IBM and Corning will work together on a project for the US Department of Energy and National Nuclear Security Administration to develop optical interconnect technology for supercomputers.
The project aims to deliver a system within 30 months using a $20m (£12m) grant from the government.
Corning and IBM hope to take advantage of the high-speed and low-power consumption of optical technology, which sends light pulses down a fibre-optic cable rather than sending electrical signals across copper wires, to transmit data.
While semiconductor companies continue to improve the performance of the processors within supercomputers and clustered servers, server and networking companies need to ensure that interconnect technology develops at a similar rate to prevent performance bottlenecks between networks.
Optical technology scales well as processors or servers are added to supercomputer clusters, and improves performance over existing interconnect technologies, IBM and Corning said.
Corning has conducted much research into optical semiconductor technologies, while IBM has experience in developing switching technology for data and telecommunications networks.
Corning will develop the optical interconnects, while IBM produces the electronic control and monitoring circuitry.
IBM's supercomputing division will also sell prepackaged clusters of its new eServer 325 system, its first server with Advanced Micro Devices's Opteron processor.
A base configuration of the eServer 1350 supercluster comes with four nodes of the eServer 325, for a total of eight AMD Opteron 246 processors.
The cluster costs $11,676 (£6,971) for the base configuration, which comes with either Microsoft's Windows Server 2003 or Red Hat Linux version 3.
Users who choose Windows for their cluster will be unable to take advantage of the Opteron's 64-bit capability, as Microsoft does not expect to release the final version of Windows Server 2003 for Opteron until the second half of next year.
IBM also sells bundles of its servers which use Intel's Xeon processors. Scientific researchers and universities with limited budgets can assemble large resources of computing power with the clusters, IBM said.
Tom Krazit writes for IDG News Service