The difference being of course that this 4.8-second compile in 2002 was done on an IBM p690 which cost half a million dollars. Whereas the 15.6-second compile was done on the EPYC 7601 which is only $4k for the CPU, ~$6k for a whole machine.
Last response from AMD: "The vast majority of users using Ryzen for Linux code and development have reported very positive results. ... A small number of users have reported some isolated issues and conflicting observations. AMD is working with users individually to understand and resolve these issues."
I doubt there's any single root cause here. "I get crashes when running big compiles" is a classic symptom of hardware that is almost, but not quite, stable. This can be due to power delivery problems, thermals, marginal RAM, overclocks, faulty CPU (yeah, it happens sometimes) and so on.
There used to be a long running joke (I think perhaps Linus even coined it?) that the kernel codebase grew about as fast as CPU performance improved. I guess that died around the time AMD released their dual core CPUs..
It seems reasonable, considering anyone working on it enough to need to recompile would have a minimum change time correlated with compilation time. (Incremental compilation aside)
Going to get a cup of coffee takes the same amount of time as it did 15 years ago, so devs still need a compile time which is roughly equivalent. 15 seconds would be much too fast, for example.
I'll admit the endless compiling was often torturous, but I have the Gentoo docs to thank for most of what I know about the innards of Linux and operating systems. Bet I'm not alone in judging the time spent worth it :)
No, I'm sure lots of things have changed. More of the code would be using C99 semantics now, and I have no idea how GCC would have improved or regressed since then. It's possible that it's trading improved runtime performance on ever-more complex hardware for longer compile times.
I see that gnu89 is the default, to avoid compatibility problems with older code, but does that mean there is no code in the kernel which requires gnu99? I saw some work was done to make the whole kernel gnu11 compatible as well.
I still remember doing a make world on FreeBSD on a 75MHz Pentium. Went out for the night and it was still going when the hangover wore off the following afternoon. This was mainly because it spent most of it's time swapping.
