Maximize your earnings
Once you have a device earning, the next question is: how do I earn more without buying new hardware? A few honest levers move the needle. Several others sound good but don't.
The single most important caveat: your earnings depend on actual marketplace demand, not your availability alone. Being online during low-demand windows earns nothing. The levers below help you intersect with demand.
Lever 1 — Be online when demand is high
The platform's pricing engine tracks live supply/demand per resource type. Demand is typically highest:
- Weekday business hours in major buyer time zones (US/EU office hours)
- Tax/quarter-end batch windows (financial workloads)
- ML training pushes (any time, irregular)
[VERIFY: scheduler exposes a demand-curve chart to device owners at GA — planned in web/src/pages/devices/]
If your day-job doesn't need the machine 9-to-5, that's the best window. Adjust your schedule (Stage 2 chapter 5) accordingly.
Lever 2 — Raise your effective capability score
The score is 0-1000, weighted: CPU 30%, RAM 30%, GPU 40% (desktop/laptop) — verified in backend/app/models/device/methods.py::calculate_capability_score. Two practical moves:
- Raise CPU/RAM caps (Stage 2 chapter 6). A 16-core machine capped at 25% looks like 4 cores to the scheduler. Going from 25% → 75% can triple your job-match rate.
- Free up RAM headroom. The agent measures what's available, not what's installed. Closing memory-heavy apps before your schedule window lifts your score.
There's a real ceiling — you can't out-score the laws of your hardware.
Lever 3 — Opt into GPU work
This is the single biggest lever for anyone with a recent dedicated GPU. ML inference, fine-tuning, and rendering jobs route preferentially to GPU-opt-in devices at a higher rate [VERIFY: GPU multiplier exact value at GA]. Turn it on in Devices → [device] → GPU opt-in.
Honest trade-off: GPU jobs run near-100% utilization. Expect heat, fan noise, and higher electricity draw. See chapter 4 (device health).
Lever 4 — Run multiple concurrent tasks (heavy machines only)
max_concurrent_tasks (default 1) can go to 2-4 on machines with 32+ cores and 64+ GB RAM. Each concurrent task gets a slice of your caps. Net: more $/hour, more wear.
Lever 5 — Reputation builds over time
reputation_score and trust_tier (0-3) are tracked per device. Higher trust → access to better-paying, lower-risk jobs [VERIFY: trust-tier rate-multiplier table at GA]. The fastest way to climb: don't drop tasks. Avoid mid-task shutdowns, keep your schedule consistent.
What doesn't move the needle
- Always-on, if demand isn't there. Idle online time earns nothing and wears hardware.
- Tiny CPU/RAM cap bumps (5-10%). The scheduler bands devices coarsely; sub-band changes rarely flip job matches.
- Fancy device names. Buyers never see them.
Realistic monthly numbers
Earnings depend on hardware × hours × demand. A mid-range desktop (8-core CPU, 32 GB RAM, no GPU) at 8 hours/night will earn meaningfully less than the same machine with a GPU opted-in. Watch your own first 30-day baseline before optimizing — that's the only number that matters.
What's next
← 1. Add more devices | 3. Troubleshoot common issues →
Last reviewed: 2026-05-21