I'm an idiot.
Yup
Not even going to edit that. Please mock me.
If you don't deal with software a lot, basically 32-bit CPUs limit you to a 32-bit memory address space. Memory in this case being RAM. That ends up being 4 GiB (2^32=4,294,967,296 = 4 GiB). GiB is Gibibytes, which is power of 2 and to a certain extent reasonably close to GB Gigabytes (4,000,000,000 Bytes).
When looking at performance you should look at core speed and core instructions per cycle (IPC). That's harder to do with ARM chips. x86 has a bunch of benchmarks that give you realistic estimates of core performance.
You know that 64-bit has to do with more than memory right? Memory addressing is just one aspect of moving to 64-bit processin. There is also 64-bit floating point precision and calculation, as well as CPU operations on 64-bit data. Epic showed off really cool Unreal tools almost a decade ago utilizing 64-bit processing... Long before 4GB workstations were even a reality.
That's true. I'm not familiar with mobile, but Apple does seem quick to ditch backwards compatibility in favor of new hardware so maybe they're using it. In desktop space with enterprise software, many apps use ancient instruction sets.