The OpenBSD build machine with a high number of CPU cores

This brief article provides a footnote for an actual state of using OpenBSD as build machine with a high number of CPU cores.

In order to perform a test case, the net/tdesktop-6.1.3 port was built, on account of the fact that it possesses sufficient complexity for the loading of the system over a period of minutes, as opposed to the hours required by www/chromium.

Tests were performed on an AMD EPYC 7502P 32-Core Processor (2500 MHz) with 128 GB of RAM and a single SSD, which is read as a Micron 5200 MTFD on the kernel version:

kern.version=OpenBSD 7.8-beta (GENERIC.MP) #31: Wed Sep 10 21:40:01 MDT 2025
    deraadt@amd64.openbsd.org:/usr/src/sys/arch/amd64/compile/GENERIC.MP

The execution of each test involved the commands: make clean; make configure; time make build MAKE_JOBS=X , where X denotes the number of parallel jobs. This allows the clean build time to be measured as follows:

It is apparent that utilising a greater number of jobs than 32 is impractical.

Here, it’s clear that using more than 32 jobs doesn’t make any sesnse.

Next, the same tests were performed on an AMD Ryzen 9 7950X3D 16-Core Processor (4200 MHz) with 128 GB of RAM and two NVMe in the RAID 1 (mirror) which reads as NVMe KIOXIA KCD8XRUG1 0105, on the same kernel. This machine roughly twice fast and it is measured as:

Subsequently, the same tests were performed on an AMD Ryzen 9 7950X3D 16-Core Processor (4200 MHz) with 128 GB of RAM and two NVMe in the RAID 1 (mirror), which reads as NVMe KIOXIA KCD8XRUG1 0105, on the same kernel. The machine in this case is approximately twice as fast, and this is measured as follows:

In this case, for a small number of concurrent jobs, the speed enhancement is approximately twofold. However, as the number of jobs approaches the limit, the enhancement diminishes to approximately one-third.