The MetaCentrum computing grid, a key resource for research and scientific computing in the Czech Republic, has significantly boosted its capabilities with the integration of new clusters owned by CESNET and the University of West Bohemia (ZČU). This upgrade enhances both CPU and GPU power, while introducing a specialized high-memory SMP node to support demanding computational tasks.
Supporting Scientific Research
MetaCentrum’s expanded infrastructure will accelerate computational projects across various fields, including high-performance computing, AI, big data, and scientific simulations. Users can explore the full list of available servers and their current utilization on the MetaCentrum Hardware Page.
New CPU Cluster: Adan
Owner: CESNET (adan[1-48].grid.cesnet.cz)
Total Capacity: 48 nodes / 6,144 CPU cores
Node Configuration:
- CPU: 2x AMD EPYC 9554 (128 cores per node)
- RAM: 768 GiB
- Storage: 2x 3.84 TB NVMe
- Network: 25 Gbit/s
- Performance: SPECrate 2017_fp_base 1,360 (10.6 per core)
The new Adan cluster replaces previous hardware and is optimized for CPU-intensive computations, offering high processing power without GPU acceleration. It is already available in MetaCentrum’s standard scheduling queues.
Enhanced AI and GPU Cluster: Alfrid
Owner: University of West Bohemia (ZČU) (alfrid[1-9].meta.zcu.cz)
Total Capacity: 8 GPU nodes / 1,024 CPU cores + 1 SMP node
Modernization Phases: May 2024 & December 2024
ZČU’s Alfrid cluster has undergone a major expansion, now featuring:
-
GPU Nodes (alfrid[1-4] & alfrid-II [6-9])
- CPU: 2x AMD EPYC 9554 (128 cores per node)
- RAM: 1,536 GiB (alfrid) / 768 GiB (alfrid-II)
- GPU: 2x NVIDIA L40 48GB (alfrid) / 4x NVIDIA L40S 48GB (alfrid-II)
- Storage: 2x 7 TB NVMe
- Network: 25 Gbit/s
- Performance: SPECrate 2017_fp_base 1,230
These nodes are ideal for AI workloads and GPU-accelerated calculations, offering researchers advanced hardware for machine learning, simulations, and more.
-
SMP High-Memory Node (alfrid5)
- CPU: 2x AMD EPYC 9554 (128 cores total)
- RAM: 4,608 GiB (~4.5 TB shared memory)
- Storage: 2x 7 TB NVMe
- Network: 10 Gbit/s
The large-memory SMP node is designed for memory-intensive applications requiring up to 4.5 TB of RAM, making it a powerful tool for tasks like genomics, data analytics, and complex modeling.
The Alfrid cluster is accessible in priority queues for ZČU and ITI researchers and in standard short queues for other MetaCentrum users.
