This is exactly analogous to Tulsa. Tulsa was the last of the V8 Interceptors, er, P4-based Xeons, tricked out with shared L3 cache but pin-compatible with its predecessors. It was the best of its line, the non plus ultra refinement, but also the last gasp before a new architecture took over.

Given that this is an MCM and the L3 is shared, the cache must be its own chip in the package... which should make for a large and interesting package. But I don't see how AMD's problems with quad cores and shared cache predict anything about Intel's prospects. Yes, increased cores and increased shared state creates more potential for deadlocks and races the rest of the multiprocessor pitfalls, but there's nothing fundamental to "4+ cores and shared L3" that makes it an automatic failure. Tukwila is already exploring that terrain too.

And yeah, it'll be expensive. But it's a server processor, so that's expected (as long as the performance seems commensurate). For tasks that can use the cores, and for customers who want to increase their CPU density, it's likely to be a compelling product -- particularly for Sun, who specializes in both.

Wow. I thought Aces Hardware had gone under. Glad to see at least some of the gang is still around! They ranked with TR among the best of the Web in terms of technical info. Now TR is in a class by itself :)

Squeezing three dies on instead of a monolithic die. I suppose that is what Intel does best.