Intel Monday said an electrical glitch in one of its server chips could end up crashing a system and wiping out its data.

The Santa Clara, Calif.-based chip making giant said researchers in its laboratories found the flaw in its 900 MHz and 1 GHz Itanium 2 processors during a series of stress tests last week. The company said it has already begun contacting partners and vendors about the problem.

The chips code-named McKinley is the current high-end model of Intel's flagship Itanium series. The chip ship in servers from well-known manufacturers including IBM , Dell Computer and Hewlett-Packard , which helped develop the processor's EPIC architecture.

According to Intel spokesperson Barbara Grimes, the flaw is only found under a certain set of instructions and conditions but could force the server to "behave unpredictably and crash." Grimes could not specifically comment on what a programmer would have to do to cause the error. She said during the testing phase, Intel researchers ruled out software and hardware bugs as the source of the problem.

"It's so instruction and data dependent that the flaw can vary from system to system and from OEM to OEM," Grimes told internetnews.com.

Intel says the flaw is not present in McKinley's successor, the 1.5 gigahertz version of the Itanium 2 (code-named Madison), which is due to be released in mid-2003. The flaw is also absent in the company's upcomming lower-voltage "Deerfield" series. Intel says Madison will be availble in 3,6 and 9MB configurations due in the next two years. Its dual core Montecito is scheduled for release in 2005.

Intel says it will replace the chip for affected customers or they can opt to use software to lower the clock speed on the Itanium 2 chips to 800 MHz.

According to Grimes, there is a third option by vendors and that is to just simply leave the flaw in. The suggestion here is that not all configurations would be impacted by the glitch.

"Some vendors won't see it happening in their systems and for them and their customer base that is fine," Grimes said. "In our operations the sequence of events may not be a risk. In other cases you might have no crashes. If a customer wants to run the safe route we will replace the chip."

Intel said it would be posting the specifics of the flaw including replacement information and workaround support on its errata site as soon as it becomes available.

Editor's note: A previous version of this story mis-identified the future releases of the Itanium 2 family.