Forgejo became a hard fork about a year ago: https://forgejo.org/2024-02-forking-forward/ And it seems that migration from Gitea is only possible up to Gitea version 1.22: https://forgejo.org/2024-12-gitea-compatibility/

What a terrible article. The margins for board partners is small, but Nvidias margin is huge.

Currently, my favorite ways of running non-Steam games are the Heroic Games Launcher and Bottles. Heroic is especially nice if you have games from GOG or EGS. However, looking at ProtonDB, it seems that both DCS and Flight Sim 2024 don’t work too well on Linux. Overall it sounds like it might be challenging for you to switch to Linux, but you can always give it a try and see how much works.

This seems to be a limitation of Intel host controllers. The USB 2.0 specification (including 12 Mbps Full Speed) allows for up to 127 devices. Each of those devices can have up to 16 IN and 16 OUT endpoints, c.f. https://www.usbmadesimple.co.uk/ums_3.htm Depending on how you count, that would be a maximum of 2k to 4k endpoints in total. I guess Intel thought it wasn’t worthwhile supporting that many endpoints.

Some quick searching turned up this post that claims that USB3 controllers often support up to 254 endpoints (in total). https://www.cambrionix.com/a-quick-guide-to-usb-endpoint-limitations/ Other posters have also said that AMD appears to have higher limits. You could also consider adding more USB root hubs to your system (with PCIe cards).

Yield is the percentage of chips that are functional. Roughly, you can think of it as the probability of a chip having 0 defects. The bigger the chip, or the higher the defect density, the lower this probability becomes. Chip designers will also include mitigation techniques (e.g. redundancy) to allow chips to work even with some defects.

Talking about the “yield” of a process doesn’t make any sense. Yield is a metric for a specific chip fabricated on a given process. This depends heavily on the size of the chip and mitigation techniques.

The “correct” metric to compare processes is defect density (in defects per square cm). Intel claims that their defect density is below 0.4 defects/cm²: https://www.tomshardware.com/tech-industry/intel-says-defect-density-at-18a-is-healthy-potential-clients-are-lining-up. This would be relatively high but not much worse than what TSMC has seen for their recent nodes: https://www.techpowerup.com/forums/threads/intel-18a-process-node-clocks-an-abysmal-10-yield-report.329513/page-2#post-5387835).

Unfotunately, I can help you with that. The machine is not running any VMs.

It’s possible, but you should be able to see it quite easily. In my case, the CPU utilization was very low, so the same test should also not be CPU-bottlenecked on your system.

I’m seeing very similar speeds on my two-HDD RAID1. The computer has an AMD 8500G CPU but the load from ZFS is minimal. Reading / writing a 50GB /dev/urandom file (larger than the cache) gives me:

• 169 MB/s write
• 254 MB/s read

What’s your setup?

With version 2.3 (currently in RC), ZFS will at least support RAIDZ expansion. That should already help a lot for a NAS usecase.

That system also sounds a lot more capable than mine. How did you end up with 25 VMs?

I’m running it in a regular mATX case (Node 804) but I think you can also get AM5 motherboards in rack-mount cases.

Perhaps my recent NAS/home server build can serve as a bit of an inspiration for you:

• AMD Ryzen 8500G (8 cores, much more powerful than your two CPUs, with iGPU)
• Standard B650 mainboard, 32 GB RAM
• 2 x used 10 TB HDDs in a ZFS pool (mainboard has 4x SATA ports)
• Debian Bookworm with Docker containers for applications (containers should be more efficient than VMs).
• Average power consumption of 19W. Usually cooled passively.

I don’t think it’s more efficient to separate processing and storage so I’d only go for that if you want to play around with a cluster. I would also avoid SD cards as a root FS, as they tend to die early and catastrophically.

It’s an Apple Silicon Mac Mini. Do you have a particular reason to think the new one is less efficient?

I do think it can achieve that while waiting for network packets (see e.g. https://www.anandtech.com/show/16252/mac-mini-apple-m1-tested).

But in terms of money savings it would rarely make sense, as you need to make it back during the time you run the system. If we assume 6 years lifetime then it would only make sense to pay $120 more. But yes, I’d also go for a system that runs regular Linux :)

I don’t have one (and I don’t want one), but Anandtech measured the M1 version at 4.2W in idle. https://www.anandtech.com/show/16252/mac-mini-apple-m1-tested I think you can also get that from other Mini PCs (e.g. NUCs).

I would disagree with idle power not being important for a home server. Most of the time, your system will be doing very little and wait for something to happen. I also don’t think a typical server has a display attached. Wolfang explains this quite well: https://youtu.be/Ppo6C_JhDHM?t=94&si=zyjEKNX8yA51uNSf

I don’t have a Mac Mini, but for always-on systems, the idle power consumption can become quite significant.

• Gaming PCs can consume up to 100W (876 kWh / year).
• My AMD B650 NAS consumes about 17W in idle (150 kWh / year).
• A NUC / Mac Mini can idle as low as 5W (44 kWh / year).

If you pay 0.30$/kWh, running your old 100W gaming PC all the time would cost you 263$ per year. My NAS is 45$ per year…

It also depends on what you need/want from the machine. The Mac Mini doesn’t have any HDDs and can’t run a regular Linux distro, for example.

It sounds like a weird idea at first, but maybe it could actually work. Kind-of like running two trains on top of each other instead of after each other. I guess the downside would be the need for bespoke rolling stock and larger platforms. I think, it would generally be preferable to double the frequency or run longer trains. But it could be interesting if you’ve already exhausted those.

The main downside of double-decker train cars is the time it takes passengers to to board them. And, since this is one of the main factors limiting metro frequencies and thus capacity, they’re not that suitable for subways. To maximize metro capacity, you want long trains with many doors and very high frequency.

Double-decker cars are much more suitable for lower-frequency service (S-Bahn, regional, long-distance,…) where they’re also commonly used.

Of course, you could still use double-decker cars in a metro (and maybe some places do), it’s just suboptimal.