IBM unveils high-end systems

As IBM contemplated a 32-processor successor and licensing the first model to other companies, it showed off its new family of Enterprise X Architecture servers at Comdex.

The servers, which use a combination of Intel chips and IBM's own chipset technology, codenamed Summit, will launch first with Intel Xeon processors.

At a later date, it is planned that the servers will adopt Intel's second-generation 64bit chip, known as McKinley.

The Summit chipset allows IBM to incorporate mainframe-like features such as the ability to swap out memory and processors while the computer is running.

Summit servers are ready to go and will debut at the same time Intel releases the McKinley chip, according to Brendan Paget, IBM's worldwide marketing manager for xSeries servers. He also confirmed that the Summit servers took about $20m and three years to develop.

IBM executives said the company is negotiating licensing the design to server makers and looking at the technology for Linux.

The four-way server shown at Comdex also provides expandable input/output (I/O) through the use of an I/O expansion enclosure, which can double the amount of I/O available to the new xSeries server.

It comes with as much as 64MB of high-speed cache memory shared by each four-processor unit.

New capabilities include an added Level 4 cache of up to 64MB to improve processor performance, and physical partitioning that allows each node to run its own combination of operating systems and applications. Hot-swappable and hot-add memory are also included.

IBM executives said users will ultimately be able to use the expansion enclosure outlet and plug the new xSeries server directly into an InfiniBand switched fabric network.

In addition, the architecture is designed to enable customers to "build as they go", as it enables users to integrate up to four individual four-way servers together into a 16-way system by using new high-speed scalability ports.