Performance needs context¶
Earlier development measurements recorded the following render times:
| Output | Historical time | Status |
|---|---|---|
| 1,000 × 1,000 (1 MP) | approximately 2.33 s | Historical; environment not recorded |
| 2,000 × 2,000 (4 MP) | approximately 4.26 s | Historical; environment not recorded |
| 16,384 × 16,384 (268 MP) | approximately 1.6 min | Historical; environment not recorded |
These values are preserved for transparency, but they are not a current service guarantee. Renderer changes, style complexity, network latency, cache state, available CPU/GPU resources, and upstream tile servers can all dominate the result.
Publication standard for new results¶
The next benchmark run should record:
- Renderer version and source commit.
- Host CPU, memory, and graphics/software-rendering mode.
- Style and source provider.
- Cold-cache and warm-cache results separately.
- BBOX, output dimensions, DPI, and output format.
- Concurrent request count and queue state.
- At least 20 samples, reporting median and p95 duration.
- Test date and whether the endpoint was local or publicly routed.
Once those results exist, this page should replace the historical table with reproducible measurements and link to the raw benchmark artifact.
Capacity is not a latency promise¶
The current source limits width and height to 16,384 pixels. That describes an accepted dimension, not an expected response time or a production service level. Large renders consume substantial memory and processing time and should be tested against the intended style and deployment.