Skip to main content
Date
2020-08-18
NVIDIA DGX A100

As part of the IIP-Ecosphere project (funded by the BMWi), the L3S research center will soon put the NVIDIA DGX-A100 high-performance server into operation to develop innovative AI solutions. The L3S will thus double its AI computing power to 10 Petaflops - an impressive 10 quadrillion floating-point operations per second. In detail, the high-performance DGX-A10 system with 8x Nvidia A100 Tensor Core GPUs and 320 GB GPU HBM2 memory offers an impressive performance of 5 PFlops AI and 10 PFlops INT8. With the Mellanox ConnectX 6-VPI, up to 200Gb/s HDR InfiniBand speed is realized. At the same time, 15 TB of U.2 NVME drives are available for data storage.

The research center is one of the first users of the system in Europe. It not only enables the fastest possible processing in the field of deep learning but also consumes only a quarter of the power of its predecessor system.

The use of the DGX-A100 intends to promote AI research at L3S - especially in the IIP-Ecosphere project - and to provide an attractive working environment for researchers at L3S.  This is crucial to remain competitive in the international field of AI research and to prepare students and PhD students optimally for their future field of activity in the context of AI.

Within the framework of IIP-Ecosphere, the system will also be integrated into a larger experimental field for intelligent production, which will enable the partners in the project and other smaller companies in the production environment to test AI approaches on a low-threshold basis. This is an essential element of the wide range of project activities aimed at facilitating access to AI and accelerating AI projects in companies. With the support of researchers at L3S, smaller companies can gain insights into AI technologies without having to invest directly in infrastructure.