# Render Sampling Budget

> Compare image size, sample workload and a simple sampling-noise estimate. Plan a local Render Desk image before starting the worker. Compare the chosen sample count with a target, see pixel and ray bounds, buffer estimates and the expected relative sampling noise under independent samples.

Part of usefulweb.site (https://usefulweb.site/). Results are deterministic calculations from the inputs and the listed assumptions; the API returns exactly what the web page shows.

## Use it
- Human interface: https://kerfwiserenderbudget.usefulweb.site/ (interactive tool at https://kerfwiserenderbudget.usefulweb.site/tool)
- JSON API: POST https://kerfwiserenderbudget.usefulweb.site/api/v1/calculate with a JSON object of inputs (schema: https://kerfwiserenderbudget.usefulweb.site/api/v1/tool)
- OpenAPI: https://kerfwiserenderbudget.usefulweb.site/openapi.json
- MCP (Streamable HTTP): https://kerfwiserenderbudget.usefulweb.site/mcp - tool `kerfwise_render_budget`

## Inputs
- width (number): Width (px)
- height (number): Height (px)
- samples (number): Samples per pixel
- target (number): Comparison samples per pixel

## Method
- Pixel samples = width × height × samples. Each pixel sample traces one primary ray and at most four secondary rays in the concept renderer. These are upper bounds, not measured render times.
- Independent Monte Carlo sampling has standard error proportional to 1/√N when variance is finite. The estimate excludes bias, geometry aliasing and correlated samples; it is not a promised quality score.
- Buffer estimate reserves six floating-point RGBA accumulators and six RGBA byte outputs. The current single-pass worker allocates less; geometry, browser and export memory are additional. Local limits: 384000 pixels and 6 million pixel samples.

## Limits
- Workload and sampling-noise estimates only. Actual render time and image quality depend on the scene and device.