ZIH-Info No. 184 November 2024

Table of contents

      1. CAPELLA among the Top500 HPC systems
      2. Operational readiness at the turn of the year
      3. Certificate issue via Sectigo discontinued
      4. Matrix 2.0 – Update for the Matrix service
      5. Energy efficiency award for LZR waste heat recovery
      6. OpenGPT-X publishes large AI language model
      7. Junior doctor offers
      8. Events (online)

Editor: Jacqueline Papperitz

CAPELLA among the Top500 HPC systems

The HPC system Capella currently installed at the ZIH achieved very good rankings in the November edition of the Top500 list of the fastest high-performance computing systems: With more than 38 petaflop/s peak computing power (number of double-precision floating point calculations the cluster can perform per second), Capella is positioned in 51st place in the global ranking and 3rd place among German systems. In addition, the system sets standards in energy efficiency, achieving 5th place on the Green500 list of the most energy-efficient supercomputers worldwide. Capella thus outperforms all other NVIDIA H100-based computers. The high-performance computing system, installed by the Saxon company Megware and designed in close cooperation with Lenovo and other companies, is equipped with more than 140 nodes. Each of these nodes has four H100 accelerators from NVIDIA and two AMD processors with 32 cores each. In addition, each graphics processor (GPU) is equipped with 94 gigabytes of high bandwidth memory. The memory system acts as a so-called "burst buffer", providing data at over 1,500 gigabytes/s for the AI accelerators, which benefits data-intensive applications such as the training of large AI models and data analytics requirements. By integrating Capella into the existing LZR infrastructure, the file systems of the Barnard HPC cluster could also be efficiently connected in the overall complex. With its hot water cooling system and waste heat recovery, Capella continues the high standards of energy efficiency at the ZIH data center. The cluster is funded in equal parts by the BMBF and the Free State of Saxony (NHR-Verbund) as well as by ScaDS.AI Dresden/Leipzig and the German Center for Astrophysics. Further information: https://tu-dresden.de/zih/news/tud-supercomputer-capella-in-top500-und-green500. (Contact: Dr. Matthias Lieber)

Operational readiness at the turn of the year

During the closing times for Christmas and New Year from December 24, 2024 to January 1, 2025, the data communication network and the communication, data and computer services of the ZIH will remain in operation. In the event of disruptions, we will only respond if there is an immediate need for action, e.g. for email distribution and to limit damage in the event of accidents. Other reports are processed after the end of the closing time. Service outages can be reported by e-mail with a detailed description of the problem to or by telephone on 0351-463-40000 (answering machine). Current fault reports are made available on the ZIH website under "Operational status" or by calling the fault announcement service on 463-31888. The Service Desk in the Andreas-Pfitzmann-Bau will close on Monday, December 23, 2024. On Friday, December 20, 2024, the Service Desk will be available until 19:00. (Contact: )

Certificate issue via Sectigo discontinued

TU Dresden provides its user and server certificates via the central service provider Sectigo of DFN/TCS Géant. Sectigo has terminated the contract with TCS Géant as of January 10, 2025. Existing certificates remain valid. However, the affiliated universities will only be able to issue new certificates once TCS Géant's negotiations with alternative providers have been successfully concluded. In order to make the transition period as smooth as possible, all certificate holders of certificates expiring in the first half of the year have already been asked to have a new certificate issued by December 15, 2024. (Contact: )

Matrix 2.0 – Update for the Matrix service

With the update of the Matrix service at the beginning of December, the start, registration and synchronization of Element will be significantly faster. Matrix 2.0 now supports modern login methods, enabling the use of Shibboleth including second factor authentication (2FA). A QR code from a current session can be used to set up new devices, including encryption, by simply scanning it. An updated technical implementation also reduces the most frequent warnings regarding encryption. Encrypted group calls are now possible via the integrated Call element and replace the Jitsi instance that was previously required internally. A prerequisite for the new features on mobile devices is the use of the "Element X" app, which has been developed from scratch and also features an improved user interface. Further information: https://faq.tickets.tu-dresden.de/v/ItemID=1250. (Contact: )

Energy efficiency award for LZR waste heat recovery

In addition to security and high availability, energy efficiency is an important goal for the operation of our data center (LZR). The ZIH and the two project partners SachsenEnergie and the Staatsbetrieb Sächsisches Immobilien- und Baumanagement (SIB) are therefore proud that the joint planning for the waste heat recovery of the LZR systems in the Dresden district heating network was awarded the Energy Efficiency Award by the German Energy Agency (dena). By reusing the waste heat from the water-cooled high-performance computing systems, LZR's overall operations have reached a new level in terms of sustainability. In future, three heat pumps will convert this waste heat into district heating. We expect over 50% of the heat generated at the LZR to be reused throughout the year. The building for the heat pumps and the civil engineering work for the media route were managed by the SIB Dresden II. Sachsen-Energie is equipping the building technically. HPC waste heat utilization can generate up to 24,000 megawatt hours of green heat per year, i.e. 3,700 average households can be supplied. The reduction in CO2 emissions due to the avoided recooling amounts to approx. 65 tons per year. Further information is available at
https://tu-dresden.de/zih/news/energiewendepreis-fuer-hpc-abwaermenutzung-im-dresdner-fernwaermenetz. (Contact: Dr. Daniel Hackenberg)

OpenGPT-X publishes large AI language model

OpenGPT-X is now making its large AI language model available for download at Hugging Face. Following the launch of the European LLM Leaderboard in mid-July 2024, the consortium of the BMWK-funded research project has now published the underlying model "Teuken-7B". It was trained from scratch with the 24 official languages of the EU and comprises seven billion parameters. As a technological basis, the free model can be adapted, supplemented and specialized for generative artificial intelligence (AI) applications. It can also be used to implement a wide range of AI applications. The highlights of Teuken-7B include the multilingual pre-processing stage ("Tokenizer"), which ensures greater efficiency in training and operation, as well as embedding in the infrastructure of the European Gaia-X ecosystem. Further information: https://tu-dresden.de/zih/news/opengpt-x-veroeffentlicht-ki-sprachmodell. (Contact: Klaudia-Doris Thellmann)

Junior doctor offers

For the 17th time, students from 3rd to 12th grade can gain insights into research, politics and culture in Dresden in the JUNIORDOKTOR program. Participants will once again visit workshops, laboratories, lecture halls and exhibitions at TU Dresden during the 2024/25 school year to get to know Dresden's scientific landscape and gain inspiration for their own study and career choices. ScaDS.AI Dresden/Leipzig is offering three dates in the field of AI research: "How Computers See Images" on December 3, 2024 and "Point Clouds in the Data Sky - How Machines Learn Independently" on December 16, 2024 and March 13, 2025. There are also two dates in the current exhibition //DataSpaces in the Andreas-Pfitzmann-Bau on December 18, 2024 and January 22, 2025. The free program ends in June 2025. Further information and registration: https://www.juniordoktor.de

Events (online)

  • 04.11.2024, 10:00–11:30 a.m.: FAIR publication of data from clinical, epidemiological and public health studies with the German Central Health Study Hub (English)

  • 03.12.2024, 15:00–16:00: Junior Doctor: How computers see images

  • 05.12.2024, 13:00–15:00: Research data management in general and in HPC

  • 05.12.2024, 14:50–16:20: Research data management in working groups - solutions for collaborative and sustainable work with research data

  • 16.12.2024, 15:30–16:30: Junior Doctor: Point clouds in the data sky - How machines learn independently

  • Onyx basic course: 16.01. and 30.01. 9:20 -10:50 a.m. each day

  • Onyx setup: 12.12.2024, 9:20–10:50 a.m. each day

  • Opal basic course: 09.01. and 23.01. 9:20–10:50 a.m. each day

  • Opal advanced course: 05.12.2024, 9:20–10:50 am

  • Softphone consultation hour: 09.12., 16.12., 06.01., 13.01. 20.01., 27.01. and 03.02. 10:00–11:00 each day

  • Open Q&A session for users of the NHR@TUD computing cluster: 09.12. and 23.12.2024 13:30–14:30 each day

  • 28.01.2025, 18:00–19:30: AI, Language Models, and Cancer Research: How do they interlink?

http://www.tu-dresden.de/zih/veranstaltungen

About this page

ZIH
Last modified: Dec 04, 2024