High-Performance-Data Analysis
Since summer 2020, the Institute of Data Science has been operating a High-Performance-Data Analysis (HPDA) Cluster to support scientific computing needs at the Friedrich Schiller University Computer Center.
The Cluster consists of:
- 116 CPU-nodes with Intel® Xeon® Platinum CPUs
- 2 Fat-nodes with 1,5 TB of RAM each
- 4 GPU-nodes with a total of 32 NVIDIA Tesla V100 GPUs
- 6 GPU-nodes with a total of 48 NVIDIA Tesla A100 GPUs.
The overall system has 52TB of RAM and a total of 6144 physical CPU-cores. Data is transferred between the computing nodes via an Intel Omnipath high-speed network with 100 Gigabits per second (GBits/s). The usable storage capacity for input and output data is estimated at 1.15 Petabytes in the current expansion stage. In total, the cluster achieves a maximum computing power of 1619 TFLOP/s or 1,62 PFLOP/s.
The site is currently connected to the German Research Netwock (DFN) via fibre optics with a 1 GBits/s connection. This makes it available not only to the DLR institue in Jena, but explicitly to other instituts with DLR, as well as other Research institutions and SMEs.
The HPDA supports research at DLR, particularly in the are of Machine Learning and Causal Inference. It is itself a research object and thus serves as a platform for the development of scalable basic software in the field of Quantum Computing, among other topics.