I noticed that when I deploy my distro there is an option between 32 and 64bit version of Ubuntu 8.10.

Which is the best version to deploy?

Will I get better performance out of one compared to the other?

I'd say 32-bit, unless you have a specific reason to use 64-bit. 64-bit software will use more RAM, there probably won't be a noticeable speed difference, and not all software will run on a 64-bit system.

Cool, thanks for the advice I will stay with 32 bit.

Do you have any statistics regarding the difference in RAM, in actual usage?

None at all.

However, there would have to be some increase, and even a small amount of RAM can matter on a VPS.

I've heard an anecdotal account/claim of some Ruby app using 50-100% more, but that's it.

I just went from Ubuntu 32 bit to Ubuntu 64 bit. Apache + Django is noticably faster, but with memcached turned on most of my site's users won't see much difference. RAM use increased on my setup by 200M+, so I went from a Linode 720 to a Linode 1080.

James

Alright. Thank you for providing some real information, instead of my guesses.

Just out of curiosity .... Why would you pay more, if there is no performance increase? Or is this is an upgrade path that you are taking to accommodate for future needs?

Good question!

I wrote this article comparing Slicehost and Linode, but the same principle would apply when comparing any 32 bit system to a 64 bit one:

http://journal.dedasys.com/2008/11/24/s ... -vs-linode

I switched to Linode because they provide a 32 bit option.

I think he said noticably faster.

Yes. However with memcached on, users don't see much benefit. I went back to 32 bit and kept the larger Linode - RAM, sweet RAM.

James
http://zunzun.com

I just did a quick python pystones(65789 vs 54000) and local http transfer test ( 640MB/s vs 500MB/s). So x86_64 seems about 20-30% faster.

Only 39M memory is used with cherokee http server running, 74M with both mysql and http server.

On 64-bit having 'bigger stones': Integer math yes, floating point no. My need is faster double precision floating point math, and going to 64-bit did not provide this.

James