Well, it does matter in that exact case where the board doesn't have that much memory: The executables get larger, and the cache becomes less efficient because every address takes more space (this is an issue also if there's enough RAM to make use of 64-bit addressing).
It's enough of an issue that the most popular executable format on SGI was n32, not n64 (
edit: nor o32 - "old" 32). Well, as I mentioned before - that n32 format does give you the best of both worlds: 32-bit addressing, more efficient cache (very important), but access to 64-bit registers for integer arithmetics. I wish there were an ARM architecture that could do that. (Maybe there is - there are so many and I don't have the full overview. But I'm not aware of one.)
And floating point - floating point registers have been handled separately for a long time on common processors, Intel ia32 has had 64- and 80-bit precision long before ia64 and amd64 (aka x86-64) were invented, and the 32-bit Raspberry Pi2 (ARM Cortex A7) has that as well - with its Neon FP unit: 32 64-bit registers, which can be used as 16 128-bit combined registers. Quad precision has been used for ints and floats since the (32-bit) VAX

. (I have used VAX, but after a little checking it turns out that the IBM System/370 had 128-bit hardware floating point before that, and some other systems of the era had 128-bit ALUs).
The Cortex A7 has another, non-vector FP unit which also does 64-bit floating point in hardware. Actually the Cortex A53 in the Pine has the same Neon and FP units as the A7..
Yes, as soon as you want to address a ton of memory (virtual or physical) you need 64-bit. But at that point you're unlikely to do your development in assembly.. so it doesn't really matter. As far as I understand, that Pine board cannot be upgraded to hold, say, 16GB of RAM. It can hold 2GB though (previously 1GB), they just updated that kickstarter page with some more info (unfortunately the second link in the OP went dead btw). So if you really need a 64-bit processor, e.g. if you do the kind of processing I do on my desktop, then you can't use that Pine board. You'll have to buy a new board anyway. And at that point it doesn't really matter if the architecture is otherwise exactly the same - because you'll be programming in a high level language on a board with an operating system. I can move transparently between a 32-bit *nix computer and a 64-bit computer with my applications, it's just that the former can only process small datasets. The code I write is exactly the same. It even builds for and runs on a Nokia N900 phone, a different architecture (Cortex A8). Not enough RAM to be useful though.
So, with the Pine, what you get compared to a Raspberry Pi2 is, roughly:
- + Slightly cheaper
- + 64-bit hardware integer arithmetic
- - Larger executables
- - Less efficient L1 cache
- + The Cortex A53 is generally faster than the Cortex A7, and is a bit more power
efficient
So it's not unlikely that due to the A53 being a newer processor than the A7, the L1 cache may be just that bit faster to compensate for the lower efficiency of 64-bit programs. But I couldn't find any numbers of that.
I fully support that ARM goes 64-bit, but that is in order to compete on the server- and desktop side.. for small memory-limited boards like this it's more of a nuisance and it doesn't in my opinion make any difference for what the user wants to do in the future, with a newer, larger board.
The Pine is a nice board. But to justify it being 64-bit, which, as far as I'm concerned, is a *limitation* for that board, if it had been designed to easily accomodate 16GB or 96GB of DRAM, and if the A53 has a cache system designed to handle that (which I'm not certain about), then yes it would be just the expandable board they claim it to be. As it is, I don't really see it as more expandable than the competitors, and I definitely don't see that their claim of it being a supercomputer can hold up.. that word should be reserved for whatever is the current architecture for weather forecast modelling and the like. My phone is faster than the Cray-1, but it doesn't make my phone a supercomputer even though the Cray-1 was one..
(But yes, if I could put *at least* 16GB on that board, I would buy it as quickly as I could. Because then I could fully replace the computer I lug around when I travel between my two home sites and elsewhere. Would just need to stash big monitors at each destination point. With 2GB, the new apparently maximum on that board, it's basically useless for the kind of 64-bit processing I need, or in fact any 64-bit real-life processing I can think of, and compared to other boards it'll need that bit of extra RAM it already has just to compensate for the extended space of the applications and operating system. When I look at that board with 64-bit eyes I simply see a severely RAM-constrained board, not useful for more than whatever any 32-bit small board could do as well as or better than the Pine. I have several small boards, as far as I can see the Pine won't outperform them on what such boards are useful for. I could be missing something, but I just can't see that it being 64-bit is a selling point. It's rather the opposite, in my opinion.)