I did software development for NonStop Kernel on Itanium for a few years. 5 years ago, Itanium was FAST. Our software was twice as fast as competitors who used twice as many processing cores. Problem was that, unlike x86, Intel's Itanium roadmap moves slow and continually plagued by problems. About 2-3 years ago, with cost reduction in mind, we began porting software over to Linux on x86. Initially we needed 2.5x as many x86 cores to reach the performance of the 1.6Ghz Itanium 2. By the time Intel began shipping Westmere-EP processors early in 2010, each core was nearly 1:1 with Itanium, and at half the cost and 3x the density.

Interesting. The NonStop has always intrigued me as a platform. Thanks.

I've never worked with it, but the idea is that it's unbelievably fast if you have a smart enough compiler. Intel's x86 chips have a lot of logic onboard that will optimize code as it's running, in real-time. Itanium uses all of its silicon to do whatever it's told really, really fast and just trusts that the compiler has figured out all the optimizations ahead of time. Also, it has mountains of cache memory, so that helps too :)

The problem was coming up with the Sufficiently Smart Compiler to turn branchy code into explicitly parallel instructions. It never really turned up for C/C++ and Fortran code.

It would probably fly for languages like Haskell or Erlang.