Re: 64-bit vs. 32-bit Computing

Jon,

I re-read your post, and realized that if you're talking about a new
Dell server, you're probably referring to the new dual core Xeon (not
"Core Duo", stupid Intel branding), right?

...regardless, to the issue of 32 vs 64bit, that really is a big
question. Obviously, as you said, in the server arena it's good to
have full 64 bit addressing. Of course, many modern "32bit"
processors can address more than 4GB of RAM, and some 64 bit CPUs
(like the G5) use only 42 bits of physical address space, for support
of "only" 4 terabytes of RAM.

But when the Athlon 64 came out, there was a lot of rumbling about
whether or not the 64-bitness of the new processors offered any real
advantage. Consensus seems to be 'not really'. There are certain
things, like decrypting 4096 bit RSA keys that can be done roughly 4
times faster in a 64 bit environment on 64 bit CPU...then there's
everything else.

http://www.pcstats.com/articleview.cfm?articleid=1665&page=8

I'd guess any computational biologists on Micronet may be able to
chime in with some real world applications, such as nucleotide
sequence matching in BLAST, or protein sequence analysis in HMMER.

But even for the "average" server app, it doesn't seem like a big leap
performance wise. At least not today.

John

Jon Forrest wrote:
> There's no doubt that for certain applications the
> increased address space provided by a 64-bit computer
> is a good thing. But, there aren't that many of these
> applications.
>
> Now that the new 64-bit Intel Core Duo/Solo processors are
> out, and now that new AMD Athlons are all 64-bit, I'm
> wondering when and why people are running them in
> 64-bit mode other than for database applications.
>
> I just did a quick test on a new Dell server with a Core Duo
> and 1GB of RAM. I loaded Fedora Core 5 in 32-bit mode and
> timed a build of a fairly big app (subversion). I then loaded
> Fedora Core 5 in 64-bit mode on the same PC, and timed the
> same build. I think this is a reasonable test since
> it does a lot of both I/O and computation. The results weren't
> what I expected.
>
> It turned out that the build took 2 minutes of elapsed time
> in both cases. However, the 64-bit build took 12 seconds less
> of usermode time, and 12 seconds more of system time. Go
> figure.
>
> --
> Jon Forrest
> forrest@ce.berkeley.edu
> Computer Resources Manager
> Civil and Environmental Engineering Dept.
> 305 Davis Hall
> Univ. of Calif., Berkeley
> Berkeley, CA 94720-1710
> 510-642-0904
>
> ------------------------------------------------------------------------
> The following was automatically added to this message by the list server:
>
> For information about Micronet, including subscribing to
> or unsubscribing from its mailing list and finding out
> about upcoming meetings, please visit the Micronet Web site:
> <http://micronet.berkeley.edu/>.

------------------------------------------------------------------------
The following was automatically added to this message by the list server:

For information about Micronet, including subscribing to
or unsubscribing from its mailing list and finding out
about upcoming meetings, please visit the Micronet Web site:
<http://micronet.berkeley.edu/>.
Received on Thu Apr 13 01:25:02 2006