Joshua Warren is bringing Omarchy up on the M6 Mac mini: bootloader in, kernel loading in RAM

The M6 Mac mini is on its way to running Omarchy. Joshua Warren, who works on Apple Silicon support in the Omarchy M team, posted a progress report on X on October 18, 2026: “Installed a bootloader, got the USB debug console working. It runs the second-stage loader, and the Linux kernel and initramfs are loaded into RAM.” His next steps, in order: the U-Boot handoff and a Linux shell, then HDMI and an Omarchy desktop, and finally the Apple Neural Engine and the GPU.

That is fast work for a chip nobody has run Linux on before. The mini was delivered on October 16, and Warren kicked off the bring-up on October 17 with a simple goal: “Let’s see how long it takes us to get from macOS About screen to Omarchy About screen.” As with his other projects, he is putting AI agents on the job, saying they were “diving in to this M6 Mac Mini to bring it up under Omarchy Linux as quickly as possible”.

Before touching the boot chain, he ran local AI benchmarks against his M1 Max, a machine with 2.67x the GPU cores and 2.35x the memory bandwidth. In his numbers, the M6 wins prefill outright: 3.1x faster, at 1,046 versus 339 tokens per second. Decode is a different story. On Qwen3.6-35B-A3B the M6 reaches 69.4 tokens per second against 75.6 on the M1 Max, which Warren summed up as “92% of the Max’s performance with only 42.5% of the bandwidth”. A 27B dense model falls to 9.6 versus 19.5 tokens per second: “That’s the wall. The M6 reads about 154 GB/s, and dense decode uses about 90% of it.” These were taken before the bring-up started, so they say nothing yet about Omarchy on the M6.

The part Warren cares most about is the Neural Engine. On October 17 he identified the M6’s ANE as a new target called H18G: two 16-core engines whose firmware is loaded by iBoot before macOS starts. He also compiled the first H18G program on it the same day, noting that the team’s “open M1 tooling splits the binary into two clean tasks”. His omarchy-mlx project already runs MLX-compatible Vulkan and ANE backends on Apple Silicon under Linux, so a working ANE on the M6 would slot straight into an existing stack.

The M6 effort sits on top of the kernel work covered here before. “We have our own Linux kernel for Apple Silicon covering M1, M2, M3 and MacBook Neo machines now for @omarchy with external monitors, thunderbolt docks, Touch ID,” Warren wrote on October 17, adding: “M4, M5 and M6 coming soon.” That kernel is Aurora, whose development has moved to omacom/linux-aurora; its most recent tagged release, aurora-2026.10.17.20, went out on October 17. Warren’s post quoted a video from teammate Chris Kearney titled “OMARCHY M KERNEL RELEASE NOTES”, in which Kearney said “We are about to launch and the team is HYPED”.

The rest of the team is not idle either. On October 18, Naeem Malik, creator of Omarchy Mac, reported that the M3 Max GPU is “on, all 40 cores working with our aurora kernel”, and said his next step is joining Warren “to push Local AI on M3 Max”. Add the Touch ID support that landed earlier this month, and the Omarchy M lineup is filling in from both ends: older Macs getting more of their hardware back, and the newest one getting its first Linux boot.

For now, the M6 has a kernel in memory and no shell. The distance from there to an Omarchy desktop with a working Neural Engine is real, but if the past week is any guide, the next update won’t be long in coming.

Source: https://x.com/JoshuaSWarren/status/2111923039378211100