Description
I saw an article citing a suite of benchmarks showing that FreeBSD 14.0 was ridiculously faster than FreeBSD 13.2, and it made no sense to me. The article has been picked up by just about every FreeBSD related site on the planet. I've been working extensively with both 13.2 and 14 and haven't really noticed a lot of difference. What made no sense was that 13.2 is very fast; 13 was a big jump over FreeBSD 12. So how could 14.0 be significantly faster?
I'm not set up to do any of the fancy benches you see on every website, so I decided to do something practical. Let me do a little test building a kernel from head to toe; something a lot of us do and a pretty good sized project.
My 13.2 disk and 14.0 disk are very similar in nature as I'm building appliances; I used a configuration that was exactly the same for both I ran the test on the following system:
The hard drives used were identical; Patriot Burst 240GB SSD s.
make clean && make depend;then:
FreeBSD 14.0
/usr//bin/time make -j 8
...much hullabaloo...
151.22 real 1025.83 user 32.37 sys
Ok, now I shut down and fired up my FreeBSD 13.2 disk on the same system and ran config, make clean and make depend on the same configuration.
FreeBSD 13.2
/usr//bin/time make -j 8
...much hullabaloo...
144.60 real 967.90 user 34.07 sys
So according to this, the FreeBSD 13.2 system was 7 seconds faster. Nothing to light your hair on fire over, but a difference nonetheless.
Now the benchmark I saw ran the benchmark on a 64 core Epyc CPU, so I thought maybe freeBSD 14 handled a larger number of cores substantially better. So I re-ran the test using 20 cores. Note I never like to use all of them because then other stuff in the system trying to run can compromise the results.
FreeBSD 14.0
/usr//bin/time make -j 20
...much hullabaloo...
109.48 real 1505.26 user 44.66 sys
FreeBSD 13.2
/usr//bin/time make -j 20
...much hullabaloo...
92.70 real 1442.11 user 46.25 sys
So using 20 cores, the difference was more; almost 17 seconds.
Conclusion
Anecdotally I didn't believe that FreeBSD 14 was much faster than 13.2, and a practical test on a 24 core system showed that to be true. Now there are differences between the 2 systems; different version of the CLANG compiler. FreeBSD 14 has 2026 object files and FreeBSD 13,2 has 1997. So that difference could be the equalizer in this particular un-scientific test.
My feeling is that there was something unusual about the published benchmarks on FreeBSD 14; perhaps just more efficiency with the Epyc cpu or a misconfigured 13.2 system. From my use of both systems for the past few months and these practical tests, I'd say FreeBSD 13.2 and FreeBSD 14.0 are both very fast, but FreeBSD 14 isn't must faster if at all.