32-bit 4eva
I see in LXF203 [Mailserver, p11] the debate goes on in terms of 32-bit vs 64-bit. Basically they work at their best on differing tasks, yes. But newer 32-bit designs are more advanced and take i386 etc, hardware Java bytecode is always 32-bit and Ethernet packets are 32-bit and big endian, still? TimoOjala, Finland, via MMS Neil says I think we’re drifting into another territory here, but the idea that 64-bit or 32-bit is more highly optimised is something of a
nonsense, as when you look at machine languages they run the same opcode. We generally talk about 64-bit in terms of addressing, but along the way the instruction architects took the chance to add improvements to 64-bit implementations.
AMD64 (x86_64) had more registers added to it for increased speed. With ARM v8 (already a highly optimised Risc language) many more registers were added, along with new SIMD instructions with twice the registers, improved 64-bit floating point pipelines, dedicated hardware crypto and, of course, assumed 64-bit addresses. All of that adds up to a 25 to 33% increase in speed over the older 32-bit implementations.
So a more modern architecture can be far faster and offers access to vastly larger memory addresses. Java is ‘stuck’ being 32-bit as it needs to be portable, which is useful for low-cost embedded systems, where memory isn’t an issue. As for Ethernet that’s a protocol, so it is irrelevant to the issue. But as we always point out, we do run 32-bit distros and indeed have 32-bit Mint on this month’s cover disc. The scary thing for 32-bit people, is examples like OpenSUSE Leap that has no 32-bit version available anymore.