For the 'Big Tin' machines (think 100 CPUs+), Itanium has its niche and Opteron isn't really a threat to it. You kind of got that backward. Itanium is currently only excellent for dedicated floating point crunch, and doesn't scale well beyond two or so processors in a single system image.
Opteron, on the other hand, has one of the fastest ccNUMA interconnects that money can buy designed into the system, and will scale almost linearly out to around 8-procs and scale well to much higher numbers. And the v2.0 of the interconnect is going to be out soon, which will increase the linear scalability significantly further.
Opteron was intentionally designed and highly optimized for massive Single System Image type computers. In this regard, it scales FAR better than Itanium, which is designed more like a workstation processor.
"You kind of got that backward. Itanium is currently only excellent for dedicated floating point crunch, and doesn't scale well beyond two or so processors in a single system image."
Huh? Err, perhaps you need to take another look at the
TPC-C benchmarks which show the Itanium-based HP Superdome systems occupying a good portion of the top 10 systems. 64-CPU Itanium 2 systems crush everything else tested. Itanium is not, and never was intended for 1-2p systems. That market, would be Xeon's. Hence the need for Xeon (1-CPU solutions), XeonDP (2-CPU solutions), and XeonMP (4+ CPUs). Itanium was designed from the start for memory addressing and scalability. While Intel does also sell them for single and dual processor machines, that's done primarily for single application solutions. Right on Intel's product overview
page for Itanium 2, they talk about its support for large SMP and clustered solutions.
Opteron does scale very well, but for non-clustered solutions, Itanium owns the big tin. I'm a big supporter of Opteron, and I absolutely adore the architecture. Opteron was, however, always intended to primarily be a 1-8 way processor. The success with clustering and the Opteron has won AMD some pretty big-time customers, but Itanium's niche is, as yet, unchallenged by much of anything. That being said, Itanium's price and performance benefits against the Sparc and Power chips are easily consumed by Opteron and - soon - 64-bit Xeons. If those two processors push Itanium into a small enough niche at the top, it's not going to be economically viable to continue producing it, unless and until Intel can bring production and development costs down significantly. Itanium was originally supposed to replace x86 altogether. Since that time, it has slowly been further and further relagated to a niche. It's now sold exclusively as an 'enterprise' processor for enterprise software. That could very well be a stable spot for it, or it would be that's what kills it.