There is no single camera-count-to-hardware rule. Size the recording host from the aggregate incoming bit rate, retention period, concurrent viewers, export workload, and the analytics rules you enable. Validate the design with the actual camera profiles before production rollout.
| Resource | Planning guidance |
|---|---|
| CPU | Modern x64 processor. Add capacity for stream conversion, concurrent exports, and analytics. |
| Memory | Allow headroom for every active stream, live-view session, background job, and analytics model. |
| Application storage | Use an SSD for the operating system, database, and application. |
| Recording storage | Use dedicated storage with sustained write performance and enough capacity for the retention target. |
| Network | Use wired Ethernet, account for camera and viewer traffic, and keep operational headroom. |
A practical decimal estimate is GB per day ≈ Mbps × 10.8 for each continuously recorded stream. Multiply by the number of retention days and cameras, then add operating headroom. For example, one 4 Mbps stream uses about 43.2 GB per day before filesystem and operational overhead.
Do not assume every stream is always copied byte-for-byte. Kalytera can use codec copy when the input and target are compatible, but processing and export paths may transcode. Measure the streams and build in capacity rather than treating a bitrate estimate as an exact result.
Use a current browser with WebRTC and Media Source support. Browser and codec behavior changes over time, so test the intended browser, camera codec, and network path together.
Analytics execute on the recording host. CPU inference is supported; optional execution providers and compatible hardware can improve throughput when included in the deployed build. Capacity depends on model, resolution, sampling rate, and simultaneous rules, so benchmark the production workload instead of relying on a fixed cameras-per-GPU figure.
Kalytera includes adapters for the following device families and standards. Actual functions still depend on the camera model, firmware, credentials, and enabled services.