Compatible software
The model, runtime, driver and operating system must work together. A powerful GPU can still be unsuitable for a particular workload.
GPU sharing guide · Early accessCan your idle GPU
Can your idle GPU
earn money?
Potentially—when compatible hardware completes work that a customer needs and pays for. IdleGPU is developing a way to connect that spare capacity with useful processing demand. Listing a GPU today does not activate jobs or earnings.
Register GPU provider interest →What GPU sharing means.
GPU sharing connects spare processing capacity with work that benefits from it. In IdleGPU’s proposed managed service, owners would run approved, bounded tasks and receive compensation for accepted results. This differs from giving a customer unrestricted remote access to a computer.
Our current marketplace is a discovery service: owners can list capacity, customers can describe requirements, and both can express interest. Commercial job dispatch, paid work and provider payouts remain in development. An explicitly started, unpaid research pipeline is available to connected participants.
What makes a GPU suitable?
Usable memory
The workload needs enough memory for its model, inputs and intermediate results. Published memory capacity alone is not a compatibility test.
Reliable availability
Correct output, stable execution, connectivity and completion time all matter. Work interrupted by gaming or disconnects may need to be reassigned.
IdleGPU’s GB10 preflight reports demonstrate small CUDA correctness checks. They do not yet establish AI throughput, compatibility across GPU models, or profitable commercial operation.
How earnings would be calculated.
Our proposed tiers link compensation to compatible, accepted productive work. Better hardware may qualify for more demanding or higher-rate tasks, but only measured useful performance can justify that classification. No paying demand means no paid jobs, regardless of GPU power.
A practical comparison is gross compensation minus electricity, transfer costs and other operating costs. Measure whole-machine electricity during a representative job; GPU telemetry alone is not the total electricity used at the wall.
Planning rates are scenarios, not income forecasts. Before participating, ask how rejected results, retries, minimum payout amounts and payment fees are handled. Read the proposed earning tiers and payout preview.
Control and data access.
Our design calls for explicit opt-in, pause controls, approved workloads and bounded inputs. Owners should understand the task before granting a worker access to their GPU. A capacity listing itself grants no machine access.
Initial processing trials would use controlled workers and public or synthetic inputs. Encrypting transfers or hiding identities does not by itself stop a machine owner from accessing plaintext processed on that machine. Confidential workloads require additional controls. Explore the security approach →
Questions GPU owners ask
Can I earn while gaming?
We do not promise concurrent gaming and processing. Competing workloads can change completion time and stability, so availability must be tested and controlled.
Are all GPUs supported?
No. Eligibility depends on the specific workload and validated software, memory and performance. Having a GPU does not guarantee admission.
Can I use a PlayStation or Xbox?
Console participation is not offered. It remains an isolated research track with no rollout timeline. Browser access or API exposure is not proof of useful GPU inference or earning eligibility.
Does registering start processing?
No. Early access records interest; marketplace listings support discovery. Neither starts GPU work or generates money.
Start with your hardware and availability.
Register interest or create a self-reported capacity listing. Describe the GPU model, usable memory, workload of interest and expected availability without publishing credentials or private information.