Archives


- Beowulf
- Beowulf Announce
- Scyld-users
- Beowulf on Debian

Why is 64 bit faster. RE: [Beowulf] Win64 Clusters!!!!!!!!!!!!

Many of your questions may have already been answered in earlier discussions or in the FAQ. The search results page will indicate current discussions as well as past list serves, articles, and papers.

Search

Andrew Shewmaker agshew at gmail.com
Sat Apr 14 00:55:21 PDT 2007


On 4/12/07, Mark Hahn <hahn at mcmaster.ca> wrote:
> > 1.) Why is a 64 bit cpu faster?  I had assumed the main benefit was the
> > memory that could be addressed, obviously a bad assumption.
>
> being able to address more memory is indeed critical for some codes.
> certainly not all; in fact, the larger pointers hurt some codes.
> 64b mode also enables a lot more registers, of which the compiler can
> make very good use.  and some codes like the extra register width.

Real World Technologies has some good threads on what makes x86-64
better than x86.

People have mentioned the extra registers you get when going to
x86-64, but I don't
think anyone mentioned how that was related to the x87 unit.  x86-64's ABI
forces people off of the x87 unit (mostly), and that's a good thing
for several reasons.

http://www.realworldtech.com/forums/index.cfm?action=detail&id=68666&threadid=68664&roomid=2

Increasing the number of logical registers beyond a certain point can
have negative
consequences (increased write-back of intermediate values).  Register renaming
in hardware works pretty well.

http://www.realworldtech.com/forums/index.cfm?action=detail&id=75516&threadid=75235&roomid=2

For those worried about 64-bit pointers, they also talk about
"compressed pointers"
(aka indexes).

http://www.realworldtech.com/forums/index.cfm?action=detail&id=75624&threadid=75463&roomid=2

And why 64-bit pointers don't necessarily hurt performance that much even if
you don't compress them.

http://www.realworldtech.com/forums/index.cfm?action=detail&id=72864&threadid=72839&roomid=2

One nice benefit of x86-64 is that it allowed the Linux kernel
developers to clean up.

http://www.realworldtech.com/forums/index.cfm?action=detail&id=64142&threadid=64137&roomid=2

There's tons of other discussion about how 32-bits of virtual space
isn't enough, how PAE
sucked, why simple flat address spaces are good, and why x86-64 made sense even
if only a small percentage of the market required 64-bits.

-- 
Andrew Shewmaker



More information about the Beowulf mailing list