The “Doppler”-cluster (named after the physicist Christian Doppler, who lived and worked in Salzburg) is a heterogeneous high performance computing system, consisting of more than 32 compute nodes, equipped with a variety of GPGPU accelerator hardware. Doppler’s maximum performance currently exceeds 20 TF/s (Rpeak: 20.000 bilion double precisions operations per second). By employing high efficiency/energy saving processors along with GPUs this high performance is accomplished at a surprisingly small „energy footprint“ and we thus regard the system a truly „green“ super computer.
Software (June 2015)
- Rocks Cluster Suite 6.1.1 (Centos 6.5), SGE scheduler
- OpenMPI, MPICH2
- Python 2.6, 2.7, 3.4
- AMD® Core Math Library
- AMD® LibM
- Oracle Java JDK
- GCC 4.4.7, 4.7.2 and others
- NVIDIA CUDA 4.2, 5.0, 6.0
- Orca, VASP
- Other software can be made available on request
Hardware (June 2015)
- Cores: 1759 Opteron Series 6200, 2.2 Ghz
- RAM: 8.6 TB
- 44 nodes, 2 – 16 GB RAM/core, 16 – 64 cores per node
- AMD® Opteron™ 6200 server (2-2.4 GHz)
- Intel® server with Xeon Phi™ 5100 Series Coprocessor, 8GB GDDR5 of memory
- NVIDIA® Tesla™ S2050 server (Fermi), 4 x 3GB GDDR5 of memory
- NVIDIA® Tesla™ K20 (Kepler), 6GB GDDR5 of memory
- GPU nodes based on Intel® Core™ i7 and Nvidia® GeForce™ GTX480, 590, 680 and 780 TI
- QDR Infiniband / Mellanox for intra connection
- 21TB RAID 10 storage + backup
- Virtual appliances (Linux, Microsoft® Windows®) can be made available on request
AMD and Opteron are trademarks or registered trademarks of Advanced Micro Devices, Inc.
Intel, Xeon and Intel Xeon Phi are trademarks of Intel Corporation in the U.S. and/or other countries.
Microsoft and Windows are either registered trademarks or trademarks of Microsoft Corporation in the United States and/or other countries.
NVIDIA, GeForce, Tesla and CUDA are trademarks or registered trademarks of NVIDIA Corporation.