5 ms·
You shouldn't. A Pi (even the 8G one) is seriously under powered and lacks a hardware clock. I recently went through the exercise of setting up a Pi for a Minec
by chillfox 4y ago
You shouldn't. A Pi (even the 8G one) is seriously under powered and lacks a hardware clock. I recently went through the exercise of setting up a Pi for a Minecraft server only have to replace it with an office desktop within the week because it couldn't handle it when my friends were on. Turned out to be an expensive lesson, the Pi is a toy for solo tinkering and nothing more.
edit: I even tried overclocking the Pi and dialing down the simulation distance in Minecraft.
- codazoda 4y agoTo be fair, Minecraft is heavy. Doesn’t mean I’d spend this money on a pi, but I’m probably not the target.
- planb 4y agoWow really? That seems to be a minecraft problem. I run openhab (which is a quite resource heavy java app), home assistant, influxdb, grafana and a bunch of data loggings scripts off a 4GB Pi and for a while even had the unifi controller running on it. All the scripts still run instantly and interactive use is possible without problems. I found the main reason for high load or lags is when the SD card breaks - which happens approximately every two years under such heavy use. Yes, I know you can boot from USB, but you should have a working and tested backup strategy nevertheless and I always keep a spare one around.
- doubled112 4y agoThis argument has been going since the Pi came out. I'd agree most people shouldn't. They should just get a NUC and save themselves the headaches. If you know what you're getting into, though, and build to it's lack of strength you can go pretty far. Current inflated prices are a hard no from me. Minecraft servers are likely the heaviest of the ~15 services I run at home. You chose a service that I wouldn't have expected to work. My Pi 4 4GB runs an MPD server that outputs to a HiFiBerry DAC, Zabbix and Grafana. It's done so smoothly for about a year now. At one time it acceptably ran Nextcloud, Gitea, Matrix Synapse, Vaultwarden and a file server in systemd containers. At another time, it replaced an old AMD e350 machine I was using for backups - the Pi 4 was faster than that old desktop. The bottleneck was the CPU load when transferring data over SSH.
- LinuxBender 4y agoJust my own preference, but I was going the route of NUC and instead landed on Protectli routers. They are meant to be routers/firewalls but work as general purpose PC's and at least at the time costed less than a NUC and have more interface options. They are air cooled and generally run cool unless doing heavy workloads but not a problem in my case. I just set them next to my HEPA filter. One can chose between CoreBoot and AMI bios and many different CPU options. [1] - https://protectli.com/ https://protectli.com/
- atmosx 4y agoThese are rather expensive for SoHo. I would go with a carambola2 or another module from 8-devices running OpenWRT.
- LinuxBender 4y agoThose are also good options for routers/firewalls. I think Protectli is aiming for devices that can handle VPN encryption/decryption at a decent rate on their Core i5/i7 models. I only mention the Protectli devices because for me at least they were a decent alternative to NUC's and for general computing whereas some of the SoHo devices are less apt for general computing in my experience.
- atmosx 4y agoRight, makes sense.
- 72deluxe 4y agoI run a Pi 8GB booting from a USB3 NVMe. It runs a PiHole, plus acts as a WireGuard server so I can VPN into it permanently on my phone and actually get useful adblocking on the go. I also run a Navidrome instance on it and use Subtracks on Android so I can get all my music whilst out and about. It also runs a Mopidy server so I can play music at home over the proper speakers using Mopidy Mobile. It also acts as a DDNS informer so that I can always get home when my ISP decides to change my public IP. It also acts as a SSH gateway and runs a RDP instance on it. I did used to use it as a desktop device and a Moonlight/Steam desktop streamer but now the PC is in the same room, and a Pi400 is downstairs with the TV. That runs much cooler. All in all a great useful device and not sure how I'd live without it.
- OptionX 4y agoSo you got a Pi and dumped an CPU-intensive, famously single-threaded game server and called it a useless toy when it didn't work. It like trying to tow a trailer with a dirt bike and complaining when it doesn't move. The PI as a server shines in running low-to-medium power services and having a low entry and upkeep costs. I run a Plex server, Gitea server, samba filesharing and qbitorrent on a 4gb rpi4 and it just sit there and chugs along using like a tenth of the power a normal computer would, taking almost no space, making no noise and costing like a fifteenth something like a NUC would. And as a bonus, if it ever does go kaput I can just get another rpi4 and the entire process of setting it all up again would take me 5 minutes max.
- dublin 4y agoI call B.S.: I've used those "toys for solo tinkering" in production control, monitoring, and data collection/communications applications in industrial, energy, and manufacturing settings. Sounds like you just don't know how to set up and build non-obese software. N.B.: An 8GB rPi 4/CM4 has more memory, more storage, more compute power, and faster networking and i/o than a Cray supercomputer had in the early 90s! It's a real computer. If you can't do most of the useful jobs on the planet with that level of performance, then you just suck at software. (Yes, I know most Linux distros are Jabba-level obese these days, but still...) Full disclosure: I have a PiBox on order, and will be prototyping a prepackaged server with it, which may well turn into a product itself.
- Abishek_Muthian 4y agoARM servers on the cloud doesn't seem to be as performant as their x86 counterparts with similar specs and same workloads (I've tried AWS, Oracle, Hetzner, Scaleway- when it had one), I think it largely comes down to the software optimizations. Where as, I continue to use several ARM home servers consisting of Pi's and Jetson with smaller workloads and applications made for them.