Comment 20 for bug 585940

Short: I agree with all the above.

Long:
The central point about 32-bit machines is that a single program cannot address more than 4GB of of memory at all (other than with tricky added hardware and software that is largely now irrelevant). In addition, to keep hardware costs down, consumer machines, especially laptops, reserve some of the RAM for video memory. Sometimes this is even effectively true when there is an added video card with its own memory. In practice, most 32-bit consumer desktop/laptop machines can use about 3.5 GB for programs. By contrast, 64-bit machines can offer several advantages. Among these are (a)they can exploit more than 4GB for program memory (b)many of their instructions can load and execute faster (c)they can compute with larger numbers more immediately, which is usually of little importance except for scientific computing; (d)more programs can remain resident in memory without swapping out to disk, so changing from program to program among several running programs is without noticeable delay. Of course, some of these accrue only if the system actually has more than 4GB of hardware on it, which is rarely true of current consumer-grade 64-bit machines. That will change as memory density increases, in its usual fashion.

In the history of digital computing, as memory addressability went from 8 to 16 to 32 to 64 bits, programming technology has always found a way to exploit the increased memory to increase speed or to dramatically support new kinds of data structures to good effect. It's typical of application programs that if recompiled on a 64-bit machine they can exploit such advantages as above. Only the few that fundamentally touch the hardware, e.g. video, may not be able to, unless the kernel programmers have completely done so at the system level, and the application programmers have not tried to circumvent the O/S for improved performance.

Probably programmers, and perhaps gamers, are typically in need of all of the aforementioned advantages, and probably most people who have such need would be capable of understanding the arguments leading to "not recommended". Whatever those are, one hopes they disappear as 10.4 matures. Meanwhile those who can benefit from 64-bit architecture and have it, would be well-served if they could learn where the sticky points are. I don't want to spend my time nudging early post-beta releases of Ubuntu, but I also don't want my 6GB systems forced back to 4GB. And I am always grumpy when I feel compelled to use more than one version of an O/S, as I do at the moment: 10.4 on my 32-bit machines, 9.10 on my 64-bit machines.