DocumentationNews
Welcome to the MetaCentrum grid computing service documentation.
06/10/2025
Presentations and video from the High-Performance Computing Seminar available
On Thursday, October 2, 2025, the MetaCentrum 2025 High-Performance Computing Seminar took place at the Lávka Club in Prague, with more than 90 participants attending in person and another 40 joining online.
The program focused on data processing and storage, security, working with containers and the cloud, as well as the use of AI models on the MetaCentrum and CERIT-SC infrastructure. Experiences were shared not only by experts from CESNET and CERIT-SC, but also by users from research groups at Masaryk University and Charles University.
The seminar presentations are available on the event page. The same link hosts a video recording of the seminar.
06/10/2025
New Stable AI Models
The AI models DeepSeek R1 0528, Qwen3‑Coder‑480B, and GPT‑OSS‑120B are publicly available through the https://chat.ai.e-infra.cz service.
19/08/2025
New GPU Cluster in MetaCentrum
We are pleased to announce that a new computing cluster fobos.meta.zcu.cz has been successfully integrated into the MetaCentrum infrastructure. Cluster Specification
- Number of nodes: 20
- Total CPU cores: 1920
- Configuration of each node:
- CPU: 2x AMD EPYC 9454 2.75GHz 48-core 290W Processor
- RAM: 768 GiB
- GPU: 4x NVIDIA L40S 48GB
- Disk: 4x 3.84 TB NVMe
- Network: Ethernet 100Gbit/s, InfiniBand 200Gbit/s
- Power (SPECrate 2017_fp_base): 1160
- Owner: CESNET
Access to Computing Resources: The cluster is available in regular queues. A complete list of available computing servers can be found here: https://metavo.metacentrum.cz/pbsmon2/hardware
08/08/2025
BeeGFS: Fast Shared Scratch
We’re pleased to announce the availability of a new fast shared scratch using the parallel distributed file system BeeGFS on our bee.cerit-sc.cz cluster. This new resource, available as ‘scratch_shared’, is specifically designed for high-performance computing (HPC) needs and offers several advantages for data-intensive and compute-intensive applications.
BeeGFS is ideal for demanding jobs that require:
- Working with large files or a huge number of small files – efficiently handle massive datasets, making it an ideal choice for applications that require fast and scalable storage.
- Utilizing many threads or processes that read or write in parallel – enables high-performance and concurrent access to data, making it perfect for applications that require simultaneous reads and writes.
- Spanning multiple compute nodes – can handle workloads that span multiple compute nodes, allowing for seamless scalability and performance.
- Sequential computations with intermediate results – well-suited for workflows where subsequent computations can pick up intermediate results left in the scratch directory, eliminating the need to copy data to permanent storage or run on the same machine as the previous step.
Typical Use Cases:
- High-Performance Computing (HPC) – BeeGFS is designed to efficiently handle large files and parallel input/output operations, making it an ideal choice for scientific computing workloads.
- Machine Learning and AI – With BeeGFS, you can train machine learning models faster by accessing large volumes of data with high-throughput and low-latency.
- Simulations, Rendering, Genomics, and Big Data Research – BeeGFS is perfect for handling massive datasets, such as those found in 3D rendering, complex simulations, genomic sequencing, and big data research.
More Information:
Last updated on
