Setup, power and thermals, and software tips for running a Mac mini as a home server or self-hosting box.
A ThinkCentre M700 mini desktop is being used as a personal server at home. It runs Ubuntu 24.04.4 LTS with an Intel i3-6100 processor, built-in Intel graphics, and 8GB of memory. The machine has 10TB of storage and runs several services: Liquidsoap for streaming, WordPress, SMB file sharing, Grafana with Prometheus for server metrics, FTP, RDP, and SSH. Remote access from outside the home network is handled through Tailscale. A line UPS is connected for power backup, and its single 12V battery keeps the server running for about three hours during outages.
The option being considered is a refurbished 2018 Mac mini. It has a 3.2GHz Intel i7 processor, 32GB memory, and a 500GB SSD. The asking price is $200. The planned use is web browsing and light photo editing because the current MacBook Air is becoming too old. Installing macOS Tahoe through OpenCore is also being considered.
The main question is whether a small home server setup should watch its own network traffic. The tools being considered are Suricata, Snort, and Zeek, which can inspect network activity and help spot suspicious behavior. The current setup is a Raspberry Pi home lab using OMV and Jellyfin to manage files and media. Network monitoring is being considered as both a personal setup and a project idea. The open question is whether this kind of monitoring is useful for a home lab or too much work for a small personal server.
A locked 9U network rack was placed in a garage for a rental home, while the Wi-Fi access points around the house were connected with CAT6A cables and PoE. The main device is an Omada ER7212PC, which combines router, switch, and management functions in one unit. If the fiber internet connection fails, the network can switch over to Starlink as a backup connection. Remote network management matters because guest Wi-Fi passwords may need to change often. A Home Assistant Green is included for basic automation, with plans to add more later. The rack also holds a water leak sensor hub and the internet provider’s fiber modem. The rack has cooling fans with a temperature threshold, a 1U power unit with surge protection, and a UPS that currently gives about 3 hours of backup power. Dust is likely the main ongoing issue because the rack is in the garage.
A home server setup began with one Apple Time Capsule in 2008, then grew to three. By 2018, it had become a RAID storage stack connected under a Raspberry Pi 3B. In 2022, it moved to a Dell T110 server, then added an HP RAID cage powered by a makeshift ATX power supply, running 12 drives in total. Last month, adding a 13th drive overloaded the power setup and caused a chain of failures. A used HP drive rack was then added, but the Seagate SATA drives ran at 3Gbps instead of 6Gbps, and drives began failing while the storage array was being built. The setup then moved to cheaper SAS drives and two Supermicro rack servers. Each server has 32GB of memory, a 500GB NVMe drive, and Linux Mint 22.3. One now works as a 33TB data server, while the other is a 62TB media server with six empty bays left. The used SAS drives have 4,000 to 20,000 hours of use, and cost about one-twentieth of the new price, so extra storage space was included to allow for failures and local recovery.
A new house is being prepared with network cables inside the walls before move-in. The goal is to make wall plates support both 10 gig copper networking and 10 gig or faster fiber networking. The first plan was to run older OM3 LC fiber alongside CAT6A cable. Later, OM5 seemed like a better option because it can support higher bandwidth. Another strong recommendation was to use single mode fiber instead of multimode fiber. Because pulling these cables through the house will be difficult, the desired setup is one that can keep using the same cables and connectors long term, possibly up to 400Gbps.
A recent search for very cheap basic desktops for a business led to the small USFF PC market. A few years ago, HP EliteBook Mini systems were often available for about $30 to $70 with memory, a drive, a drive caddy, and a CPU already included. The power adapter was often the only missing part, so many listings were close to ready to use. Recent listings in the same general price range often lacked the drive, drive caddy, and AC adapter. A decent ready-to-run system under $90 was hard to find, even before adding the cost of the AC adapter.
The new home server setup moves to newer hardware that uses less electricity. The goal is also to run local LLMs at home. Setup is still in progress, with Terraform being prepared so the server can be managed through IaaC. k3s nested nodes are also being explored. Docker, pg, and a full set of home services are part of the planned stack.
Several virtual servers were moved from an old enterprise R620 server to a laptop. The old machine had 128GB of DDR3 memory and eight 800GB SAS drives. The new machine has an i7 processor, 64GB of DDR5 memory, and one 1TB NVMe drive. Large storage had to be moved somewhere else. Even with that tradeoff, the laptop is better in almost every other way. Old enterprise hardware can be fun to use, but the electricity cost made it hard to justify.
A personal server rack is running several computers. The main machine uses Windows 11 as a daily computer, while two other machines run Ubuntu. Pi-hole is already used to reduce ads and tracking requests at home, and OctoPrint is used to manage a 3D printer. AMP is being used so friends can play games together, and Plex/Jellyfin is planned for shared movies or media. The next goal is to find more services that make life easier for friends and family. A small LLM is also being considered so private data can stay closer to home, but its real usefulness and setup difficulty are still uncertain.
Canadian Prime Day deals showed much higher storage prices for someone trying to build or expand a home server. The Samsung 9100 Pro 4TB cost 399 Canadian dollars each in sales last year, but the current sale price was 1,200 Canadian dollars. Its listed regular price was 1,850 Canadian dollars, so even the sale price felt about three times higher than last year. Four of those drives had previously been used in a RAID 10 setup as a very fast main boot drive for a Dell Precision 7920, costing about 1,600 Canadian dollars before tax. Large hard drives were also more expensive. WD Elements 18TB external drives, once bought from Best Buy for about 750 Canadian dollars, were now close to 1,000 Canadian dollars. Those external drives had been opened up and used inside a NAS, and two spare drives had already been used, making replacements necessary.
The planned home server is not meant to transcode media. It is also not meant to run machine learning inside Immich. That makes the workload lighter than a server that converts video while people watch it or automatically analyzes photos. The available text does not include hardware specs, storage layout, operating system, or network setup.
Minimus Community Edition has opened its container registry so anyone can pull the images they need for free. The images are minimal containers, meaning they include only a small set of files needed to run software. They are also hardened, so they are designed to reduce avoidable security risks. Some are close to a distroless style, where common operating system tools are removed. These images were previously limited to enterprise customers, but they are now available to general users.
Two HP Gen8 microservers are being prepared for separate jobs: one as a NAS and the other as a backup machine. Both have been upgraded to their maximum 16GB ECC memory, had their internal batteries replaced, and received fresh thermal paste on their Xeon processors. The processors are still capable and fairly low-power for their age, but the passive cooling may cause performance to drop quickly when they get hot. The NAS is expected to run only monitoring tools and possibly Syncthing, not a large stack of extra services. Ubuntu Server is being installed temporarily on a hard drive so the BIOS, iLO, Smart Array, and networking firmware can be updated first. The machines include several enterprise features, but most of those may be avoided for security reasons. The intended final setup is FreeBSD or XigmaNAS, which is based on FreeBSD, mainly to learn ZFS and BSD systems. The storage idea is to use two 500GB drives in a mirror for personal data and two 4TB drives in striping for media.
tux2lab is an open-source tool for building a small virtual datacenter inside one Linux computer. It uses KVM to create and manage Linux virtual machines for learning, testing, and experiments. It creates a private virtual network, its own internal DNS domain, and a central infrastructure server virtual machine. That central server provides lab services such as DHCP, DNS, PXE, HTTP, NTP, and NFS. With golden images prepared for each Linux distribution, new virtual machines can be deployed in under a minute, and full PXE installs are also supported. One CLI handles the whole lifecycle: install, reimage, resize CPU, memory, and disks, add or remove disks and network cards, create snapshots, run health checks, and manage DNS. Multiple hostnames can be passed at once, so the same action can run across several virtual machines. Supported systems include Alma, Rocky, Oracle, CentOS Stream, and RHEL 10/9/8, Ubuntu LTS 26.04/24.04/22.04, and openSUSE Leap 16.0/15.6/15.5.
Running text to speech at home without a graphics card is possible, but the best choice depends on whether sound quality, response speed, or small size matters most. Three open-weight text to speech models were tested on a 4-core Intel Xeon system with 15.6 gigabytes of memory and no GPU, across 150 runs. For the best sound quality, Kokoro 82M with the ONNX build is the strongest choice. It produced audio at about 1.8 times real-time speed on this CPU, while the PyTorch build was slower at about 1.3 times real-time speed with nearly the same quality score. For voice assistants or chatbots where quick replies matter more, Supertonic 3 at the 5-step setting is practical. It ran at about 3.2 times real-time speed and scored 4.37 for quality, but its OpenRAIL-M license has commercial limits that need checking before public or business use. For the smallest setup, Inflect-Nano-v1 is very fast and light, with 4.6 million parameters and about 7.3 times real-time speed, but the voice can sound buzzy and robotic.
A lightweight control room is being built to check home server infrastructure and private services from a phone. It covers a Synology NAS, Proxmox VE machines, Ubuntu and Linux hosts, Docker containers, websites, and private services. The goal is broader than a simple uptime page, because it brings device health, private monitoring, status history, alerts, and recovery visibility into one place. Internal services can be checked without opening ports to the internet, because an outbound-only Docker Probe sends health information from inside the network. Current checks include HTTP, HTTPS, TCP, UDP, DNS, Ping, SSL certificate and domain expiry, container status timelines, CPU, memory, storage history, and overall health summaries.
There is demand for a self-hosted replacement for Google Keep. The wanted app should feel close to Google Keep and support simple notes as well as doodles or sketch notes. Two possible options are Anchor and react-glass-keep. Anchor appears to be maintained more actively, but it lacks doodle or sketch note support. react-glass-keep seems closer to the desired feature set, but it appears not to have been maintained since January. The goal is a Google Keep-like service in the same spirit as Immich replacing Google Photos for self-hosted photo storage.
A router port forwarding rule usually needs a service name, the inside device address, an inside port, an outside port, and a protocol. The main confusion is that people say to open “TCP and UDP,” but the router screen may not make it obvious where those choices go. Some routers let you choose TCP, UDP, or both in one protocol field. Others require one separate rule for TCP and another separate rule for UDP. A Mac mini server uses the same basic idea when it needs to be reachable from outside the home network.
Remux is a self-hosted media server. It can stream video from Stremio add-ons, local files, or WebDAV storage in one place. Existing Jellyfin clients can keep working as they are, including browsing, search, and playback. It is built in Rust with performance and low resource use as goals. The dashboard was mostly made with AI tools, while the core parts were mainly written by hand. The code is now public, so it can be installed, tested, or inspected.
A local marketplace deal offers 7 Beelink Mini PC SER5 Pro machines for $1,000 total. Each one has an AMD Ryzen 7 5825U processor, 16GB of DDR4 memory, and a 500GB SSD. A similar single unit appears on Newegg for $550, so the bundle price looks unusually low. The original plan was to buy one machine for $200 as a first home server, but the low bundle price raises the idea of buying all 7 and reselling the extras.
The main need is a local agent that can help with personal tasks while keeping private data under control. Privacy is the biggest concern. Free company-hosted models may still have hidden tradeoffs, even when a company says it does not train on user data. The main work is managing Obsidian notes, with some light code-related experiments. The practical goal is a tool that can handle notes and small tasks without sending sensitive personal material to an outside service.
TensorSharp now supports image editing and image generation with Qwen Image Edit 2511 models. The comparison tested TensorSharp against stable-diffusion.cpp with the same input image, prompt, resolution, step count, CFG setting, and seed. The run used CUDA, a 544x1184 image size, and 4 steps. On a server that was already running, TensorSharp finished the full warm request in 40.44 seconds, while stable-diffusion.cpp took 48.16 seconds. Per sampling step, TensorSharp took 7.57 seconds and stable-diffusion.cpp took 9.43 seconds. Text encoding was slower on TensorSharp at 7.45 seconds, compared with 4.47 seconds on stable-diffusion.cpp. Image encoding and decoding were faster on TensorSharp, at 0.54 seconds and 1.51 seconds, compared with 1.92 seconds and 2.57 seconds. TensorSharp’s first cold request took 54.11 seconds because it had extra setup work on a fresh server.
A Debian 12 server is sharing several network folders and whole drives through the official Samba package. Windows, Android, and Linux devices all connect to those shares. The basic setup works, but the first configuration was hard because the available documentation felt scattered, repetitive, and mixed with old settings. Sometimes all connections drop, and the Samba daemon has to be restarted by hand. Windows user access is also troublesome, because Samba-created users are not working as expected, so access is handled by forcing one user in the config or allowing guest access. A separate file browser and sync system are already in place, so the need is only a simple UI for managing Samba shares, not a large all-in-one server platform. The practical question is whether to keep running Samba directly on the server or move to a management UI or another tool.
An M4 Mac Mini connected to a Sony ZV-1 through Camlink is having random video dropouts during Riverside recordings. The camera still works on the local side, but Riverside treats the video as dropped and creates a new track. Audio keeps working, but co-hosts cannot see the video until the studio is left and rejoined. Replacing all cables did not reveal the cause. The issue seems more common in recordings longer than one hour, although some sessions can run for hours without trouble. Recording is usually done in Google Chrome.
On a Mac mini M4 running Sequoia, a Focusrite 18i8 3rd Gen can be connected and working while Mix Control says no hardware is connected. Music playback, WAV playback, and DAW recording still work normally. The blocked part is changing routing or input settings inside Mix Control. Restarting the Mac, trying different USB ports, and restarting related services in Activity Monitor did not fix it. Reinstalling the driver and Mix Control makes the controls work again. The lack of recent driver or Mix Control updates raises a concern that the device may no longer be actively supported, though that is not confirmed.
Sync-in 2.4 is a new release of a self-hosted platform for file storage, syncing, sharing, and collaboration. Euro-Office integration now allows collaborative document editing. The file task manager has been redesigned so running jobs are easier to track, organize, and cancel. Full-text search now works with Chinese, Japanese, and Korean. The server can now create ZIP archives directly, in addition to the existing TAR and TGZ formats. Security hardening and reliability fixes cover downloads, archive extraction, server startup, and configuration. Several important security vulnerabilities are also fixed in this release.
After moving to macOS Golden Gate, some apps on a Mac mini stopped installing or running correctly. Adobe apps and JetBrains apps had worked before the update. After the update, some apps cannot be installed, and some installed apps crash with an “app has quit unexpectedly” error. The cause is not confirmed, and no clear fix is included.
The NICGIGA S100-0800T is an unmanaged switch with eight 10 gigabit wired ports. It may share the same circuit board and main chips as the SR-ST3008P, a managed switch based on the same hardware family. The board includes a Realtek RTL9303 SoC, RAM, and the same type of flash chip. The visible difference may be the software installed on the device, not the hardware itself. The practical question is whether managed switch software can be loaded onto this model to unlock management features, and whether anyone has a known working method.
The need is a self-hosted music search app that can find individual songs. Lidarr is known for searching albums, but that does not cover this use case. For a Mac mini used as a home server, managing music may require a tool that searches at the song level, not only by album.