none (none@none.com) on 2/11/10 wrote:
---------------------------
>nemlis (noone@void.com) on 2/11/10 wrote:
>---------------------------
>>While I generally agree with your position, there's still one argument in favor
>>of Itanium: The compiler's probably much better at optimizing for the Core 2 than
>>it is at optimizing for the Itanium (which has been a headache from the start).
>>Intel excels at improving their x86 spec results by just issuing a new compiler version.
>
>I think it's a point against Itanium: If Intel could improve
>their compiler to get much better results on Itanium, don't
>you think they would have done it? Given that Itanium
>already is more than 10 years old, icc probably is very
>mature.
>
>Improving compiler technology was one of the things that
>Itanium design decisions relied upon. The problem is that
>in real life, the dynamic behaviour of a program is hard to
>predict, so that having OOOe is a much better choice as your
>processor will kind of adapt itself to the load.

Right, one can actually view it from two different perspectives: The architecture prohibiting optimization beyond a certain degree (your point) or the compiler not being able to leverage the cpu's full potential (my point).

I think your point generally bears more weight, but in certain cases (code tailored to the target platform/HPC), the SPEC_CPU-based clock-for-clock comparison would unjustly favor x86, because it's a too general benchmark.