First Impressions: Lenovo T14s with Qualcomm Snapdragon ARM64 CPU

The M3 Max laptops can cross-build FreeBSD at a fraction of the time of the ThinkPad, being at around 791 seconds for `make -j17` versus the T14s being at 3210 seconds (with `make -j12`) according to the post above.

No idea about power consumptions.

It still scares me.

Do you have more details or a source on this? I'd like to learn more about the build process and timings.

Sure.

I basically did the following on trunk:

  $ CPP=/usr/bin/clang MAKEOBJDIRPREFIX=/private/var/tmp/obj ./tools/build/make.py  TARGET=arm64 TARGET_ARCH=aarch64 --host-compiler-type clang --debug -j17 --clean buildworld

You probably can follow build(5) from FreeBSD hosts instead.

NetBSD is similar, but you need to edit `tools/llvm/Makefile` and make sure that you use the following target for `support-modules` instead:

   support-modules: module-test.cpp Makefile
  -       if ${HOST_CXX} -stdlib=libc++ -c -fmodules -fcxx-modules -fmodules-cache-path=./module.cache \
  -          ${.CURDIR}/module-test.cpp  3> /dev/null 2>&1; then \
  -               echo HOST_SUPPORTS_MODULES=yes > ${.TARGET}; \
  -       else \
  -               echo HOST_SUPPORTS_MODULES=no > ${.TARGET}; \
  -       fi
  +       # Just don't use modules pre for C++20 targets. Some compilers cannot support them.
  +       echo HOST_SUPPORTS_MODULES=no > ${.TARGET};

You can further speed up NetBSD builds by editing `share/mk/bsd.sys.mk` and removing the workaround for SunPro's cc. The repeated invocation of /bin/mv for each object file really does add up.

I have not tried cross builds of OpenBSD from other operating systems.

One is $1000 and the other is around $3000. So performance per dollar looks about right.

Word of warning, I ended up getting a lot of strange compiler segfaults within xgcc when using when using `MKGCC=yes` instead `MKLLVM=yes` with NetBSD, specifically with floating point heavy code. I never did end up finding out why that happens.

Not only performance per dollar but also battery duration/consumption.

Not so on the M3 generation (nor the newer M4s). M3 Pro was particularly hobbled in this regard with only 6 performance cores vs 10 on the Max.

my m4 pro has 10 p cores as m4 max has ,get your facts right at least.

No, M3 Pro is available with 11 or 12 cores. M3 Max with 14 or 16:

https://en.wikipedia.org/wiki/Apple_M3#Variants

Similarly, the M4 Pro is available with 12 or 14 cores, the M4 Max with 14 or 16:

https://en.wikipedia.org/wiki/Apple_M4#Comparison_with_other...

You can get the M4 Max with 12 performance cores.

Comparing unbinned to unbinned, you get 10 P cores on the M4 Pro and 12 on the M4 Max. Regardless the original comment was regarding the M3 series, where there was an even larger difference between the M3 Pro and M3 Max (6 vs 10 P cores).