[FFmpeg-devel] Inline ASM vs. Intrinsics

Trent Piepho xyzzy
Fri May 11 23:17:35 CEST 2007


On Fri, 11 May 2007, Michael Niedermayer wrote:
> On Fri, May 11, 2007 at 07:44:53PM +0100, M?ns Rullg?rd wrote:
> > > IA64 is a complete failure with and without intrinsics
> >
> > IA64 is a complete commercial failure.  Its performance is far better
> > than any x86-based CPU of the same time period.  The main reason it
> > failed was lack of good x86 emulation, and people insisting on
> > continuing to run the same old rubbish non-portable software.
>
> really?
> can you point to some benchmarks? (not from intel of course)
> i thought it was significantly slower than compareable CPUs (same time period
> and same price range) even when both run natively compiled code
> (with similarly good compilers of course)

I designed a Beowulf cluster for the company I worked for shortly after the
time the AMD Opteron came out.

At the time, Intel's fastest Itanium2 was faster than AMD's fastest Opteron.
Though I think the Opteron had a slight edge in SPECint with the Itanium2
better by a non-small margin in SPECfloat.  It's been a while, I don't
remember exactly.

For the code we wanted to run Itanium2 _was_ faster, I know that much.  Still,
the Itanium2 system was the easiest to remove from consideration.  The math
ended up that for our budget we could buy a system with X Opteron CPUs, or
something like X/4 Itanium2 CPUs that were 20% to at best 50% faster than the
Opterons.  The equal cost Itanium2 system was well under half the speed of the
Opteron system.  And that's just capital cost!  Supporting ia64 in addition to
the existing ia32 infrastructure has a cost too.

The Opteron was quite a bit better in facilities too.  You could fit a dual
CPU system in one rack unit.  I don't think Itanium2 would pack better than 1
CPU per rack unit.  Power consumption (and thus A/C needs) were also quite a
bit better for Opteron vs Itanium2.




More information about the ffmpeg-devel mailing list