Paradox (paradox1000@gmail.com) on 2/9/10 wrote:
---------------------------
>Michael S (already5chosen@yahoo.com) on 2/9/10 wrote:
>---------------------------
>>gcc for x86 is significantly closer to state of the art than gcc on Power.
>>
>>Compiling with the same compilers as in the best vendor submissions, but with different
>>set of flags, in particular without PGO and without autopar, would be fairer. IMHO.
>>
>
>I don't know the internals of gcc (or have compared gcc/icc/xlc results), but that
>is likely true. However I still think the results would be interesting. (Even though
>it of course is a mixed measure of both cpu and compiler.)
>
>Is it also not true that icc is (or maybe was, some time ago) very heavily tuned for SPEC?
>
>If anybody has a link to SPEC results using gcc on modern x86 CPUs, then please share.
>

Well SUN made som specfp_2006 tests on Intel Xeon X5570 in a SUN Blade X6270, without autopar.

Compiler cfp cfp_base
Gcc version (4.4) 23.1 21.1
SUN Studio XX.X 29.1
with auto par.
SUN Studio 50.4 45.0

Just shows how totally useless non rate spec cpu results have become.

I haven't really been able to find any spec_int results, but it might even be worse.

// Jesper