Rambus Meltdown a sorry tale of fudge, mudge and kludge

The aptly named "Rambus Meltdown" er, Plugfest, was hosted by Intel in a Silicon Valley hotel Tuesday and Wednesday of last week. Revision B0 (B-Zero) silicon has just started sampling, but it will not go to production as planned. Intel calls pre-production silicon A0, A1, A2 until it is certain that the next rev will really work - then it calls that rev B0. Well, you guessed it, Camino rev B0 didn't quite make it. Now Intel is hoping that Rev B1 (sampling in July) will be the real thing. But Intel has not yet officially changed its already thrice delayed launch date. Intel has instead enhanced the schedule to "late in September," rather than just "September". At the meltdown, Compaq and Dell each slaved over three unique board desings each with the aid of Intel and Rambus personnel, working out the details and trying to get them to run up to speed with RIMMs from several different DRAM suppliers. IBM unenthusiastically showed up with only one board design. No other OEMs were present. All of the boards are six layers - now accepted as the minimum required to get 400MHz running. The challenge now is to get Rambus running at 400MHz on the same board with a 133MHz CPU bus and a 133MHz AGP4x bus. It seems doubtful that even Intel can pull this one off with its Vancouver OEM board. If rev B1 does not fix these minor annoyances, AGP will be the first to go - dropping the platform spec back to AGP2x (66MHz). Then if it cannot be made to work at 133/800, the RDRAM speed will be dropped to 712MHz. If that doesn't work, then "late September" might as well become "Next Millennium". A couple months ago, the Intel proposed solution was to drop Rambus to 600MHz instead of 800MHz and to shrink the maximum system configuration from a rather slim 256MB to a paltry 192MB. Surprisingly, no one seemed to complain about the DRAM capacity issue, but after loads of benchmarking, the OEMs came back yawning over the performance figures. Couldn't even match the BX at 100MHz. It was then that Intel started shifting gears to define the 712MHz and 533MHz speed grades. 533MHz offers bandwidth equal to PC133, but worse latency. Who would want this? 712MHz is a rather odd speed derived as follows: 133.3*8/3*2. Synchronous? Yes. Logical? No. 666MHz (133*5) would have been a more logical choice, but a little too Satanic for Intel marketing. It is amazing what lengths Intel will go in an attempt to avoid the already obvious associations. Now, with a veritable smorgasbord of Rambus speed grades, CPU bus speeds and AGP speeds to choose from, Intel is hoping that somebody will be able to ship something this year. The question remains - Will anybody want to buy it? ® * Rambus rose on Wall Street yesterday. See also: What the hell is Rambus and Camino all about?