Monitor device health
Honest framing up front: this isn't free money. Running sustained compute on your hardware wears it. The platform is designed to make the trade-off worthwhile — earnings outweigh wear for most modern hardware in most usage profiles — but you should monitor what's actually happening on your machine, not assume.
This chapter covers the four health indicators that matter and how to set a sensible "power budget" so the wear stays well below the earnings.
Indicator 1 — Temperature
Sustained heat is the single biggest factor in component lifespan. Healthy operating ranges (rough industry consensus):
| Component | Healthy | Caution | Reduce load |
|---|---|---|---|
| CPU under load | 60-80 °C | 80-90 °C | > 90 °C |
| GPU under load | 65-85 °C | 85-90 °C | > 90 °C |
The agent reads OS-level temperature sensors and surfaces them on Devices → [device] → Health [VERIFY: thermals tab implementation status — planned, not all platforms expose readings reliably]. If your dashboard shows a sustained > 90 °C, lower CPU/RAM caps or pause GPU opt-in.
Laptops with dust-clogged fans run hotter. Schedule a yearly cleaning if you intend to earn 24/7.
Indicator 2 — Power draw
A modern CPU+GPU at full load draws 300-500W. At US-average residential rates (~$0.16/kWh), 8 hours/day at 400W is roughly $15/month in electricity.
The dashboard's Net earnings widget subtracts an estimated electricity cost based on a power-draw measurement taken during your first few jobs [VERIFY: power-draw measurement feature ship date — implementation in progress in backend/app/services/analytics/]. Until that ships, do the math by hand: hours/day × watts × $/kWh × 30.
Net rule of thumb: if your gross earnings don't exceed 2× your estimated electricity cost, you're losing money. Pause or recalibrate caps.
Indicator 3 — Disk wear (SSD only)
SSDs have a finite write-cycle budget (TBW — terabytes written). Compute tasks write task inputs, outputs, and Docker layer cache to your work directory. Tens of GB per day on a 250 GB consumer SSD is fine; hundreds of GB per day will materially shorten lifespan.
Mitigation: run the agent's working directory on a dedicated cheap SSD, or a hard drive if you're OK with slower task setup. Set the work directory in Devices → [device] → Storage [VERIFY: work-directory relocation UI at GA].
HDDs don't have a write-cycle limit but care about mechanical hours. Modern enterprise HDDs are rated for 24/7 operation; consumer HDDs are not.
Indicator 4 — Sustained-load impact
CPUs and GPUs handle 100% load fine in short bursts. The wear comes from thermal cycling (going from cold → hot → cold over and over) and from sustained near-thermal-limit operation. Two practical implications:
- A schedule that runs the device 8 hours straight is gentler than one that turns it on/off every 15 minutes.
- A device running at 80 °C all day is gentler than one occasionally spiking to 95 °C.
The agent throttles to keep below a configurable thermal ceiling — default 85 °C [VERIFY: thermal ceiling configurable in agent settings at GA].
Set a power budget
Pick a monthly $ ceiling for electricity. Calculate hours/day × watts allowed:
$15/month budget ÷ $0.16/kWh ÷ 30 days ÷ 400W = ~8 hours/dayEnforce via the schedule (Stage 2 chapter 5), not by manually pausing.
Warning signs to act on
| Sign | Action |
|---|---|
| Fans running constantly off-schedule | Check for runaway background apps; the agent should be idle outside scheduled hours |
| Temps creeping up week-over-week | Clean dust from fans and intakes |
| Random reboots under load | Lower CPU/RAM caps; check PSU age (5+ years on a daily-driver desktop is suspect) |
| SMART warnings on SSD | Replace before it fails; move work directory |
What's next
← 3. Troubleshoot | 5. Get paid: payouts and taxes →
Last reviewed: 2026-05-21