February 22, 2000 8:50 AM PST
Bugs found in servers, workstations using Intel chipset
- Related Stories
Intel unveils 750, 800-MHz CoppermineDecember 20, 1999
Bug in new Pentium IIIs hampers boot-upDecember 1, 1999
Pentium IIIs in short supply, Intel confirmsNovember 17, 1999
After many delays, Rambus PCs are releasedNovember 15, 1999
PC makers struggle with Coppermine strategyOctober 25, 1999
The Santa Clara, Calif., chipmaking giant has discovered a bug that affects some server and workstation computers incorporating recently released Intel chipsets, just when it seemed like the company was digging out from a series of manufacturing snafus in 1999. Though the glitch occurs somewhat rarely, three circuit board ("motherboard") designs have been canceled in response.
Intel is moving to correct the error and will work with computer makers to resolve any current product issues, according to spokesman Dan Francisco.
Still, the bug is not likely to sit well with hardware manufacturers, who had to endure shortages and product delays last year. The shortcoming also could prove a boon to chipset start-up ServerWorks, which makes components that compete with the problematic Intel parts.
Some server and workstation makers are experiencing data corruption errors with systems containing Intel's high-end 840 and 820 chipsets and also one of two ancillary chips, the Memory Repeater Hub (MRH) or the Memory Translator Hub (MTH). The latter two chips are actually to blame, Intel said.
The two chipsets were designed to "talk" to an advanced memory design called Rambus, and are typically put into computers that also contain Rambus memory. Rambus, however, has been at the center of controversy because of its higher price, and manufacturers have resisted using the design despite Intel's endorsement. The MRH, which is paired with the 840, and MTH, which comes with the 820, essentially let computer makers use standard computer memory because the chips can take signals from standard memory and translate them to Rambus signals for the chipsets.
Further, the error only occurs when a system also incorporates Error Correction Code (ECC) technology. Server makers generally adopt ECC, which helps prevent data from becoming corrupted as it shuttles between the processor and other components, while workstation vendors occasionally use it.
The number of systems affected by the bug is limited, asserted Intel's Francisco. The majority of customers that have adopted the 820 and 840 chipsets are using Rambus memory, and therefore aren't using the translator chips. In addition, "the majority of the 820 customers do not enable ECC," he said.
Intel said it will correct the defect the next time it revises its manufacturing process, in the next "spin," to use industry parlance.
Nonetheless, some will likely be unhappy. Hewlett-Packard is one high-profile company that makes products within the danger zone. HP and others also had to deal with last year's problems with the 820 and 840 chipsets, which were delayed more than once for different reasons, forcing manufacturers to alter their road maps.
Intel also had difficulties manufacturing adequate volumes of the high-end "Coppermine" Pentium IIIs. The shortage continues to linger, although the company maintains that the end is in sight. "In Q1 we will catch up on everything," said Pat Gelsinger, vice president of Intel's desktop products group.
While painful for Intel, ServerWorks may benefit. The start-up specializes in chipsets for Intel-based servers. The key difference is that the company's chipsets don't speak Rambus. Instead, they are designed to work with standard memory and don't require a translator chip.