Anyone self hosting on Mac mini M1/M2?

https://programming.dev/post/15872064

Anyone self hosting on Mac mini M1/M2? - programming.dev

I currently have a server, a Dell T310 with an SSD in it and 12Gig of ram (weird config, I know I messed up but it works fine so I can’t be bothered to change that for now), with all my dockers running in it. It runs mostly fine, with Debian 11, a VPN so that I can block public ssh and allow it only on the VPN network, an nginx proxy to have services like a forgejo and a music library (ampache). However it can’t run a Minecraft server with more than a single person on it without stuttering ; so I was considering changing it maybe next year, after more than 3 years of services, for something beefier but also consuming less W/h (current consumption is 80W), and since I already have a Mac for work I was wondering how suitable a Mac Mini M1/M2 would be for a homelab? Does anyone have such a configuration and how does it work for you? Any hurdle that you should be aware of?

An m2 / m2 runs Debian?
InstallingDebianOn/Apple/M1 - Debian Wiki

I mean technically it does. Isn’t this the Ashai Linux and isnt it very Alpha / Beta stage? As far as I’m aware this is not recommended for anything critical ( per the project maintainers) and unless OP loves tinkering he is most likely going to hit a lot of unique edge case scenarios that require a lot of time to unwind. If you have a very narrow use case this might work but is probably not worth the headache.
Asahi is not at all “alpha” and I’d hesitate to describe even parts of it on the first and second generation Apple silicon as “beta.” Its daily driver levels of stability and I’m constantly impressed by it.

From the Ashai website - “Asahi Linux is a work in progress. Many hardware components are not yet supported!”

Just be warned a lot of people have bricked or nearly bricked their computers just trying to get Ashai installed on their Mac. Daily driver this is not but they are making great progress! It will get there eventually.

It’s a fair warning, but on my M2 MBA the only things that don’t work are the microphone and some elements of graphics acceleration. I keep macos on a tiny partition for firmware updates and, I guess, to recover in the event of a catastrophic failure, but … it’s been rock solid. Most of the software I use has compatible builds, which might be the most surprising part.

BULLSHIT

No one has hard bricked a device, you can always flash MacOS back with a tool. Any issues installing are years old. OFC it’s a work in progress, so is all of Linux even RHEL. It is 100% ready to daily drive and many people do.

All the hardware support for the Mac Mini is complete and working.

I’ve had no problems running Asahi Linux on an M1 Mac Mini.

I was thinking more like just having dockers on macOS

But running a Linux like asahi is an option

IMO venturing out in the unknown using fringe case hardware/software is a hobby by itself. It’s my 2nd hobby besides self hosting. Being more about experimenting than stability and ease of use, it’s not compatible with self hosting so I keep them separate

I host using an M1 Mac Mini using Fedora Asahi Linux. Installed easily, no problems. Fast and quiet!

I ran a Minecraft server for a while. Worked fine.

since I already have a Mac for work I was wondering how suitable a Mac Mini M1/M2 would be for a homelab?

Suitable yes, if you want to do it… maybe or maybe not. Here’s a few pointers:

  • Debian can be installed on those machines, however I’m not sure how power management works properly:
  • Installation isn’t as straightforward and as easy as in another systems because Apple decided to keep pushing the usual ARM bullshit of not including a proper UEFI with the system;
  • Some stuff will be broker, but you most likely don’t need it for self-hosting;
  • If you keep macOS around you may have good luck with virtualizing Debian using UTM or VMware. Debian’s arm64 images will run at optimal performance on that hardware.

If you’re about to spend money I would grab an HP Mini unit with a “T” CPU, those will downclock really hard and you can get a i5-10500T (on ebay) for around 250€… and everything will work fine out of the box. An i7-8500T model also sells for 150€ or something like that.

Have a look at those CPU benchmarks (last one is probably yours):

If you’re looking for power efficiency the newer CPUs are always better. Those mini units will downclock and idle at around 9-12W depending on hardware configuration but Apple should be able to do better - at least assuming you’ve power management working.

cpubenchmark.net/…/Apple-M1-8-Core-3200-MHz-vs-Ap…

InstallingDebianOn/Apple/M1 - Debian Wiki

The problem with Mac hardware is that it’s ARM and vertically integrated with everything Apple. Not all hardware is supported by Linux because Apple won’t write any linux drivers and everything is reverse engineered. You’re better off buying something non-Apple which linux properly supports.

If power consumption is an issue for you get, a R9 7950X consumes as much and at times less power than an M1/M2 (I think even M3). Check out GamerNexus’s charts. IINM AMD in W/Ghz performs better than Intel across the board.

No idea where you are, but you can get a small factor PC from one of the vendors that preload linux, or configure a small form factor PC of your liking for cheap and put linux on it. You’ll get more out of your money for the same or better performance with about the same energy consumption (or a bit more).
Somebody I know who happens to live in Hungary got himself this cheap beauty. They deliver all over Europe, but if you live elsewhere on this planet, there probably is something similar like this out there.

Anti Commercial-AI license

GN Mega Charts: CPU Power Consumption | GamersNexus

Features CPUs GN Mega Charts: CPU Power Consumption November 30, 2023 Last Updated: 2023-12-08 CPU all-core power consumption benchmarks across generations The Highlights This data has been collected from our CPU power consumption testingData here is directly comparable and can be cross-compared between charts on this pageCPUs have been split by release year and/or generation to make filtering easier. Ctrl+F for the CPU of choiceThis is a large, ongoing effort and will get updates at this URL permanentlyPlease consider supporting this effort Table of Contents AutoTOC Visit our Patreon page to contribute a few dollars toward this website's operation (or consider a direct donation or buying something from our GN Store!) Additionally, when you purchase through links to retailers on our site, we may earn a small affiliate commission. Intro This is part of our long-term reference series of component charts. We will roll these out over time. This page will be updated regularly with our latest power consumption figures for CPU testing. There will be a slight delay to incorporate the latest data following a CPU review, so the charts on this page may not be as up-to-date as our latest CPU review; however, it will contain more data than is presented in the reviews. If we were to make a software analogy, think of this page as the “LTS” (or long-term support) version of the charts: They won’t be updated as often, but will contain the most stable set of data.  Credits Test Lead, Testing, Writing Steve Burke Testing Patrick LathanMike Gaglione CPU Power Consumption Charts Gallery There are important notes below this; however, we want this page to be as useful as possible, so we're putting the charts right at the top in a gallery format for quick reference. Please scroll down and read the methodological details for further information, and also note that the Legacy & Miscellaneous chart has an important bit of information about BIOS controls and test parameters accompanying its section at the bottom of the page. There is also an update log at the bottom. Here's the quick-reference gallery! Click to enlarge. Left-to-right is sorted as newest-to-oldest generations of CPU. Our fully custom 3D Emblem Glasses celebrate our 15th Anniversary! We hand-assemble these on the East Coast in the US with a metal badge, strong adhesive, and high-quality pint glass. They pair excellently with our 3D 'Debug' Drink Coasters. Purchases keep us ad-free and directly support our consumer-focused reviews! CPU Power Efficiency Chart Lower is better (more efficient) The above is the current (as in 'today,' not as in P=VI) power efficiency chart. Unlike our above (and below) power consumption snapshot tests, this chart cannot be compared across multiple test generations. That's because it relies on a unit of work (a Blender render) to complete for the calculation, and that unit of work gets updated for our review cycle every 1-2 years. Everything on the chart is directly comparable, but it cannot be compared to older versions of itself. This gives us a measure for a known and fixed unit of work (the render) versus the power required to complete that unit of work. As an example: Something that's faster and draws more power could be less efficient than something that's slower, but ultra low power. Using This Page This page contains a mix of test methodology information, limitations of testing data, and the data itself. We also present links and resources for the original reviews (or related content) and to the Amazon or Newegg pages, which contain affiliate links to help support this page. Of course, we'd recommend reading the limitations and methodology section first. Once done there, we'd recommend Ctrl+F to find the CPU you care about. that will jump you to a table naming the CPUs in each chart. We intend to update this page quarterly. The results are dependent on methodology. As long as the method doesn’t change for each chart, it can be appended going forward. We suspect these long-term reference pages will get higher traffic volume than typical. The intent of these pages is to serve as a quick reference, permanent URL for performance data. Because we currently are ad-free on this website, we ask that if you find this page regularly useful, you please support our maintenance of this information. You can support us a few ways: Patreon: This is a monthly subscription of any amount to help fund our continued effortsDirect donation: This is a one-time, direct donation of any amount and will ensure we incur the lowest feesGN store purchase: This is the best way to support. You can get a PC building accessory or tool, a mouse mat, a t-shirt, glassware, or any other item you find useful while also supporting us Testing Methodology Unless stated otherwise, all power tests on this page were conducted using the same testing methods. The test benches have changed (naturally) for each CPU tested; however, because we isolate to only the EPS12V cables and test at a full and fixed load, the only meaningful impact to power consumption would come through motherboard changes. Methodologically, we control all motherboard power settings (unless otherwise stated for the DUT) to match the official guidance from AMD or Intel. That means boards which run MCE out-of-box will be tuned down to operate at the proper guidance and closer to the commonly understood “spec” (although that specific word is arguable) by the community.  We are advancing and changing our testing processes for power all the time, but for purposes of maintaining a directly comparable dataset, we maintain these tests long-term. Software Workload The test uses a full, all-core Blender rendering workload using the Cycles renderer to load the CPU to 100%. We use an “impossible to render” GN logo frame with high sample count and resolution. Timing & Turbo This set of tests includes a 5-minute wait period (under load) before taking a steady state power figure. This does a few things:  First, it ensures we’ve had some time for leakage to set in and for power to stabilize. Power will bounce more at the beginning of a workload, especially on CPUs that may have features like Precision Boost or may boost to a particular thermal ‘value’ for Tdie.  Secondly, and most importantly, it ensures we are measuring after any Turbo time limits on processors that have them. We do this to make sure we are consistent in measurement. Different tests are performed for the Turbo power consumption (this occurs within the first 45-50 seconds of boost on older Intel CPUs). For all-core, steady state testing, we measure only after the CPU has stabilized. The Turbo/Tau tests are not on this page yet, but again, that primarily affects Intel CPUs dating back a couple generations. Measurement Although we have begun the process of changing our power measurements to use an interposer, this dataset uses a current clamp to measure the current at the EPS12V rails. We then multiply that current by the known voltage going into the EPS12V rails to land on the total power consumption in Watts. This does not include total system power, and as a result, allows us to compare even relatively ‘ancient’ data since we are eliminating the impact from GPU components. The motherboard will affect the data (via VRM efficiency losses), but we’re still getting pretty close to the ‘true’ power consumption of the CPU without moving to software measurement, which introduces new issues. The interposer method further improves this, but for purposes of this long-term reference table, we will be presenting and maintaining the clamp method data. Environment Testing is conducted in a 21C environment with controls to keep it +/- 1 degree Celsius. Temperature has an impact on power and leakage. All tests since ~2020 are conducted with a 360mm Arctic Liquid Freezer II AIO, with all tests prior to that conducted on a 280mm Liquid Freezer II AIO. Prior to the existence of this Arctic solution, we used the Kraken X62 CPU coolers. In all scenarios, we ensure that the CPU does not thermal throttle. If a CPU is pushing too much power for its cooling, in years past, we’d move to a 360mm CLC from a 280mm solution. Now, all tests are done on a 360 (and that seems to prevent any throttling). For CPUs wherein a 95C target is part of the design, we let them boost within the limits of their cooling. Limitations This testing has limitations, and it’s important that you understand them. We want our audience to be aware of the value offered by the data we present and the limitations of the data. First, this page currently only presents all-core, 100% load data. That means we are not presenting, for example, the extreme power consumption that might be seen in a Prime95 workload (although Blender is still a heavy, realistic workload).  Additionally, we are not currently presenting gaming power consumption in these charts (although that is something we’ve recently added, it isn’t public yet). Gaming typically uses less power in total than production workloads due to the more variable nature of gaming impact on a CPU. This becomes more complicated if the bottleneck shifts: If a CPU is fully GPU-bound due to its ultra-high performance, it might appear to draw less power than alternatives that are fully CPU-bound. Again, that’s something we’ll soon be talking about in video format. It’ll make its way to articles later. There are realistic ability and time constraints with a small team, and as such, we focus on presenting the data we find to be the most reliable and which we can maintain with the highest quality. The current approach enables those goals. Although we collect it, we are also not currently presenting idle power consumption. This is because the idle measurements are often more complicated than stock due to Windows behaviors. We might add these in the future. CPU Power Consumption - Blender All-Core The below charts are our latest vetted, full power consumption charts. Data dates back several years. Because it is measured at the EPS12V cables and not total system draw, we are able to provide like-for-like comparisons between these CPUs (sans VRM efficiency losses). Keep in mind that if retested with the newest BIOS, it is possible some of these numbers change. To keep things legible, we will be breaking these into charts based on generations and architectures that we think people will most commonly compare. The data between all charts should be almost completely cross-comparable, so you can pull two up side-by-side if both parts you're interested in aren't in a single chart. The axis has been kept at 400W for all charts to make it easier to spot-check comparisons between charts. The one exception is for the Legacy & Misc. chart, which goes to a higher wattage. Gaming PC Power Supply Wattage - "How Many Watts?" These CPU power consumption comparison charts will help you understand the full load power requirements of AMD vs. Intel CPUs for each generation. These don't necessarily show power efficiency (that's a different test and requires a unit of work for comparison), but will at least give you a better idea for what kind of power supply you need for your gaming PC build or workstation build. As general guidance, our recommendation for how many watts your PSU needs is to spec for at least maximum power consumption (+ some overhead) to ensure you have enough capacity during peak loads. If you're wondering how big of a power supply to get, start first by looking at benchmarks for CPU and GPU power consumption, then go from there. Although it's unlikely you'd hit 100% load on both components simultaneously, that would allow prep for the worst case scenarios. AMD Ryzen 7000, Intel 13th Gen, Intel 14th Gen Power Draw Grab a LIMITED RUN GamersNexus T-Shirt in Charcoal Heather or 100% Cotton! This shirt features band-style "tour dates" on the back, detailing the most disappointing moments in PC hardware for all of 2023. They also have an explosion diagram of a RAM module on the front. Buying these directly funds our AD-FREE website reviews and our in-depth testing. Charcoal Heather is ultra comfortable and still in stock! Once these are gone, we will not be restocking them! For search reasons and Ctrl+F users, the above chart contains the following CPUs: CPUArchitectureRelease DateGN Original Review*Intel i3-13100FRaptor Lake2023Intel i3-13100F CPU ReviewIntel i5-13400FRaptor Lake2023Intel i5-13400F CPU ReviewAMD Ryzen 9 7900Zen 42023AMD R9 7900 CPU ReviewAMD Ryzen 7 7800X3DZen 42023AMD R7 7800X3D CPU ReviewAMD Ryzen 5 7600Zen 42023AMD R5 7600 CPU ReviewAMD Ryzen 7 7700Zen 42023AMD R7 7700 CPU ReviewAMD Ryzen 5 7600XZen 42022AMD R5 7600X CPU ReviewAMD Ryzen 7 7700XZen 42022AMD R7 7700X CPU Review & BenchmarksAMD Ryzen 9 7900XZen 42022AMD R9 7900X CPU ReviewAMD Ryzen 9 7900X3DZen 42023AMD R9 7900X3D CPU BenchmarksAMD Ryzen 9 7950XZen 42022AMD R9 7950X CPU ReviewAMD Ryzen 9 7950X ECO ModeZen 420227950X Eco Mode BenchmarksAMD Ryzen 9 7950X3DZen 42023R9 7950X3D CPU ReviewIntel i5-13600KRaptor Lake2022Intel i5-13600K CPU ReviewIntel i5-14600KRaptor Lake Refresh2023Intel i5-14600K CPU ReviewIntel i7-13700KRaptor Lake2022Intel i7-13700K CPU ReviewIntel i7-14700KRaptor Lake Refresh2023Intel i7-14700K CPU ReviewIntel i9-14900KRaptor Lake Refresh2023Intel i9-14900K CPU ReviewIntel i9-13900KRaptor Lake2022Intel i9-13900K CPU BenchmarksAMD Threadripper 7960XZen 42023AMD Threadripper 7970XZen 42023AMD 7970X & 7980X ReviewAMD Threadripper 7980XZen 42023AMD 7970X & 7980X Review* Note that we update our opinions of products as pricing changes (or as the products receive important updates). As such, our original review opinions may not be reflected in our more recent content. We'd recommend checking the latest videos for the most up-to-date opinions. AMD Ryzen 5000, Intel 10th, 11th, & 12th Gen Power Consumption This chart contains the following CPUs: CPUArchitectureRelease DateGN Original Review*AMD Ryzen 5 5500Zen 3 Cezanne2022AMD R5 5500 ReviewAMD Ryzen 5 5600Zen 3 Vermeer2022AMD R5 5600 ReviewAMD Ryzen 5 5600GZen 3 Cezanne2021AMD R5 5600G APU ReviewAMD Ryzen 5 5600X + OCZen 3 Vermeer2020AMD R5 5600X CPU ReviewAMD Ryzen 5 5600X3DZen 3 Vermeer2023AMD R5 5600X3D CPU ReviewAMD Ryzen 7 5700GZen 3 Cezanne2021AMD R7 5700G APU ReviewAMD Ryzen 7 5700XZen 3 Vermeer2022AMD R7 5700X CPU ReviewAMD Ryzen 7 5800Zen 3 Vermeer2021AMD R7 5800 CPU ReviewAMD Ryzen 7 5800XZen 3 Vermeer2020AMD R7 5800X CPU ReviewAMD Ryzen 7 5800X3DZen 3 Vermeer2022AMD R7 5800X3D CPU BenchmarksAMD Ryzen 9 5900X + OCZen 3 Vermeer2020AMD R9 5900X ReviewIntel i5-10400F + APEComet Lake2020Intel i5-10400 CPU ReviewIntel i5-10600K + OCComet Lake2020Intel i5-10600K CPU ReviewIntel i5-11400Rocket Lake2021Intel i5-11400 CPU ReviewIntel i5-11600K + OCRocket Lake2021Intel i5-11600K CPU ReviewIntel i5-12400Alder Lake2022Intel i5-12400 CPU BenchmarksIntel i5-12500Alder Lake2022Intel i5-12500 CPU TestIntel i5-12600KAlder Lake2021Intel i5-12600K CPU BenchmarkIntel i7-11700KRocket Lake2021Intel i7-11700K ReviewIntel i7-12700KAlder Lake2021Intel i7-12700K BenchmarksIntel i9-10850KComet Lake2020Intel i9-10850K CPU ReviewIntel i9-10900KComet Lake2020Intel i9-10900K CPU ReviewIntel i9-11900KRocket Lake2021Intel i9-11900K BenchmarksIntel i9-12900KAlder Lake2021Intel i9-12900K ReviewIntel i9-12900KSAlder Lake20225800X3D vs. 12900KS * Note that we update our opinions of products as pricing changes (or as the products receive important updates). As such, our original review opinions may not be reflected in our more recent content. We'd recommend checking the latest videos for the most up-to-date opinions. AMD Ryzen 3000, Intel 8th & 9th Gen Power Consumption Buy a GN 4-Pack of PC-themed 3D Coasters! These high-quality, durable, flexible coasters ship in a pack of 4, each with a fully custom design made by GN's team. You'll get a motherboard-themed coaster with debug display & reset buttons, a SATA SSD with to-scale connectors, RAM sticks, and a GN logo. These fund our web work! Buy here. The CPUs in the above chart include: CPUArchitectureRelease DateGN Original Review*AMD Athlon 3000GZen Raven Ridge2019AMD Athlon 3000G ReviewAMD Ryzen 3 3100Zen 2 Matisse2020AMD R3 3100 CPU ReviewAMD Ryzen 3 3300XZen 2 Matisse2020AMD R3 3300X CPU ReviewAMD Ryzen 5 3400GZen+ Picasso2019AMD Ryzen 5 3500XZen 2 Matisse2019AMD R5 3500X CPU ReviewAMD Ryzen 5 3600Zen 2 Matisse2019AMD R5 3600 CPU ReviewAMD Ryzen 5 3600XZen 2 Matisse2019AMD R5 3600X CPU ReviewAMD Ryzen 5 3600XTZen 2 Matisse2019AMD R5 3600XT CPU ReviewAMD Ryzen 7 3700XZen 2 Matisse2019AMD R7 3700X CPU ReviewAMD Ryzen 7 3800XZen 2 Matisse2019AMD Ryzen 7 3800X CPU ReviewAMD Ryzen 9 3900XZen 2 Matisse2019AMD R9 3900X ReviewAMD Ryzen 9 3900XTZen 2 Matisse2019AMD R9 3900XT ReviewAMD Ryzen 9 3950XZen 2 Matisse2019AMD R9 3950X ReviewAMD Ryzen 7 3800XTZen 2 Matisse2019AMD R7 3800XT ReviewIntel Pentium G5600Coffee Lake2018Intel Pentium G5600 ReviewIntel i3-9100FCoffee Lake Refresh2019Intel i3-9100F CPU BenchmarksIntel i5-9600KCoffee Lake Refresh2018Intel i5-9600K CPU TestsIntel i7-8086KCoffee Lake2018Intel i7-8086K Binning TestIntel i7-8700KCoffee Lake2017Intel i7-8700K CPU ReviewIntel i5-8600KCoffee Lake2017Intel i5-8600K ReviewIntel i5-8400Coffee Lake2017Intel i5-8400 ReviewIntel i7-9700KCoffee Lake Refresh2018Intel i7-9700K CPU ReviewIntel i9-9900KCoffee Lake Refresh2018Intel i9-9900K CPU Review * Note that we update our opinions of products as pricing changes (or as the products receive important updates). As such, our original review opinions may not be reflected in our more recent content. We'd recommend checking the latest videos for the most up-to-date opinions. AMD Ryzen 1000, 2000, & Intel 7th Gen Power Consumption We don't have as many power consumption numbers dating back to Ryzen 2000, 1000, or the Intel 7th Generation CPUs. We have some random/legacy results (below) that were from various revisits at points in time, but the Ryzen 1000 era and Intel 7th generation would have been when we just started collecting power numbers. CPUs in the list above include the below, plus various overclocking tests: CPUArchitectureRelease DateGN Original Review*AMD Ryzen 3 1200Zen Summit Ridge2017AMD R3 1200 ReviewAMD Ryzen 3 1200 AFZen+ Pinnacle Ridge2017AMD R3 1200 AF ReviewAMD Ryzen 7 2700Zen+ Pinnacle Ridge2018AMD R7 2700 ReviewIntel i7-7700KKaby Lake2017Intel i7-7700K ReviewAMD Ryzen 5 1600 AF (2600)Zen+ Pinnacle Ridge2019AMD R5 1600 AF CPU ReviewAMD Ryzen 5 2600Zen+ Pinnacle Ridge2018AMD R5 2600 Review * Note that we update our opinions of products as pricing changes (or as the products receive important updates). As such, our original review opinions may not be reflected in our more recent content. We'd recommend checking the latest videos for the most up-to-date opinions. Miscellaneous, Older HEDT, & Legacy CPU Power Consumption The next chart is for miscellaneous and legacy results. These are 'orphaned' tests: We ran them at one point or another, but they don't necessarily belong anywhere else. These results are much more variable based on platform than our newer data. That's because our inventoried platform options were more limited on some of these, like FX. IMPORTANT: With the oldest CPUs in this table, there is a higher likelihood than typical of a result which does not necessarily match the expectations of the time. The result would be 'accurate' in the sense that the current clamp would have been accurately used, but the BIOS settings may not match whatever the "guidance" was from manufacturers at the time, as we were not receiving guidance in the pre-2015 era for AMD or Intel. In other words, we would have exercised less control over BIOS as we did not have official guidance on the expectations (since we were not part of the reviewer circuit). The above chart contains these CPUs: CPUArchitectureRelease DateGN Original Review*AMD Phenom II X6 1090TThuban20102020 Phenom II X6 1090T RevisitAMD FX-8150Bulldozer Zambezi2011Intel i7-4770KHaswell2013AMD FX-9370Piledriver Vishera2013AMD FX-9590Piledriver Vishera2013Intel i7-4790KDevil's Canyon20142020 4790K RevisitAMD FX-8370Piledriver Vishera2014Intel i7-5775CBroadwell2015Intel E5-2678 v3 (x2)Haswell-EP2015Dual-X99 Motherboard BuildIntel i9-7980XESkylake-X2017Intel i9-7960X & 7980XE ReviewIntel i9-7960XSkylake-X2017Intel i9-7960X & 7980XE ReviewAMD Threadripper 2990WXZen+ Colfax2018Intel i9-9980XESkylake-X2018Intel i9-9980XE ReviewIntel i9-10900XCascade Lake-X2019Intel i9-10900X ReviewIntel i9-10980XECascade Lake-X2019Intel i9-10980XE ReviewAMD Threadripper 3970XZen 2 Castle Peak2019AMD Threadripper 3970X ReviewIntel Xeon W3175XSkylake-W2019Intel Xeon W-3175X CPU ReviewAMD Threadripper 3990XZen 2 Castle Peak2020AMD TR 3990X ReviewMany of these CPUs are from before our time reviewing CPUs, so our reasons for having them are generally being harvests from personal machines or from later revisits. Update Log The below contains an update log of changes to this page. The format is MM/DD/YYYY: 12/03/2023: Added table of contents feature to ease navigation11/30/2023: Created page with initial dataset following Threadripper 7000 reviews Grab a GN Soldering & Project Mat for a high-quality work surface with extreme heat resistance. These purchases directly fund our operation, including our build-out of the hemi-anechoic chamber for our acoustic testing! (or consider a direct donation or a Patreon contribution!)

There are plenty of Linux containers available for ARM in part because a lot of developers want to run Linux containers within macOS on Apple Silicon.

That has had the effect improving the experience of running Linux directly on ARM servers.

That may be, but buying a Mac Mini is like buying a device made from the ground up for Windows, where any other operating system has to reverse engineer 100% of the things to work well, or you have to emulate another OS on it (which comes with its own pitfalls), and it’s 200+€ more expensive than its nearest equivalent.

Every single company I’ve worked at which introduced Apple Silicon to its developers has had headaches with compatibility. The worst I’ve seen was it taking a developer a month to get up and running because the specific component we used didn’t have a build for the specific ARM architecture. Multipass, UTM, podman, docker desktop, all didn’t work until colima and forcing the VM to emulate x86 + forcing docker in the VM to use the x86 image worked. There was a persistent problem with disk IO since it used 9p or whatever. Installing dependencies from scratch meant waiting 30 minutes on the M2.

Why pay a premium for less compatibility and worse specs? Just get yourself something that works, which is cheaper, maybe even supports a company that invests in Linux and its ecosystem, and be able to ask an existing developer community instead of asking the subsection of linux users that run your specific app on however you’re running linux on Appe hardware.

Anti Commercial-AI license

Deed - Attribution-NonCommercial-ShareAlike 4.0 International - Creative Commons

I didn’t pay a premium, I got a great deal.

The reverse engineering work was already complete, and all the containers I needed for ARM were available.

These have great performance per watt.

That’s great if it’s your experience.

I’m just saying me and others have consistently had different experiences, and OP can get a better experience at half the price, with the same (or better) energy consumption, all while supporting the Linux ecosystem directly.

Anti Commercial-AI license

Deed - Attribution-NonCommercial-ShareAlike 4.0 International - Creative Commons

I’ve donated to marcan to work on Asahi Linux, which gets upstreamed. That’s direct.

What has better performance per watt than M1 at a better price?

PassMark CPU TDP Chart - Performance / Power of available CPUs

CPU performance / power chart for CPUs. Updated daily

Other efficiency benchmarks place Apple Silicon and AMD chips ahead of Intel chips:

cpu-monkey.com/…/cpu_benchmark-cpu_performance_pe…

That may be, but buying a Mac Mini is like buying a device made from the ground up for Windows, where any other operating system has to reverse engineer 100% of the things to work well, or you have to emulate another OS on it (which comes with its own pitfalls), and it’s 200+€ more expensive than its nearest equivalent.

Every single company I’ve worked at which introduced Apple Silicon to its developers has had headaches with compatibility. The worst I’ve seen was it taking a developer a month to get up and running because the specific component we used didn’t have a build for the specific ARM architecture. Multipass, UTM, podman, docker desktop, all didn’t work until colima and forcing the VM to emulate x86 + forcing docker in the VM to use the x86 image worked. There was a persistent problem with disk IO since it used 9p or whatever. Installing dependencies from scratch meant waiting 30 minutes on the M2.

Why pay a premium for less compatibility and worse specs? Just get yourself something that works, which is cheaper, maybe even supports a company that invests in Linux and its ecosystem, and be able to ask an existing developer community instead of asking the subsection of linux users that run your specific app on however you’re running linux on Appe hardware.

What benchmark do you have that says a R9 7950X “consumes as much and at times less power than an M1/M2”??? I mean, a) of course that’s a crazy comparison because the M1/M2 is an entry-level laptop chip and the 7950x is AMD’s flagship high performance desktop chip and b) the M1/M2’s power consumption at full load is ~20W and the 7950x idles at 20-30W.

A reasonable comparable is one of AMD’s monolithic laptop chips like the 7840U or (if you’re willing to consume more power) 7840HS.

Some actual useful charts: notebookcheck.net/Apple-M3-SoC-analyzed-Increased… www.techspot.com/review/2499-apple-m2/

Apple M3 SoC analyzed: Increased performance and improved efficiency

A Notebookcheck analysis of the performance and efficiency of the new Apple M3 SoC, Apple's first 3nm chip, in comparison with AMD, Intel and Qualcomm.

Notebookcheck
Got an M2 mini and at first had it running with MacOS and k3s (mini kubernetes). What annoyed me a bit was that if you restart you need to login to get kubernetes/docker starting up again. You could remove the login but then your setup is completely unsecured against local access. After waiting a bit asahi linux supported the M2 mini and I gave it a try. It’s a fedora linux, which I’m not super familiar with, but installing k3s was easy and I’m only really using the server via kubectl which is fine for me. It even uses less watt in this config what I remember after measuring. Restarting also starts k3s without login as with any linux. So quite happy with this config.
Asahi Linux

I have an old mac mini running nextcloud on Ubuntu, I did upgrade memory and plug in external ssd

I have a Mac mini m1 and run several containers to manage media, arrs and vpn. I also have a Debian vm for homeassitant. I had used UTM prior but now moved the vm to VMware fusion.

Main issue I run into is not enough ram as the base model only has 8gb. And I do not have redundancy optimised when I reboot. I have it locked with a password so it doesn’t auto run the containers or the vm.

I double the mac for light browsing too. I considered getting a dedicated Linux mini pc for the services next to a desktop Mac. But my usage is too sparse at both ends of the spectrum.

I don’t recommend Debian for Apple Silicon, just stick with Asahi Linux. There aren’t any big issues, except the fact that not all Docker images are built for arm64.

Everyone else has described the complications that a Mac mini would have. So why not consider something else? Lenovo, HP, and Dell make 1l ultra small form factor PCs and they’re pretty cheap on eBay. They’re also low power. Search for Tiny Mini Micro to find information.

I have three Lenovo Thinkcentre machines - two with 32gb RAM and one with 64gb RAM - running my Proxmox VE cluster. Highly recommend using those small machines instead of a Mac mini.