FoldingOS is built for dedicated Folding@home compute nodes — small, efficient machines that fold 24/7 in a closet, shelf, or compact rack. Start with hardware you already have, then scale up as your farm grows.
Dell OptiPlex Micro, HP EliteDesk / ProDesk Mini, and Lenovo ThinkCentre Tiny are the sweet spot for home labs. They are inexpensive on the used market, sip power, and stack neatly on a shelf.
Intel NUC, Beelink, Minisforum, and similar compact x86 boxes work well when you want newer silicon in a smaller footprint. Good for adding capacity without a full rack rebuild.
Used 1U/2U servers and workstation towers deliver the highest core counts. Best when PPD per dollar matters more than noise, power, or desk space.
ARM SBCs can run Folding@home but generally produce far less science per dollar than x86 nodes. Treat them as learning platforms or ultra-low-power edge nodes, not primary farm hardware.
Folding@home nodes run at full tilt for days or weeks. Prioritize sustained performance, thermals, and reliability over peak benchmark scores.
| Component | Minimum | Recommended | Notes |
|---|---|---|---|
| Architecture | x86_64 | x86_64 | Primary FoldingOS target. 64-bit Linux with modern CPU features. |
| CPU | 4 cores / 4 threads | 6+ cores / 12+ threads | More physical cores generally means more PPD on CPU work units. Avoid thermally starved ultrabook chips in enclosed spaces. |
| Memory | 4 GB | 8–16 GB | FAH and the FoldOps agent are light; extra RAM helps if the supervisor also runs on the same host. |
| Storage | 32 GB SSD | 128 GB NVMe SSD | No spinning disks required. SSD keeps installs fast and reduces failure risk on 24/7 nodes. |
| Network | 100 Mbps Ethernet | Gigabit Ethernet | Wired preferred. Stable connectivity matters for agent heartbeats and apt updates. |
| GPU | None | Optional NVIDIA | GPU work units can dramatically increase output but add heat, driver complexity, and power draw. CPU-only farms are simpler to operate. |
| Cooling | Stock cooler + open air | Cleaned fans, intake/exhaust airflow | FoldOps reports CPU and chassis temperatures — watch sustained loads above ~85°C. |
Most farms separate folding compute from the machine that runs the dashboard. FoldOps makes both roles easy to deploy.
Every machine that crunches work units. Runs FoldingOS, fah-client, and foldops-agent. Configure for headless operation — no monitor, keyboard, or desktop session needed after setup.
Examples: fah-02, fah-03, fah-04
One host in the fleet runs foldops-supervisor and the web dashboard. It can also fold — many home labs use fah-01 as both supervisor and a full compute node.
Example: fah-01
Consistent hostnames (fah-NN) and static IPs or DHCP reservations make the dashboard, alerts, and SSH access far easier as the farm grows.
Tip: label the physical machine to match its hostname
You do not need a datacenter to contribute meaningful science. A few well-chosen nodes add up quickly.
One micro desktop plus a supervisor is enough to learn the stack. Add nodes when power, cooling, and monitoring are dialed in.
Four micro PCs at ~60 W each is roughly 240 W continuous. Map outlets and use a PDU with metering if you are filling a rack.
A node that thermal-throttles earns less science and wears faster. Dust filters, fan cleaning, and a few inches of clearance matter.
Enable auto power-on in BIOS, disable sleep, and use wired Ethernet. The goal is an appliance that recovers from outages without intervention.
Track PPD, temperatures, offline nodes, and pending apt updates from one dashboard instead of SSHing into every box.
Pick one model family when possible — identical BIOS settings, spare parts, and rack rails simplify a growing farm.
Three common starting points. Adjust based on budget, noise tolerance, and available space.
fah-01 supervises; fah-02–fah-04 fold flat out