Home Lab!
From top to bottom:
- Ryzen 9900X + RTX 5000 ADA in a 3U Sliger CX3171 case
- Raspberry Pi 5 Cluster on a Racknex SBC rack mount case
- Ubiquiti UDM Pro
- Hooked up to a Comcast Docsis 4.0 modem with 2-gig symmetric
- And a Netgear LM1200 LTE modem as WAN2 failover on Google Fi data SIM
- Ubiquiti USW-Pro-Max-24-PoE Switch
- Ubiquiti UNAS Pro
- Ubiquiti USP-PDU-Pro
The Raspberry Pi cluster is running a Docker Swarm to host various self-hosted services, including the Ghost instance powering this blog. All of these services are running on top of my NAS as a shared storage, although there are were a few quirks about the UNAS Pro which made this setup a bit jank (specifically around root squashing).
The main rig on top is both my AI server as well as a gaming rig, utilizing Parsec over a local 2.5G wired network to access it from my home office laptop. It's built with power efficiency in mind, which is why I went with the 9900X which is known for its performance per watt, as well as an RTX 5000 ADA (not to be confused with the RTX 5000 series geforce cards), which is workstation GPU that is roughly equivalent to an RTX 4080, but with a 250W TDP and 32GB of VRAM. In the real world, this is closer to an RTX 4070 Ti in gaming performance, but with double the 4080 VRAM for running larger AI models.
The system only goes up to about 350 Watts under maximum realistic load, measured directly from the PDU, which is quite crazy for how much performance it delivers.
It's sometimes scary to think about how my NAS is the single point of failure, which can lead to a loss of terabytes of critical data for my friends and family. I'm planning on setting up a backup plan through AWS S3 Deep Glacier and S3 Storage Gateway, as the Ubiquiti UNAS Pro doesn't really have great backup options considering it's still a very new product. I'm still exploring the most seamless and cost-effective NAS backup solution which can integrate nicely with the rest of my homelab.
