OK, but wait. 2.9BSD is from like 1982. And I think SLIP was later than that, and PPP was definitely much later than that. So you don't have a "pure" 1975 machi

OK, but wait. 2.9BSD is from like 1982. And I think SLIP was later than that, and PPP was definitely much later than that. So you don't have a "pure" 1975 machine; you're using later software. What does "support" mean, then? If you're allowed to use software that's later than your hardware, then you can write a stack for almost anything, so the idea of a machine "supporting IPv4" becomes kind of meaningless. You need 576 bytes of memory for a packet buffer, plus your program. Any mid-sixties mainframe, maybe even older. If you do count the age of the software in deciding what's "supported", and you require that the TCP/IP stack run locally and communicate with modern computers without translation, then you'll have a hard limit at about 1980 or 1981, since the TCP and IPv4 protocols weren't defined until then, and there was therefore definitely no software. You could allow using NCP as a "legitimate Internet precursor" to TCP/IP, but if you do that there's no real reason not to just allow serial terminals as "legitimate precursors" to NCP. So 1980 or 1981 seems like the limit for a "pure" system with its own software. If you allow software newer than the hardware, but limit yourself to not writing that software, my guess is that the oldest thing would be a KA-10, which is a 1968 computer. RFC 801 says that as of 1981, there was a stack for TENEX, which you could run on a KA-10. You could at least do Telnet, maybe FTP. That's assuming you can find another old computer to gateway whatever ancient serial protocols it was using at the physical layer.