It would be interesting to know what share of servers have that enabled.

I have it enabled on Debian and Ubuntu and I’m quite happy with it. But I had pretty bad experiences with reboot-less updates on other (desktop) distros.

Agreed.

You almost need an AI agent in a sandbox to download and run all the POCs to weed out the fake ones until one of them becomes well known.

Sounds like a good malware vector to target security researchers…

Yeah, assuming everyone installed the patch is quite optimistic. But censys probably also undercounts the number of fixed systems, as some Linux distros (e.g., Ubuntu) don’t change the version when shipping security fixes. I’m running the sury.org package which would report as 7.4.33 but has been patched. Still it looks like there’s tons of unpatched services.

Yes, hopefully most people had time to install the updates. And maybe the false exploit reports helped slow down malicious people too.

So it seems like it was rather easy for an IT security professional to create an exploit within 24h of the disclosure. Not sure if every script kiddy can do it, but this means that there are certainly malicious people/organizations who can.

Ultimately, it seems to me that the only real choice is to patch our (public) systems ASAP these days.

I think for me spin-down was working initially. But then I added some Docker containers and those could only be run on the HDDs, blocking spin down. With Debian I just put the containers on the smaller SSD, allowing the HDDs to spin down.

I like being able to drop in whatever disk I need to for expansion or replacement.

I didn’t move that far away (to Debian + ZFS), so that limitation still remains. Although the addition of Raid-Z expansion should alleviate things quite a bit.

My NAS/home server takes about 180 kWH/year, or 20W on average (costs me about 5 Euro/month). This feels like a lot to me and makes up a good chunk of my total energy usage.

What I did to optimize it is mostly select a “low-power” x86 platform, and limit the number of disks. I also decided not to use TrueNAS because it couldn’t let the disks from going to sleep.

The machine hosts 10 TB of usable storage (+RAID 1), syncthing, immich, file server, Influx DB, Grafana, and a few other services.

The kernel has supported it for a few years now and it’s a supported architecture for Debian.

For me, the main thing holding me back from getting a RISC-V SBC/computer is the lack of high performance processors. But they seem to be making progress there too.

They also looked at Gnome and raw DRM. As mentioned in the talk, there’s no significant difference between Gnome and KDE. Without VRR, they each loose one frame compared to the raw DRM. With VRR all three are on par.

The Gnome results are available in the repo: https://gitlab.freedesktop.org/klysik_nv/xdc_2026_latency/-/blob/main/Charts/heatmap_path_comparison.png?ref_type=heads

I don’t think people with new hardware are typically more technically competent… I think it depends more on the use case and how much money people are willing to spend. I think there are a lot of gamers that are quite new to Linux. But, e.g., right now Debian stable doesn’t support the Nvidia 50 series at all (released in January 2025).

To be honest, it sounds like you’ve tried the prime candidates. You probably know better than us which one fits you best. Ultimately, I also think that it doesn’t make that much of a difference for an advanced user. You can probably get almost any distro to do what you want.

I guess they’re very similar overall. The main advantage I see for Mint/Ubuntu over LMDE/Debian is hardware and driver support. On Ubuntu you get much more recent versions for the kernel, Mesa, Nvidia drivers, … from the HWE stack.

Mostly, I’ve only had minor issues: Old packages on Linux Mint, missing packages on Debian and Arch, system instability after updating without a reboot on many distros (NixOS, Fedora, Alma, …). But I’ve had two major issues that caused me to abandon a distro:

1. On Fedora KDE, I felt that my system was overall not as stable as other distros. Then they decided to ship (as a rolling update to a “stable” version) an experimental KDE patch that broke the lock screen for a week. The breakage itself was not so bad, but shipping a patch before upstream merged it to their nightly builds irked me quite a bit.

2. On NixOS, I was not able to get development stuff to work (I’m probably just too dumb for Nix/NixOS). First, I tried to set up a dev workflow that has quite a few external constraints (glibc version, Python version, running on other distros, …). I failed on multiple fronts and gave up on this. Then I tried to build my own KDE/Kirigami app on NixOS and also struggled a lot with it. I struggled for quite a while but gave up on it in the end. I moved to CachyOS afterwards and it worked flawlessly 🤷

To be fair, $600 for 64GB of RAM is a steal in this economy (you could probably sell the RAM alone for $800). So if you want that much RAM and are ok with the rest of the system - go for it. I mostly wanted to point out why I personally would not like such a machine (why I’m “sleeping” on it).

Mine was the equivalent of $660 at the time with 32GB of RAM (excl. HDDs). In this RAMpocalypse it would probably cost around $1k.

I am interested in trying out an Arm/RISC-V PC for a Linux server. However, what stops me from buying almost all of them is that they require you to use a dubious, poorly-maintained manufacturer fork of some Linux distro.I don’t want my hardware to become e-waste when the manufacturer looses interest in 10 months, so running an upstream distro (like Fedora/Debian/NixOS) is a hard requirement for me. It seems that they half-solved that issue for this machine, so that’s actually good progress.

For this particular machine, Jeff Geerling also reported that the idle power draw is 17W. That’s higher than my faster, fully-upgradable mATX AM5 x86 NAS…

Ok, thanks! Thaf sounds quite advanced, but I’ll have a read afterwards :)

I run it using LM Studio, which defaults to Q4 quantization, I think. I was able to put about 10 layers on the GPU with 64k token context. That put me at about 9.1 GB VRAM usage, leaving some room for Video playback xD

What version do you use and how do you run Qwen3.6? I’ve played around a bit with the Q4 version in LM studio +Zed, but I was not happy with the results. It looses track very often and often enters infinite loops or just stops…

I can run both the 27B and 35B version of Qwen3.6 on my 10GB VRAM 3080, and they run ok. You don’t need to put the entire model in VRAM, even if it’s probably beneficial.