Fujitsu announced that it has completed the enhancement of the High-performance Computational Server System of the Research Institute for Information Technology at Kyushu University by adding 256 GPU Computing Cards. The theoretical peak performance of the system was boosted from 510 teraflops to 812 teraflops.
Since July 2012, Kyushu University has been operating the Supercomputer System that consists of FUJITSU Supercomputer PRIMEHPC FX10 nodes, and the High-performance Computational Server System that employs a cluster configuration of FUJITSU Server PRIMERGY CX400 x86 servers. Although these new systems provide more computing resources than the previous system, Kyushu University decided to add GPU-based accelerators to its PRIMERGY CX400 cluster to meet the expected increase in computing demand. The high performance and excellent power efficiency of GPUs make them ideal given the limited power capacity and floor space of the facility.
With 256 NVIDIA Tesla K20X and K20 GPUs, the High-performance Computational Server System increased its theoretical peak performance from 510 teraflops to 812 teraflops. As a result, the system was ranked 43rd on the latest Top500 List announced on June 2013, with 621 teraflops of LINPACK performance and 76.48% of computing efficiency.
This is a summary of the press release jointly issued by Kyushu University, Fujitsu and NVIDIA on August 22, 2013. Read the original here (in Japanese).
Date: 22 August, 2012