DIY Mini Data Center: MacBook as Thin Client + RTX 3060 Windows Laptop (GPU Workstation) + NAS + Tailscale + Moonlight

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

isowankenobi

New Member
Aug 29, 2025
3
0
1
Hello everyone,


I’m planning to build a small “headless homelab / mini data center” setup at home and would like some advice before I start wiring everything up.

Current idea / architecture

  • Backend (headless room):
    • Windows laptop with RTX 3060 → acts as GPU workstation for CUDA / ML training / rendering
    • NAS (still deciding: Synology box vs. DIY TrueNAS SCALE on an old 2007 PC case)
    • Router + UPS (2000 VA / 500 W)
  • Frontend (thin client):
    • MacBook Air / Pro (and sometimes iPad) → remote access only
  • Access stack:
    • Tailscale → devices appear as if on the same LAN, secure connectivity
    • Sunshine (Windows) + Moonlight (Mac/iPad) → low-latency desktop/graphics streaming
    • Termius + SSH (or Tailscale SSH) → terminal access
    • NAS shares (SMB/NFS) → projects/data accessible on both Mac and Windows

Essentially the MacBook becomes a pure thin client while all heavy workloads (CUDA, ML, rendering, storage) happen in the headless devices. My own mini data center in another room.


Questions for the community

  1. NAS choice
    Would you recommend going with a Synology (simplicity, power efficiency, polished UI) or reusing my old 2007 PC case and building a TrueNAS SCALE box (flexibility, Docker/Kubernetes support, expandability)?
  2. UPS runtime
    The UPS is rated 2000 VA / 500 W. With both devices connected (RTX 3060 laptop ~60–80 W under load, old PC ~120–200 W), what kind of realistic runtime should I expect?
  3. Performance & latency
    Does anyone here use a similar Moonlight + Sunshine + Tailscale stack for daily dev/ML workloads? How usable is it over LAN vs. over the internet?
  4. Data management
    Any best practices for organizing datasets and code projects across NAS so that both Mac (frontend) and Windows (GPU box) environments stay in sync smoothly?

Goal

I want a clean setup where my main device (MacBook) is just a lightweight access point while all the “heavy lifting” happens in the backend. Basically, my own small datacenter/homelab with GPU + NAS + secure remote access.


Would really appreciate feedback from anyone who has tried similar builds or sees pitfalls in my approach.


Thanks in advance!
 
READINGPILEOFBOOKS.jpg
 
I do this as my daily driver but using a slightly different tech stack.

gpu acccelerated Remote Desktop is better than merely usable but certainly you can tell it’s not the same as being in front of the physical computer. You can stream 4k YouTube over the remote session but you’ll know it is a remote session because it’ll drop a few frames and the audio won’t always be in sync with the video and the quality is less good (because you’re encoding and decoding a video that was already encoded, sort of like making a copy of a copy).

as I said I use it as my daily driver so it’s not that bad but I have every expectation that looking glass on Linux with kvm would be dramatically better.

but compared to regular RDP it’s fantastic.

I’d say it’s about as good as using a virtual machine with software accelerated graphics (ex: VMware workstation or parallels desktop).
 
I see no earthly reason to put tailscale in the middle if both the remote workstation and the MacBook are in the same house.
 
I see no earthly reason to put tailscale in the middle if both the remote workstation and the MacBook are in the same house.
Thanks a lot for sharing your experience, really appreciate the detailed perspective! In my case, the main reason I’m planning this setup is because I’ll often be physically miles away from the system — I only want to carry a MacBook with me, while offloading all the heavy workloads (engineering applications like MATLAB, AutoCAD, and also ML training) to the headless GPU box + NAS back home. The idea is to get a better experience and efficiency for those workloads without needing to lug around a heavy workstation. That’s why I’m leaning toward this architecture with Tailscale + Moonlight.
 
What you’re proposing will work. Mind you the last time I checked (which was years ago) moonlight was intentionally hamstrung by nvidia to break desktop keyboard shortcuts (like WIN+R to open the run dialog) because…I have no idea why.

for me that was a dealbreaker for a remote workstation use case because I can’t live without keyboard shortcuts.

Maybe things have changed since last I looked.

it would be worth you looking also at Parsec.
 
Thanks for pointing that out! I wasn’t aware of the keyboard shortcut limitation with Moonlight in the past. I’ll definitely take a look at Parsec as well and see how it compares. Appreciate the tip