IB Sees Adoption Growth on TOP500 Y/Y
Connects 16 of 33 petascale-capable systems.
This is a Press Release edited by StorageNewsletter.com on June 25, 2013 at 3:03 pmThe InfiniBand Trade Association (IBTA), an organization dedicated to maintaining and furthering the IB specification, announced that the newly released TOP500 shows IB as the interconnect of choice for the world’s most powerful supercomputers, connecting 16 of the petascale-capable systems on the list, while Fourteen Data Rate (FDR) 56Gb/s IB was the fastest growing interconnect technology on the list.
"The increase in IB-connected Petascale systems demonstrates that IB provides the highest performance for clustered computing," said Jim Ryan, co-chair of the IBTA marketing working group. "Furthermore, the significant adoption rate of FDR IB technology among the TOP500 shows that institutions are relying on the newest generation of IB technology to power their mission critical systems. IB continues to help drive the TOP500 performance curve."
IB is a non-proprietary interconnect to deliver petascale performance, and the number of IB-connected petascale systems grew at a faster rate than the overall growth of these systems. FDR, the latest generation of IB technology, grew 3.3X year-over-year from 20 systems in June, 2012 to 67 in June 2013, and connects the four fastest IB-connected supercomputers.
Clustered computing, connecting off-the-shelf servers to form one system capable of parallel processing, continues to dominate the TOP500 with an 83% share of systems. Clusters listed on the TOP500 with the highest computing efficiency utilize IB interconnects. IB is also the preferred interconnect for accelerator-based systems, connecting 80% of those systems.
Delivering bandwidth of up to 56Gb/s with application latencies of one microsecond, IB technology is to carry multiple traffic types (clustering, communications, storage, management) over a single connection. As a mature and field-proven technology, IB is used in thousands of data centers, high-performance compute clusters and embedded environments that scale from two nodes up to clusters with thousands of nodes.