High Performance Computing with Graphical Processing Units

golem

In time for the 100 year anniversary of the Hamburger Sternwarte, the GOLEM GPU-Cluster enables its theory group to benefit from the use of Graphical Processing Units (GPUs) for high performance astrophysical computing. 

The cluster has been delivered and assembled by SYSGEN in Feb 2012.

The GOLEM GPU Cluster

  • Head node (golem.hs.uni-hamburg.de):
    • 2x Intel XEON Quad-Core CPUs
    • 24 GB RAM
  • 32 diskless compute nodes:
    • 2x Intel Xeon Hexa-Core E5645 CPUs, 2.40 GHz
    • 1x NVIDIA Tesla C2075 GPU, 6 GB GDDR5 memory
    • 24 GB RAM
    • Mellanox Infiniband 40 Gb/s QDR network cards
  • A total of approx. 30 TFlop (single precision) or 15 TFlop (double precision), respectively
  • 40Gb/s QDR InfiniBand Switch QLOGIC 12200-BS23
  • 40 TB RAID storage on Supermicro AOC-SAS2LP-H8iR 6Gb/s Controller
  • Operating sytem: CentOS 6.2
  • Kernel: 2.6.32-220.4.2.el6.x86_64

Sponsors

Logo Universität Hamburg

Logo Behörde für Wissenschaft und Forschung