101254461
• Name: Slovenian AI Factory
• Acronym: SLAIF