Intel Tries to Give it Away - AMD Says "No Way"

PC Week] - Intel Corp. has disclosed a problem involving its 820 and 840 chip sets that has spurred the company to scrap plans for three motherboards it had on its server road map. But while an Intel representative stressed that the problem would likely affect only a few users, some analysts said the trouble may indicate a flaw in a crucial component used to configure the boards with SDRAM.

The trouble arose when Intel's 820 and 840 chip sets, which were designed to take advantage of speedier Rambus memory, RDRAM, were configured to work with SDRAM.

According to Intel, memory errors may result from the combined use of the memory repeater hub (known as MRH-S), which is used to translate native RDRAM support to SDRAM, and ECC (error correction coding). Intel said it believes few customers will be affected, since most 820 and 840 chip-set customers using ECC, commonly in workstation and servers, are likely to have gone with Rambus memory for greater performance.

However, customers seeking a lower-cost server or workstation solution may have elected to go with less-expensive SDRAM over RDRAM in order to get the highest amount of memory for their money.

"We are informing vendors that certain server platforms using the memory repeater hub as well as the error correction coding have been experiencing memory issues or errors," said Dan Francisco, an Intel spokesman.

If the problem is actually tied to just the memory translation hub, it could spell big trouble for Intel, said Mike Feibus, principal analyst at Mercury Research in Scottsdale, Ariz. "We're predicting that more than half of the 840 chip sets that will ship this quarter will be configured with SDRAM," Feibus said. "So if there's trouble with the memory translator, that could pose major problems for Intel."

A spokesman for Dell Computer Corp. said the company has avoided packaging SDRAM with 820 and 840 chip sets after it became aware of the problem a couple of weeks ago. A representative at Compaq Computer Corp., which sells workstations with SDRAM, said the company has been made aware of the problem and is working with Intel to resolve it. Officials at Hewlett-Packard, which also sells systems packaging SDRAM with the 820 and 840 chip sets, were not immediately available for comment.

Industry analyst Kevin Knox said that, while he was unaware of the problem, he's not surprised. "Anytime you put a translator in there (i.e., the memory translator hub), you're asking for trouble," he said.

"The 820 and 840 chip sets were not designed for SDRAM," Knox said. "So trying to make them work with SDRAM posed an obvious risk of introducing errors."

The problems involving the memory translator and error correction coding came to light after Intel informed vendors that it was scrapping plans for three chip sets targeted for use in servers.

The chip sets, known by the code names Pine, Hemlock, and Willow, were developed based on the 820 and 840 chip-set designs. Intel had planned to design the boards to accommodate dual processors configured with SDRAM. "We are working to resolve the issue and expect to fix it in the next 'stepping' of the parts," Francisco said. The 820 chip set's introduction was initially delayed last fall due to a platform integration issue involving Rambus. The problem was addressed by altering the motherboard from its original three-slot RIMM configuration to a two-slot RIMM configuration.

Market Impact

This is a short-term black eye for Intel, but in the longer term it will weather the impact from this problem. If Advanced Micro Devices (AMD) had a server offering, Intel would pay a significantly higher price, both in terms of the hit to its reputation and the advantage AMD would gain. Since Intel is the only game in town for the x86-architecture servers, AMD will not gain anything other than a PR bullet for taking potshots at Intel. (More on this below.)

Market growth will be dampened for the next few months, until Intel can fix the problem. It is unlikely that non-x86 servers will be able to capitalize on this issue, since the two architectures generally compete on other measures than whether SDRAM works in the motherboard.

Regarding the Intel-AMD battle: until recently, Intel was perceived as being nearly invulnerable in the CPU and system markets - AMD competed on price and performance, but customers still tended to go with "Intel inside", especially when Intel kept lowering prices in response to AMD.

In the last six months, Intel has started to show signs of weakness; specifically, the 'Saber' eight-way main board issues, the 'Coppermine' problems, and the supply chain problems which led Gateway to go back to AMD. Alone, these issues would be annoying to customers, but taken together they seem to point to Intel having taken their eye off the ball.

Users who see AMD as a potential server alternative to Intel have been waiting for about a year since the "Poseidon" chipset was first discussed. The recent (since July, 1999) silence from AMD makes us wonder whether Poseidon (now called HotRail) will become reality, and thus give Intel some multiprocessor competition.

SOURCE:http://www.technologyevaluation.com/research/articles/intel-tries-to-give-it-away-amd-says-no-way-15294/