The computing history at the “Rechenzentrum Garching (RZG)” started with the procurement of an IBM 7090 system in 1961, one of the most powerful compute systems available at that time, with start of system operation in 1962. The installation of an IBM 360/91 system followed in 1969.

In October 1979 the first vector system world-wide for basic research was installed, a Cray 1 system. This marked the evolvement of the vector technology, which dominated supercomputing for nearly two decades with follow-up systems Cray XMP-24 in 1986 and Cray YMP-4 in 1991.

After exploring the potential of massively parallel processing (MPP) systems with the first German nCUBE2/64 system in 1991, the transition from vector to MPP technology occurred in 1995 with a Cray T3D/128 system, followed by a Cray T3E/128 system in October 1996 which was upgraded to 816 PEs in February 1998. MPP-based supercomputer technology was continued with an IBM Power4 based system with initial shipments in Oct 2001, and upgrades to 3.8 TFlop/s in 2002 and to 5.2 TFlop/s in 2003. The Power4 nodes were connected by an HPS "Federation Switch" as a fast communication network.

RZG also hosted a 86-node IBM Power5-based p575 cluster. In September 2007, the worldwide first IBM Blue Gene/P system (2 racks) was installed and extended to four racks in 2008, with a peak performance of 56 TFlop/s. In spring 2008, an IBM Power6-based p575 system (named vip) was installed with 125 TFlop/s peak.

An IBM iDataPlex system with Intel Sandy Bridge processors (named hydra) was installed in 2012 and upgraded with Ivy Bridge processors and 676 NVIDIA Kepler K20x GPUs and 24 Intel Xeon Phi (Knights Corner) accelerators in 2013:  4300 nodes, 84560 cores with an aggregated peak performance of 2.8 PetaFlop/s. Hydra was decommissioned in 2018.

In 2016 an HPC extension cluster (named draco) was installed, based on Intel Xeon Haswell and Broadwell CPUs and Nvidia GPUs (GTX980): 30.688 CPU cores, 128 TB RAM, 1.12 PetaFlop/s peak (FP64), 212 GPUs. Draco was decommissioned in 2022.

In 2018,  the next supercomputer (named cobra) was installed based on Intel SkyLake processors with OmniPath interconnect. It was extended with NVIDIA Tesla V100 and Quadro RTX 5000 GPUs and has a peak performance of over 11 PetaFlop/s. The second phase of the procurement added the HPC system (named raven) which started as an interim system based on the Intel CascadeLake-AP processor in 2020, and was replaced by the final system based on Intel IceLake CPUs and Nvidia A100 GPUs in 2021. Together, cobra and raven provide an aggregate HPL performance of about 20 PFlop/s (FP64) which is the equivalent of a Top15 supercomputer in the Top500 list of June 2021. Cobra was decommissioned in 2024.

Go to Editor View