Extreme HPC power on Arctur-1 supercomputer

20. 12. 2011 / Industry

Certain large-scale simulations require a lot of RAM memory available to analyze very complex and highly inter-related data. Such research tasks cannot be accelerated through the use of multiple CPU cores since they are not parallelizable.

In these cases, a very fast infrastructure such as Arctur-1 supercomputer is not sufficient. It offers 84 IBM iDataPlex dx360 M3 servers as compute nodes, with two Intel Xeon X5650 processors each (6 cores per processor at 2.66 GHz) and 32 GB RAM. Nodes are interconnected with InfiniBand QDR 40 Gbps fast connection. This means 1008 CPU cores available altogether, with 2.66 TB RAM and a capacity of 10 TFlops, but "only" 2.66 GB of RAM available per core.

Reconfiguring our Arctur-1 supercomputer nodes enables us to fulfill the needs of researchers and achieve outstanding computational capabilities. By installing 128 GB of RAM on a single node, 10.66 GB of RAM is made available per CPU core. Or in the best scenario all 128 GB of memory can be made available for a single core. This means that you can have 128 GB of RAM / core available, which is (according to published data in the TOP 500 list) much more than the most powerful supercomputers in the world can offer.

By offering custom supercomputing capacity Arctur helps the research community fulfill their computational needs and achieve great results.

 

Featured by:

HPCwire: Arctur-I Supercomputer Reconfigured for Large-Memory Simulations