Does it really work. I've tried it and it just barely works for a few simple tests, for anything else it fails.

Most software is written for 32-bit. (Think about it. If it were written for 64-bit, how would it run on 32-bit systems?) Your speed increase is primarily in math, since the math can be done in 64-bit, even if the instructions are 32-bit.

However, when choosing 64-bit OS, all your pointers are twice as big, which eats into your L1 and L2 cache. If you can only store half the data (because each 64-bit element is twice the size of a 32-bit element) in the cache, obviously the cache has to be refreshed twice as often, which slows things down.

Bottom line: IMNSHO, if you have more than 2 GB memory and you do a lot of gaming, ray-tracing, video rendering, or other mathematical operations, you can benefit from 64-bit. Otherwise, you might as well stick with 32-bit. YMMV

Are you sure you gain in math operations? Maybe in precision, but in speed?
I thought you gain when copying/moving a lot of data (like when you move a part of a large pictures with the Gimp), because you move 64 bit at once instead of 32 bit. Am wrong about that?

Databases seem to gain the most at the moment.

Most software is not optimised for 64 bit right now.I am not talking about compiling but the source code.
The question that makes more sense would be:

Are all the apps I need working with the distro I plan on running with 64 bit?

In this context "math" typically means floating point. In floating point, you will NOT gain precision switching to 64-bit, you might loose precision. You probably gain speed.

The speed gain is because in 64-bit the compiler generates code for SSE floating point instead of x87 floating point and/or because 64-bit mode has twice as many SSE registers.

Both 32-mode and 64-bit mode support 32-bit and 64-bit floating point via SSE and both support 32-bit, 64-bit and 80-bit floating point via x87. There is no difference in the precision supported.

When the program specifies 32-bit or 64-bit floating point and the compiler uses x87, some intermediate values will be 80-bit. For example, if you compute (A*B-C*D) with 64-bit floating point and A*B is nearly equal to C*D then x87 would give a more algebraicly correct answer by using 80-bit intermediate values, while SSE would generate the IEEE standard (algebraicly incorrect) answer. So 32-bit mode may have more precision (due to the fact that the compiler uses x87).

Wrong about that too.

Both 64 bit mode and 32-bit mode can move data in 64-bit chunks and at the same speeds.

I have been using oss 64 on both my laptop and desktop for about a year. The reason I chose 64 bit is an article I came across that said the "retro attack" only affects 32 bit systems. On the down side, some software has no 64 bit version and the 32 bit version does not work in 64 bit environment. I just presume the software is not secure.

Well yes, it has its limitations. When I bump into one of those, I switch to 32 bit Opera. There is a static binary available that does not require installing 32 libraries.

Most 32-bit programs will work just fine through emulation as long as the 32-bit dependencies are met. So far my only problem is with java applets, that's why I have to keep a 32-bit FF around, so when I use java applets (rarely), I'll start that FF.

RAM is cheap these days.

Ramdisks are one of the best things you can learn to utilize.

You might want to consider if you can make use of enough RAM to support going to a high enough point to need a 64-bit OS over a 32-bit OS simply to address the extra space you could have. Using ramdisks for programs and often read data is a huge performance boost for most things, let alone if you have the extra few gigs to put that entire new id or epic fps in ram! There's more to it than the standard arguements. If you have the RAM and aren't using it, consider if it might benefit you more to start using it.

Also, keep in mind it's: 4GB - "all other addressed memory from other places (video cards, sound cards, controllers, etc.)" = available RAM
A simple 2 video card setup with 512MB on each card would drop you down to 3GB before even considering anything else in the system! The cheap entry cards are really pushing these sizes, too. I've seen things like the 8600GT-512 pushed heavily as a easy eye-catching big numbers is better, end-cap/bargain bin money maker.

Just some thoughts.

EDIT: OS into ram howto started by an LQ gent here: http://www.linuxquestions.org/questi...ilence-662116/