Intel Tuesday revealed a new component design that will allow computer makers to build super-slim, energy-saving server blades that slide into specially configured racks. At the same time, IBM is now talking publicly about a new server that it says will eliminate the performance bottlenecks typically associated with large symmetric multiprocessing configurations.
Intel is combining a low-voltage Pentium III processor with its 440GX chipset to create a subsystem that will gird forthcoming server-blade offerings from major computer makers. A blade is essentially an entire server resting on a single board, and it can be combined with other boards in a holding rack to create a more powerful system. Compaq has said it plans to incorporate the new Intel subsystem into its Quick Blade architecture, which it plans to begin selling early next year. Dell Computer and Hewlett-Packard will follow suit.
Intel says the low-voltage Pentium III solves overheating problems normally associated with dense server configurations by running at what it says is a server industry low of 1.1 volts. The processor is from a product family known as Tualatin, which the chipmaker originally sold for use in notebook computers. The 440GX chipset will be able to address up to 2 Gbytes of high performance error-correcting-code memory. Intel officials say the low-cost, low-power attributes of blades make them ideal for front-end applications like Web serving.
However, not all large vendors are jumping on the blade bandwagon--at least not right away. Tom Bradicich, IBM's director of server architectures, says IBM will forego the first generation of Intel-based blade servers in favor of more robust future offerings. "We need to make sure everything is there in terms of power, reliability, and redundant features that people would expect in an IBM server," Bradicich says. He says the company will introduce a blade server labeled Xcaliber next fall. That's roughly when Intel plans to introduce a 900-MHz, dual-processor version of its Tualatin chip.
But IBM is preparing a new server architecture of its own. Servers built on the company's forthcoming Enterprise X-Architecture--formerly code-named Summit--will feature a new IBM chipset that uses copper and silicon-on-insulator technologies to maximize speed and reduce heat emission. Enterprise X-Architecture servers, such as the X-352 Crusader, which sources say IBM will unveil next month, will also have remote I/O capabilities that will allow systems to be easily combined using a new approach to symmetric multiprocessing clustering. Remote I/O will let users connect up to four Crusader servers together via an expansion port to create a 16-way system. IBM officials say that yields better performance than simply ganging 16 processors in a single box. "It gets rid of a lot of the internal bottlenecks," says Bradicich. He adds that IBM hopes to achieve 32-way scaling on X-Architecture in about two years.
Although IBM will formally launch the architecture next month, it's doubtful that the company will be able to sell Crusader systems in significant volume until next year. That's because Crusader will be powered by Intel's Xeon Foster MP chip, which use the same NetBurst architecture found on Pentium 4 chips. Sources familiar with Intel's plans say the company will not begin volume shipments of Foster MP until late March. Regardless, analysts say the new innovations surrounding Intel chips will boost the company's fortunes, and that of its OEMs, in the server market. Says Technology Business Research analyst Brooks Gray, "It certainly improves the value proposition vs. Unix."