Intel cited several factors for its decision: a flawed strategy of matching Timna with the Rambus DRAMs; and a preference by many OEMs to use Celeron processors with 810 or 810e chip sets for value PCs, rather than the integrated Timna.

Timna — the name stems from a national park in Israel — was to integrate a processor, North Bridge, graphics engine, and a channel I/O hub on a single die.

"The standalone Celeron, combined with the 810 and 810e chip sets with integrated graphics, gave OEMs more design flexibility and the ability to design products with longer cycles," an Intel spokesman said. "Also, there was more cost flexibility with the standalone products. From a customer standpoint, they like the standalone Celeron better."

Timna might have survived but for another dagger, the need to combine it with a separate ASIC — a memory translation hub (MTH) — in order to translate Timna's Rambus signaling to enable support of standard SDRAMs. Despite several attempts and a notable delay, Intel hadn't yet manufactured a bug-free MTH for the 820 chip set. The Timna MTH "would have required another stepping [mask set]," the Intel spokesman said. "The month-long delay, from a customer standpoint, was significant."

Dean McCarron, an analyst at Mercury Research (Phoenix), said Intel's decision is "ultimately is a good one."

While some PC makers liked Timna, others disliked it because its high integration left little room for differentiation, McCarron said. "Some OEMs were fine with the inability to differentiate at that level, for PCs moving through the channel," he said. "We know of tier-one OEMs that were delighted with it. For some other OEMs, it was an issue, and that led to a high level of resistance to Timna. It was definitely a mix. In a nutshell, since the entire market was not receptive, that would have limited OEM demand, which was not good for Intel."

Also, the more conservative PC vendors were nervous about Intel's ability to create a reliable MTH for Timna-based systems. Following the embarrassing and costly recall of systems featuring the 820 chip set with an MTH, Intel was clearly not going to rush a faulty MTH chip to market.

"The cost savings of integration in Timna would have outweighed the cost of creating the MTH. But in retrospect, Intel probably wished that it had targeted Timna at SDRAM, though at the time there were significant technical concerns about the number of pins that would have been required to interface to SDRAM," McCarron said.

On top of OEM resistance and the MTH problem was a third issue — Intel's lack of manufacturing capacity. Intel planned to make Timna in a 0.18-micron process, would have consumed more silicon than Celeron manufacturing. "That would have created a situation of using more silicon to create fewer products, which is not good particularly when Intel is in a capacity-constrained situation," McCarron said. "Intel was explicit about saying that in their next [0.13-micron] process, Timna would have been die positive [consuming less silicon for more products]. But given Intel's position in the market today, Intel would have used more silicon to manufacture its cheapest product."

Despite the failure of Timna and similar integrated CPU projects in the past, McCarron said he believes Intel, Via Technologies and other suppliers will eventually succeed with a high-integration PC-on-a-chip.

"I'm sure you'll see Intel try again," he said. "Integration is critical to getting to a minimal cost solution. I don't see this changing Intel or Via's plan for integration.

"What this decision does do is validate the strategy of an appropriate standalone CPU with an integrated chip set," McCarron said. "This validates that that's the approach to use right now."

It also demonstrates Intel's newfound attitude of being flexible in response to market demand, he said. Rather than cling to strategies that are no longer valid, Intel is reversing course with Timna, as it did in an earlier decision to have the Pentium 4 processor support SDRAM.

While Intel never declared a formal price range for Timna, "there was an assertion that Timna would be somewhat cheaper than the cost of a Celeron and a separate chip set," McCarron said. "Given that particular guidance, we expected sub-$100 type pricing when Timna would have achieved significant volumes, perhaps the $80-to-$90 range for high-volume, low-cost products, and $115-to-$120 range for the faster parts. That compares with Celerons in the $60 range at the bottom part of the market and volume Celeron products in the $80-to-$90 range."

McCarron noted that Intel had planned to introduce Timna next year in the 700-MHz speed range. "It definitely was not intended as a bottom-feeding product, though it would be at the low end. It was never a compromise product."

The Intel spokesman said Intel does not currently have any integration projects under way, though he did not preclude another project at a future time. "The market has moved on to other ways of saving costs," he said.