Carmel ain't the solution to Intel Rambus SNAFU

The much-anticipated announcement of the 840 (Carmel) chip set next week may not do much to soften the pain of the Camino snafu. The 840 works, the 820 does not - but that difference may have more to do with the platforms than the chip sets. Both chip sets were designed using the same Rambus channel interface logic, so the differences must lie somewhere else. Further investigation reveals that there are several key system level trade-off between Cost, Performance and Reliability that allow 840 system to run, while 820 systems fail. 1. Carmel platforms use a six layer motherboard. This adds cost but improves signal integrity -- one of Camino's problems. 2. OEMs may play it safe by using 600MHz RDRAM on 840 production platforms. This improves timing margins and reliability at the cost of performance. This trade-off has proven unacceptable for uniprocessor PCs running business apps. Workstation and server buyers prefer ECC (requiring 18-bit RDRAM rather than 16-bit). ECC helps with the reliability issues, but at an additional 20-30 per cent premium over the already high cost of Rambus. The 840's dual Rambus channels are interleaved, and reportedly cannot operate in single channel mode. Each channel has only 2 RIMM slots, which is good, but configuration options are limited. RIMMs must be installed in matched pairs. 2 RIMMs or 4 RIMMs - thats it. Not quite flexible enough for low cost systems. These 4 trade-offs are unacceptable for the mainstream PC market - but they are why the 840 works and the 820 does not. Catch-22? This perhaps explains why Intel seems prepared to launch Camino platforms entirely void of Rambus RDRAM as a "Time To Market" solution (see: Intel i820 update leaks). Intel's internal memos point to the '0+2' Platform as the preferred near term solution. This suggests that the first generation of Camino systems (presumably in Q1) would have no RDRAM slots at all, just an MTH chip and 2 SDRAM slots on the motherboard. Amazing… It is hard to anticipate any demand for Rambus in the PC market with a strategy like that. The final question then is: "Will the 840 inspire a quick upturn in Rambus demand?" The outlook here is not exactly bright either. First, the 840 is currently aimed at comparatively small market segments (workstations and small servers). Second, buyers of these do not usually rush to adopt a new and unproven platform technologies - particularly in light of the well publicized reliability and performance problems with Rambus. The near term elasticity of Rambus in these markets is questionable. If Intel does not fully productize the 820 for RDRAM soon, it could stand accused of leaving its many supporters in the lurch. Chipzilla is between a rock and a hard place: launch Rambus-enabled Camino early, and it risks humilation. But cut Rambus out of the loop altogether, and then the Cow Pie will really hit the fan. ®