Cluster Computing:

INFINICON PROVIDES INFINIBAND SWITCHING ARCH FOR AMD CENTER

InfiniCon Systems, the premier provider of shared I/O and switching solutions for next-generation server networks, announced that it will provide a 128-node InfiniBand-based network fabric to support testing and scaling of high- performance computing (HPC) clusters at the AMD Developer Center in Sunnyvale, California. InfiniCon's dense switching technology, superior software scalability, and strong success record in the HPC segment contributed to its selection.

InfiniCon's InfinIO‰ Switch family will be used as the base for the high- speed HPC fabric connecting the compute nodes. The collaboration between the companies combines AMD's state-of-the-art 64-bit processing with InfiniCon's high-speed networking technology ­ capable of up to 30Gbps ­ affording users the

dual opportunity to prototype both advanced processing via the AMD Opteron‰ processor and advanced networking via InfiniBand“ for their application sets across large, production-class fabrics. These new technologies are expected to be well received by the growing HPC market. According to industry analyst firm International Data Corporation (IDC), the HPC market was expected to reach $4.7B in 2003 alone.

"The cost of profiling the effect of new technologies on existing applications can be prohibitive for end users, and carries an element of risk," noted Vernon Turner, Group Vice President of Global Enterprise Server Solutions at IDC. "By giving users the ability to see, touch, and rigorously exercise environments that employ next-generation technologies, AMD and InfiniCon help mitigate that risk and make more informed investment decisions possible."

InfiniBand and High Performance Computing The high bandwidth and low-latency attributes of the InfiniBand Architecture are exceptionally suited to match the demands of high performance computing ­ making it possible to design highly efficient, low-cost clusters from standards based, commodity components. Texas A&M University recently installed a 130-node HPC cluster that deploys InfiniCon's InfiniBand-based switch technology and servers based on AMD Opteron processors.

The industry-leading capabilities of InfiniCon's InfinIO Switch family will e nable AMD to offer a benchmarking environment in which 10Gbps, full bisectional bandwidth is available to every port, with latency characteristics reaching as low as 5 microseconds for message passing applications, even as fabrics scale to hundreds and thousands of nodes. These numbers represent marked enhancements over the performance yielded by current proprietary and Ethernet interconnects for HPC applications.

"The high performance computing industry is one of many segments experiencing the benefits of the AMD Opteron processor," said Ed Gasiorowski, director, developer relations, AMD's Microprocessor Business Unit. "Incorporating an InfiniBand-based technology platform like InfinIO into our developer center enables AMD to truly shine the light on the performance and scalability advantages of Opteron, while giving users the chance to set expectations for real world tasks in advance of funding their own programs."

About InfiniCon Systems

Leveraging the breakthrough advances of the InfiniBand architecture, InfiniCon Systems offers a complete systems solution for enabling data centers to migrate to a fabric/grid model of computing. Our award-winning InfinIO product family provides the dynamic infrastructure required for designing and managing highly scalable, highly available clustered server environments that integrate seamlessly into today's Fibre Channel and Ethernet networks.