On 5/9/05 8:03 AM, "Michael Kluskens" <email@hidden>
wrote:
I've discovered that a certain unnamed commercial Z compiler can
produce executables that are as much as 2.5x slower than another
compiler for the same Z language (at maximum optimization).
Obviously not clear enough. Compiler 1 produces an executable that
takes 488 seconds on a dual 1 GHz G4. Compiler 2 produces an
executable that takes 201 seconds to do the same task on a dual 1 GHz
G4. Lets say a 1.8 GHz P4 using an Intel compiler takes 133 seconds
(estimated based on previous tests).
If all you have is Compiler 1 on OS X, odds are a 2.5 GHz G5 will not
beat an obsolete P4.
On May 9, 2005, at 9:06 AM, Brian Raymond wrote:
I assume you mean across platforms because otherwise you wouldn't have
the
problem you do.
Wrong assumption.
It's accepted that
the G5 is faster clock for clock but when you can get Xeon chips at
almost
1GHz more then the G5 they compensate with the extra clock cycles.
Depends on the compiler and language you use. Assume the language is
not C, then you're dependent on a narrow range of compilers on OS X,
some of which don't do that well in which case the Xeon will beat the
G5 every time on every problem, it's not the G5 it's the compiler.
I made the mistake of trusting my compiler vendor, trusting that they
put serious work into their OS X compiler. On workstations in the past
the compiler vendor and the workstation vendor were one and the same
and we made purchasing decisions based on benchmarks we ran with
existing codes on their machines (SGI & DEC in the past).
Not much different then the point that Office 2004 Excel is slower than
Office XP on Virtual PC 7 on the same hardware.
Michael
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Fed-talk mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/fed-talk/email@hidden
