I usually use something like virtual core count plus half. If I'm compiling the Linux kernel my system might slow down in other areas for a bit, but it's still usable depending on the compile task. Personally, I want to pick up one of those new 64 core Ryzen cpus. Hear it can compile Unreal in like a minute. On Wed, Mar 31, 2021 at 6:33 AM Archange via arch-general < arch-general@lists.archlinux.org> wrote:
Le 31/03/2021 à 15:25, u34--- via arch-general a écrit :
I think there are no real machines with 5 cores, or 5 cpus. If dragon.archlinux.org actually has 5 cpus then do skip this message.
Actually this box does not exist anymore, but IIRC it had 32 cores.
Refering to https://wiki.archlinux.org/index.php/DeveloperWiki:PKGBUILD.com : It states it describes the build server at dragon.archlinux.org. At
https://wiki.archlinux.org/index.php/DeveloperWiki:PKGBUILD.com#Packager_&&_Makeflags
it suggests to use
MAKEFLAGS=-j5
https://wiki.archlinux.org/index.php/Makepkg#Parallel_compilation suggets to use the number of cores here. I don't know what the implications of using a wrong number in MAKEFLAGS. Or even if there are any implications at all. Still, why not state the correct number, or use $(nproc)?
Implications are only on build time. Generally it’s a good idea to use nproc+1 to avoid waiting threads, but using a lower value could be for ressources sharing. We are numerous users on the build machine, so not using all the ressources for very big projects is a good idea (even if they are very few of them that would block the machine for a long time). Also, they are not many projects than can leverage the 64 threads available.
But I actually built ParaView using -j65, and it was amazing to have it built in ~10 min.
Regards, Archange