Scientists in the Czech Republic get the most advanced AI system and two new clusters for demanding technical calculations

Prague, 1 June 2023. Scientific and research institutes in the Czech Republic, which use the e-INFRA CZ national e-infrastructure for science and research, now have the most advanced system of today for computing in the field of artificial intelligence NVIDIA DGX H100. It has been installed in the CERIT-SC centre at Masaryk University in Brno and expert teams will connect to it via the MetaCentrum operated by CESNET. This collaboration will significantly strengthen the environment for innovative research and development in artificial intelligence (AI) and accelerate progress in areas such as machine learning, big data analytics and AI application development.

MetaCentrum users can additionally start using two brand new computing clusters (connecting a number of computers via a fast network). The first CESNET was launched at the Institute of Molecular Genetics of the Academy of Sciences of the Czech Republic in Prague-Krč under the name TURIN, and the second at the Institute of Computer Science of Masaryk University in Brno under the name TYRA.

The NVIDIA DGX H100 system was purchased at a cost of CZK 11 million, while the cost of the TURIN and TYRA clusters exceeded CZK 20 million. Both investments were co-financed by European funds.

NVIDIA DGX H100 system

The NVIDIA DGX H100 solution deployed at CERIT-SC in Brno is the most powerful NVIDIA DGX system in our country and the first of its kind in Europe. It brings extreme computing power and innovative research capabilities. The architecture of the system enables processing of huge data volumes and significantly accelerates computational tasks. With high-performance memory subsystems in graphics accelerators, it provides fast data access and optimises performance when working with large datasets.

One of the exceptional features of the NVIDIA DGX H100 is its parallel processing capability, which enables rapid development and training of complex AI models.

“Users can achieve unparalleled efficiency and responsiveness in their AI tasks. The solution is equipped with the latest NVIDIA H100 GPUs, which provide up to ten times more performance than the previous generation Ampere A100 processors,” says Professor Luděk Matyska, Director of CERIT-SC.

The system will find applications in many fields, for example, in spoken word analysis, 3D image reconstruction and even in the detection of neurodegenerative diseases.

NVIDIA DGX H100 system configuration

• 8 GPU H100 80GB SXM5
• 135,168 CUDA cores
• 640 GB of GPU memory
• 2 TB of RAM memory
• 84 TB NVMe for OS
• 30 TB NVMe for data

TURIN and TYRA clusters

Scientists have found Elvish names for two new clusters launched by CESNET. The Prague TURIN cluster has 52 nodes, each with 64 CPU cores and 512 GB of RAM. Its Brno colleague TYRA consists of 44 nodes and otherwise has identical technical specification.

“More than 6,000 new CPU cores appropriately complement graphics accelerators and contribute to a wide range of computing capacity to meet the diverse needs of research teams. The direct connection to the national 100Gb network and the location of the clusters closer to the research teams significantly strengthens our capacity to process the data produced in these organisations,” says Miroslav Ruda, Head of Distributed Computing at CESNET.

Both clusters are equipped with AMD processors along with AMD 3D V-Cache technology. These are the most powerful server processors designed for technical computing. They are designed to speed up the product development process and increase productivity. The system enables 200 times higher interconnect density compared to current 2D technologies and more than 15 times higher interconnect density compared to other 3D technologies using so-called solder outputs.

The solution is suitable, for example, for virtualisation, hybrid cloud or software-defined infrastructure.

Configuration of TURIN and TYRA clusters

• 6,144 CPU cores in total
• 96 nodes in total, each with 64x AMD EPYC 7543@2.80GHz, 512 GB RAM, 7TiB NVME scratch filesystem
• 10Gb uplink to CESNET backbone network
• CESNET e-infrastructure integration, allowing immediate access to other services, such as direct connection to CESNET data storage

    

Project E-INFRA CZ: Modernisation (CZ.02.1.01/0.0/0.0/18_072/0015659) is financed by the OP VVV and co-financed by the European Union.

 

The CESNET association was founded by various Czech universities and the Czech Academy of Sciences in 1996. Its objective is the operation and development of ICT services that comprise the national e-infrastructure for research, development and education e-INFRA CZ. Integral components of its activities include research and development in information and communications technology. CESNET is an active partner in numerous international research infrastructures, such as the Pan-European GÉANT research, development, and education network, the European Grid Infrastructure EGI.eu and the European infrastructure for open science EOSC. Read more at www.cesnet.cz.