Connect → Ready → Start
Connect asks the browser for a compatible WebGPU adapter and creates a device. The equipment’s readiness feedback follows actual device readiness. It does not start a background job or award money.
When Ready appears, press Start. Short bounded GPU jobs run, their results are read back, and the server validates them. Rewards appear only after acceptance. Repeat connections keep this preparation brief.
Read the production state
| State | What it means |
|---|---|
| Ready | The device is prepared. No production is running; Start is available. |
| Working | Your computer is processing bounded production work. Pause is immediately available. |
| Verifying | Submitted results are awaiting server acceptance. Pending work is separate from balances. |
| Paused | No new jobs are being scheduled. A submitted result can still settle once. |
| Disconnected | The device has been released. Retained adapter details are labelled Last detected. |
| Unavailable / connection error | The browser could not provide a usable device. Read the explanation and use recovery or explore as a guest. |
What the browser can tell you
The HUD uses the best adapter name the browser exposes. A broad identifier such as “nvidia / ampere” is a valid fallback; it is not an exact retail model. Some browsers disclose very little.
Inspect shows available adapter details, session earnings, recent accepted work and measured production activity. WebGPU buffer limits are API limits, not installed VRAM. OVERCLOCK does not measure whole-device utilisation, temperatures, fan speeds, clocks or physical wattage.
A compatible non-NVIDIA adapter is not deliberately blocked. Physical verification for this build covered NVIDIA on Windows/Edge; AMD, Intel and Apple combinations remain unverified.
Low, Balanced and High
These presets change the size of bounded work batches and the rest between them. They are not percentages of your GPU’s total utilisation. OVERCLOCK changes no driver, clock, voltage or operating-system setting.
Use a lower preset if you want more room for other applications. Performance depends on the browser, adapter and other work on your computer; a higher preset is not a guaranteed proportional increase in CR.
What happens when you stop
Pause stops new scheduling. A GPU dispatch already running cannot be recalled from the hardware; work that has not been submitted is discarded when paused. A submission already in flight can still be accepted once. That final settlement does not mean production has restarted.
Disconnect releases the device. Hiding the page, changing wallet identity or network, losing the device, refreshing and reopening the browser do not silently resume production. There is no offline production income.
Accepted rates are history
CR/min and Charge/min use rewards accepted in the last 60 seconds. They include completion payments, so CR can arrive in bursts. A fresh session has little history; a displayed rate is not a promise for the next minute.
When paused or disconnected, the HUD prominently says so and labels the retained rate as history. Inspect’s graph measures accepted production activity, not whole-device GPU utilisation. Session totals describe the current client session; saved balances belong to the account.
What the validation proves
The server issues the challenge and checks its returned results. Client-reported timings, adapter names and claimed work amounts do not determine rewards. Duplicate or retried submissions cannot produce a second payment.
This proves a result matches the challenge. It does not prove ownership of a particular GPU, brand or unique physical device. A modified client can reproduce these small kernels on a CPU; OVERCLOCK is not a hardware-attested compute marketplace.