Program

All times are CEST.

TimeContentPDF/Video
09:00 – 09:10IntroductionPDF
09:10 – 10:00Keynote presentation:
Deploying and Managing the LUMI Supercomputer, Sustainably
Pekka Manninen, LUMI Leadership Computing Facility, Finland
PDF
Video 1/2
Video 2/2
10:00 – 10:30Paper presentation:
Rule-based Thermal Anomaly Detection for Tier-0 HPC Systems
Mohsen S. Ardebili, Andrea Bartolini, Andrea Acquaviva and Luca Benini
PDF
Video
10:30 – 11:00Invited talk:
A Conceptual Framework for HPC Operational Data Analytics
Michael Ott, Leibniz Supercomputing Centre, Germany
PDF
Video
11:00 – 11:30Coffee Break
11:30 – 12:00Paper presentation:
Wholistic and Physics-Based Data Center Monitoring
Hilary Egan, Avi Purkayastha and David Sickinger
Video
12:00 – 12:30Invited talk:
Opportunities & Challenges with Quantitative Codesign
Terry Jones, Oak Ridge National Laboratory, USA
Video
12:30 – 12:55Panel / Participant Discussion:
Recent Developments in MODA
Video 1/2
Video 2/2
12:55 – 13:00Closing

Keynote presentation:
Deploying and Managing the LUMI Supercomputer, Sustainably
Pekka Manninen, LUMI Leadership Computing Facility, Finland

EuroHPC Joint Undertaking’s supercomputer LUMI, operated by CSC Finland, is a GPU-accelerated HPE Cray EX system with 2,560 AMD MI250X GPU nodes with 375 Pflop/s sustained performance, supplemented with a number of other capacities. LUMI has been deployed throughout 2021 and 2022, and going into full service in this autumn. In this talk, we will review the LUMI system and LUMI datacenter together with related monitoring  building automation systems. We are collaborating with the system vendor in development of the monitoring capabilities, which will be addressed in this talk. What makes LUMI unique is its 100% carbon-neutral operations, which can be also counted as net negative due to the waste heat utilization in district heating. Hence the third theme of the talk is sharing some thoughts – and hopefully raising discussions – on the minimization of the environmental footprint of HPC installations and their operations, and how to evaluate it.