SCIENCE
Germany’s Newest, Fastest, Greenest GPU / CPU Supercomputer Built, Managed with Bright Cluster Manager
Leading TOP500 and Green500 Supercomputer at Goethe University Frankfurt combines complex high-performance computing with unique cooling model to set new PUE standard.
Goethe University Frankfurt is using Bright Cluster Manager to manage its new LOEWE-CSC supercomputer. In the latest release of the TOP500 supercomputer rankings, LOEWE-CSC is listed as the 22nd fastest in the world. It is also the fastest ATI GPU based system in the world and the fastest x86-based system in Germany.
LOEWE-CSC also ranked 8th in the Green500, a comparison of energy efficiency based on the same Linpack benchmark performance as the TOP500.
LOEWE-CSC, built by ClusterVision, comprises 20,880 AMD Opteron 6172 processor cores and 772 ATI Radeon HD 5870 GPUs using Supermicro servers and Mellanox Infiniband. This unique architecture is the world's fastest system based on AMD/ATI GPUs, and the world's fourth fastest GPU-accelerated system. The system reaches 60.7% efficiency, far ahead of the other leaders in the TOP500 list.
ClusterVision used Bright Cluster Manager in both the deployment stage and as the management solution for the supercomputer.
"Bright Cluster Manager played a key role in building the LOEWE-CSC system," said Christopher Huggins, Sales and Marketing Director at ClusterVision. "During the deployment phase, it was critical that we had real-time insight into all of the hardware and software metrics. Bright gave us a clear view of the total picture, instantly."
LOEWE-CSC running at less than full capacity achieved 285.2 TeraFLOPS Linpack at 736 MegaFLOPS/Watt and a 1.1 Power Usage Effectiveness ratio (PUE). PUE, a metric created by the GreenGrid organization, is determined by dividing the amount of power entering a data center by the power used to run the computer infrastructure within it. PUE is expressed as a ratio, with overall efficiency improving as the quotient decreases toward 1. According to the Uptime Institute, a typical data center has an average PUE of 2.5. Uptime estimates most facilities could achieve 1.6 PUE using the most efficient equipment and best practices.
"Needless to say, our uniquely complex cluster represents a difficult management challenge," said Prof. Volker Lindenstruth, Chair for HPC Architecture at the Frankfurt Institute for Advanced Studies. "That is why we chose Bright Cluster Manager. We needed a solution we could trust, and Bright's proven success in managing complex systems combining x86 and GPU technologies won us over. We are thrilled with how Bright's comprehensive, yet easy-to-use solution helps us manage and control our very important supercomputer."