Intel server strategy crashes as Xeon roadmap changes

INTEL IS A mess in the server sector. Yesterday it tore up its roadmap and all its great white(field) hopes have vanished in a puff of silicon dioxide powder. With them, CSI and TT appear to have vanished, as has chance of catching AMD before the Nehalem cores.

CSI, which Intel won't discuss, is a next generation interconnect architecture, something akin to AMD's hypertransport.

The changes Intel is forecasting are stunning. Whitefield is up and gone, replaced by Tigerton. Montvale moved back a year and so did Tukwila but that is old news. A thing codenamed Dunnington looks to have moved on a year too, which may be good or bad depending on whether 2008 means early or late in the year.

But it's the sub text that's the problem because Intel's fresh server roadmap has no details but codenames. Important elements like CSI on X86 and TT appear to be casualties. If this is the case, AMD has pretty much a clear path for three to four years, barring the recurrence of the AMD "shooting itself in its own foot" syndrome.

The Intel Paxville chip was the paper catch up chip, Dempsey was the real catch up chip, and Whitefield was the move ahead to a real chip, and that was because of CSI and TT.

But both seem to be dead. It is an open secret that the big problem with Whitefield was CSI integration, and the problems seem to have gotten the better of the part, or ex-part as it is. The Tigerton chip that replaces it is most likely the long rumoured Plan B, a Clovertown MP variant. This was a stopgap in case Whitefield slipped again. But it seems to fallen off the cliff and since it was the vehicle for TT and CSI, those went too. This means no HT-style bus for Xeons in the near future. Worse, this may mean the integrated memory controller (IMC) Intel was designing has gone as well.

Since Dunnington was related to the Caneland platform, it shares whatever memory and IO that Tigerton had. This signals that Dunnington is not the same Dunnington as it was last week, but what it is now is anyone's guess. There is talk of a new connect to the rest of the system, but Intel won't pin it down to a specific core.

It also could be a tarted up Pentium 4 with a better front side bus.

And there's more fallout on the Itanium. Without CSI bearing X86 CPUs bumping up the volume on the platforms, Itanium suddenly looks, if possible, less attractive than it was before. You need to do a high speed interconnect for something that sells 10s of server per company per quarter with no external volume to defray the cost, and no other teams to lean on for knowledge. This leaves SGI without a raft as Itanium crashes right into the priceberg. The raft called Dunnington was Intel's route to safety, and if it's true that CSI is gone, Silicon Graphics may be left in cold icy waters. µ

See Also
Intel performance lead unlikely until 2007