OK, I'm honestly curious. Why do you dislike 64-bit?
I know, as a scientist I should be basing my opinions on data. Meaning
CURRENT and
ACCURATE data.
My dislike of 64-bit mode comes from the early days of it, about 90s and early 2000s. We all used 32-bit machines on our desks and in our data centers until about 199x. CPUs with 64-bit instruction sets showed up in servers around the mid 90s (Power PC and PA-RISC), and then Intel came out with the Itanium (my neighbor Nick, a famous CPU architect, actually coined the name "Itanic" for it). One of the first things we noticed was that executables became significantly larger. That sadly makes some sense (and the Itanic is a specially bad case, as its instruction set was qualitatively different, neither CISC nor RISC). Then we started getting the first good machines for benchmarking, and we noticed that the CPI was actually somewhat worse. Our educated guess (later confirmed by experts, I'm not a CPU person) is that the machine used twice as much memory bandwidth for simple integer operations that went to memory, and for certain workloads the CPU was memory starved.
Add to that: I grew up in the era of 8-bit micros; I learned how to do productive work on a Z80 with 64K under cp/m, on a VAX 11-780 with a whole megabyte. And I shared an IBM 370/168 with 500 other users on 8 MB of memory (it did get pretty slow when all 500 people were logged in). So my instincts were honed to be as conservative of memory use as possible, which is why it emotionally hurts me to waste 32 extra bits in a pointer or integer on something that will never be used.
BUT: Those negative experiences and emotional fears are founded on experiences from 25 years ago. And on RISC machines plus that architectural abomination Itanic. In those days, performance-critical code was written in C, usually by people who looked at the assembly listings of the compiler. Today, the architecture I run on is the AMD 64-bit adaptation of the Intel x86 instruction set (different animal). And much code today is written to not be CPU efficient anyway (a lot is run in a JVM or a Python interpreter). Yes, I know shops where the biggest consumer of CPU cycles are Python programs ... so worrying about a 10% or 20% effect on CPI is completely irrational. My dislike of 64 bit mode is based on
EMOTIONS and
FEAR.
In the particular case of my server at home (a physically small machine, the size of a shoebox, with an Atom 4-core CPU running at 1.8 GHz), what really worried me that it was memory starved to begin with: It physically has only 4 GB of RAM (and I can't install more), and when running FreeBSD 12.X in 32-bit mode, I could only use 3 GB of that. So before I installed FreeBSD 13.1 in 64-bit mode, I was worrying that I would run out of memory (and have to swap more). In reality, that was DUMB: in 64-bit mode, the OS can actually use all 4 GB of the hardware, so I got 33% more memory for free.
You don't have to bother telling me that I need to upgrade (the hardware of) my home server, since I already know that. I'm busy with other projects right now, and it is working fine.
