🏗️
Dedicated hardware guide

Build a mini data centre at home

Already running the agent on your PC? This guide is for anyone who wants to go further — buy dedicated hardware, build a multi-GPU rig, or set up a small cluster to maximise earnings. We cover what to buy, what it costs to run, and how long until it pays itself off.

Choose your build tier

All three tiers run the same NexusCompute agent. Pick based on your budget, available space, and earnings ambition.

Enthusiast build

Purpose-built dual-GPU rig — your main earner

~$1,500–2,500
build cost
$200–500/mo
est. earnings
ComponentExample~Cost
GPUs (×2)RTX 4080 / RTX 3090 / RTX 3090 Ti pair$800–1,400
CPURyzen 9 5900X or Intel i7-12700K$150–200
MotherboardATX with dual x16 PCIe slots$120–180
RAM32–64 GB DDR5$80–160
Storage1 TB NVMe (OS) + 2 TB secondary$80–120
PSU1000W+ 80+ Gold (dual GPU demand)$120–180
Cooling240 mm AIO for CPU + case airflow fans$60–100

Pros

  • Strong earnings — two GPUs handle more concurrent jobs
  • Purpose-built means easy to maintain and expand later
  • Still fits in a standard mid-tower or full-tower case

Watch out for

  • Higher power draw — electricity cost matters more
  • Requires a dedicated 20A circuit or confirmed home panel capacity

What will it cost to run?

Adjust the sliders to match your setup. Electricity cost is subtracted from estimated gross earnings to show your real net income.

Running-cost calculator

All calculations are estimates. Actual earnings vary with demand.

Build tier

System TDP (watts)550 W
503000
Hours online per day18 hrs
124
Electricity rate ($/kWh)$0.13
0.050.5
$39
Electricity / mo
550 W × 18 h
$200–$500
Est. gross / mo
Enthusiast tier
$161–$461
Net after elec.
gross minus electricity
7 mo
Hardware payback
at $311/mo net

Electricity formula: (watts / 1000) × hours × 30 × rate. Earnings are midpoint estimates based on current platform averages for the selected tier. Your actual results will depend on GPU demand, your idle schedule, and local electricity costs.

Practical setup checklist

1

Ventilation and cooling

GPUs under sustained load dump serious heat. A single RTX 3090 can output 350 W continuously — equivalent to leaving a halogen heater on low. Position your rig near an air return, add case fans for positive pressure airflow (fresh air in the front, hot air out the back), and keep ambient room temperature below 25 °C / 77 °F.

Garage / loft installs

Summer ambient temperatures can exceed GPU safe limits (85–95 °C junction temp). A portable AC unit pointed at the rack is cheap insurance against thermal throttling and hardware damage.
2

Electrical — check your panel first

A standard 15 A household circuit can safely run ~1,440 W continuously (80% of 1,800 W breaker capacity). A dual-GPU enthusiast build at ~550 W sits well within a single 15 A circuit. A 3–4 node rack drawing 1.8 kW needs its own 20 A circuit — budget $150–400 for an electrician to add one.

Starter
~200 W
Existing 15 A
Enthusiast
~550 W
15–20 A dedicated
Mini Rack
~1,800 W
20 A per node
3

Home network

A wired Gigabit Ethernet connection is strongly recommended — Wi-Fi introduces latency spikes that can cause jobs to time out. Use a Cat 6 cable from the node to your router or a cheap managed switch. For a multi-node rack, a Gigabit switch (TP-Link TL-SG108, ~$20) is all you need.

Upload speed matters most

AI inference jobs send large tensors back to the requester. Make sure your ISP plan includes at least 25 Mbps upload per active GPU. Check at fast.com or speedtest.net.
4

UPS — protect your hardware

A brief power cut without a UPS can corrupt your OS drive and stress GPU capacitors. A 1500 VA UPS (~$150) provides 10–15 minutes of runtime for a single-GPU rig — enough for the agent to finish its current job and shut down gracefully. For a rack, budget one UPS per 1,000 W of load.

5

Noise — planning for livability

GPU fans at full load are loud. Strategies that help: undervolting the GPU (often 10–15% less power draw with identical performance using MSI Afterburner or AMD Adrenalin), replacing stock fans with Noctua equivalents, and placing the rig inside a cabinet with acoustic foam. A single undervolted RTX 3090 in a cabinet runs at roughly the volume of a quiet desktop.

Frequently asked questions

Ready to list your hardware?

Once your rig is built and running, registering it takes 5 minutes. The NexusCompute agent handles all job routing — no configuration needed.

Also in the Home Compute guide series:

NexusCompute © 2026
TermsPrivacyAPI Docs