32-Bit Is Very Fast
Firstly, and the most obvious reasons, 64-bit systems aren’t necessarily faster in “native” performance. It probably depends on the app and the hardware, different operations may be faster one way or the other.
Whereas the common logic in PC hardware is more=better this doesn’t apply universally especially if the system has some other necessary restrictions.
32-Bit Is Often Faster In Modern Use Cases
I use a lot of RISC-V hardware for development, almost all of my machines are RISC-V, but obviously sometimes I need to develop for an old platform or just run some old app.
In these cases, which are probably increasingly common use cases (and with other devices too), 32-bit is the clear winner for fully emulated guest operating systems. It’s just plain faster.
I Dare Say 32-Bit Is More Efficient
Especially if you’ve got a machine that needs to be responsive when needed but is idling most of the time, all of those high bits in your registers and your reserved memory are idling too… I don’t know the numbers but I can’t see any way that a 64-bit machine could be as or more efficient in many cases (except when actually making use of all that extra bus width and maths capabilities).
Can We Compete With Reliability?
A system that has it’s core rewritten every ten years for new architectures might get substantial improvements every time, but it doesn’t really give customers or distributors much guarantee that it will work for years. Rather it kind of feels like even new hardware is entering a planned obselecence cycle as soon as any modern software touches it. This even happens with new architectures, everyone buys the first decent model then the distros stop supporting that model.
So it’s not so much about whether there’s any direct commercial utility in supporting very old hardware, we know it’s better to just sell upgrades, but why is anyone going to buy a new system knowing their new CPU that still works fine will start being phased out unnecessarily in the software as soon as there’s a better model to sell?
New Hardware Often Isn’t 64-Bit First
We already have open source CPUs that are compatible with a lot of 32-bit software, whereas 64-bit CPUs are typically less practical for bootstrapping open source projects (although we do have those too, it’s probably easier to demo or to sell ao486 than it is to demo or sell a newer experimental system, and I’m a retail guy).
This is also a problem for people developing new CPUs: Do you try and make a 64-bit version that won’t fit in any cheap FPGA or do you try to make a 32-bit version that will probably work better for early testing but will be snobbed as outdated by much of today’s software?
Also For Nostalgia & Old Apps
I think this is the only thing other vendors are considering when they drop 32-bit support. Although I’m more interested in 32-bit systems because they’re very reliable & flexible, it’s also obvious that newer compatibility layers won’t be a complete solution and sometimes you’ll have to run a real OS for compatibility.