Poweredge R620 Server

I received an old Dell Poweredge server from my boss at Soundnine. I explored Linux, networking, and server management over the next 2 years. System specs include 128gb of RAM, 2 intel Xeon CPUs, and an NVIDIA Tesla P4 hardware accelerator. The server runs a standard Ubuntu headless configuration.

I had originally set the server up for use as a NAS system for my music (since I collect physical media as FLAC), however I did end up doing some other things with the server. The server is connected to all of my computers using TailScale (a free private VPN service) to allow filesharing and management from any of my devices.

Server Gutshot

Game Servers

I have primarily run 2 game servers thru this machine. One of them was a minecraft server I ran for a few months, however I never really had the chance to invite friends to play a lot for me. That one was pretty easy to set up, I used a virtual network to prevent random people from joining.

The other game server I used to run was a fork of Single-Player Tarkov (SPT), which was a modding framework developed for the game Escape from Tarkov. The name of this fork was Stay in Tarkov, which has since been abandoned by the developers. The installation, as I remember, was pretty intensive, requiring docker containers and several other python packages. I played on this server for a few months with my friends in 2024, since none of them wanted to play multiplayer Tarkov.

Neural Nets/AI

When I first started Uni, I remember I discovered local LLM models somehow. I promptly installed a 9k Ollama model to mess around with. I don't think I ever got around to training the model, because of the dataset compilation. During SP26, I took a neural network class. I learned how to make classification and regression models using PyTorch. I used this server for some of that computing, as it was very resource intensive. Even if my laptop could handle it, I would have rather used my laptop for something else while the model was running. The class was mostly image and simple LLM prediction. The Tesla P4 was able to handle it.

Tesla P4