Does anyone have any data (anecdotal or otherwise) about
relative speeds of 32- and 64-bit executables (both compiled
with optimization using Intel compilers) on the new Intel
Nocona processors?
Background: On SGI/MIPS and IBM/pwrN architectures, I have
found that 32-bit executables run a bit faster than 64-bit
executables, due presumably to greater cache warmth when
integers and pointers are 32 rather than 64 bits in length.
This is a small effect -- maybe 5% in speed.
Opteron is different. As discussions here about a year
ago pointed out, Opteron 64-bit executables have access to
several registers that 32-bit executables do not. Therefore,
on Opteron, 64-bit executables run somewhat faster than
32-bit executables. A 12% speed improvement was cited as
either typical or maximal (I forget which).
Does anyone here have data for a similar comparison for the
new Intel 64-bit chips (Nocona)?
Thanks,
-P.
--
Peter S. Shenkin Schrodinger, Inc.
VP, Software Development 120 W. 45th St., 32nd Floor
646 366 9555 x111 Tel New York, NY 10036
646 366 9550 FAX USERID: shenkin
http://www.schrodinger.com DOMAIN: schrodinger DOT com
Pre-arranged conf. calls: 816-650-0812 or 888-667-9995; passcode 060428#
|