The coffee window is closing. Going back many years with the time it's taken to compile the Linux kernel has been a well known opportunity for a coffee break or even eating a meal during the lengthy Linux kernel build process. Even when the Linux kernel was much smaller than it is today, with the hardware at the time it would often be a very time consuming process... Thus incredible to think that we are now on the horizon of a ten second clean, default kernel build on Linux x86_64.

Paired with very capable processors and the latest software innovations in the Linux kernel, building the latest Linux kernel on available hardware can be done in now roughly 15 seconds. Though with next-gen processors I am betting on 10 seconds. Phenomenal for being able to iterate faster from clean builds, if bisecting Linux kernel regressions, etc. The latest pending kernel build software improvements also help in proportionally reducing the kernel build times for those not on leading edge processors.

Absolutely mind boggling to think compiling a defconfig x86_64 Linux kernel build in a matter of seconds -- especially for those that have been in the Linux kernel space for a while. Or for an allmodconfig build with every possible kernel module, now dropping under two minutes per build for significantly speeding up kernel build times there too. Big win for increased productivity and crazy to think about considering the Linux kernel source tree as a whole is nearly 41 million lines. The time it takes to compile the Linux kernel has been one of my favorite benchmarks over the 22+ years running Phoronix and filled with excitement each time reviewing a new generation of CPUs to see the time consistently dropping even with the Linux kernel continuing to grow larger in step too.

On the software side making such speedy builds possible moving forward has been a patch series by Linux MM developer Lorenzo Stoakes for optimizing kernel build times. Lorenzo with the assistance of AI/LLMs took to addressing many bottlenecks in the Kbuild code that were reducing the parallel potential of the kernel build process. I've been covering Lorenzo's progress over the past month that began with AI Made A Lot Of "Hideous" Code But Found Major Bottlenecks For Faster Linux Compilation and has continued being revised and looking like it may land for Linux 7.4 with kernel builds ~36% faster and incremental kernel builds around 70% faster. Huge relative time savings and not being dependent upon just the newest or highest-end processors out there.

Lorenzo this week posted his v4 patch series of these kbuild improvements. Currently at 22 patches with some patches being picked up so far in kbuild-for-next and there also being a few other patch series noted in that cover letter worth pulling in.

That patch series noted an AMD EPYC server with GCC having seen defconfig build times drop from 28 to 20 seconds. That was with a dual AMD EPYC 9754 server... The EPYC 9754 Bergamo is a nice Zen 4 processor but my current favorite CPU for code compilation workloads and many HPC tasks is the AMD EPYC 9575F.

The EPYC 9575F with its 64 cores / 128 threads per socket allows for up to a 5.0GHz boost clock and all-core boost speed of 4.5GHz. Paired with 256MB of L3 cache, the AMD EPYC 9575F remains a very beautiful piece of silicon even two years after launch. (Coming up is the AMD EPYC 9686F high frequency Venice part that features 96 cores / 192 threads at up to 5.0GHz boost clock with 384MB of L3 cache... Paired with MRDIMM-12800 memory, I am already drooling for that one.)

At least until compilers achieve parallelizing compilation of individual source files, the very highest core counts aren't necessarily the best for build servers. In the past there has been some work on GCC parallelizing the compilation of individual files but years later nothing has crossed the finish line there.

In Linux kernel build time benchmarking, the AMD EPYC 9575F 2P consistently ranks the fastest setup for compiling the x86_64 defconfig kernel. Typically at around 20 seconds. That's already a great milestone considering and recalling the days when reviewing AMD EPYC Milan in finally breaking the 30 second mark for a kernel build.

So with two AMD EPYC 9575F processors and on an Ubuntu 26.04 LTS setup, I decided to try out Lorenzo's v4 kernel patches as outlined in this week's LKML patch series.