Frigate Camera Calculator
How many cameras will that detector actually keep up with? Pick the hardware, set your camera count and detect frame rate, and see the arithmetic Frigate itself documents.
Frigate's default detect rate is 5 frames per second per camera. Raising it multiplies detector load directly.
Cameras this detector handles
10
at 5 fps each, keeping 20% headroom
Comfortable. This detector has room to spare.
The arithmetic
- Inference time
- 15 ms
- Detector ceiling, 1000 divided by that
- 66.7/s
- Your 4 cameras demand
- 20/s
- Utilisation
- 30%
- This hardware runs only one detector instance, so you cannot scale it by adding a second process.
- Frigate notes the N100 can only run one detector instance.
- The detector never decodes video. Stream decoding is CPU work, so hardware acceleration still matters even with an accelerator fitted.
- Raising camera resolution or frame rate increases decode and motion-detection load independently of this figure.
The formula, and why it is not a guess
Frigate publishes inference times for specific hardware and model pairings, and it states the capacity rule directly: a detector reporting a 10 millisecond inference time tops out at a thousand divided by ten, or one hundred detections per second. Divide that by the detect frame rate each camera uses and you have the camera count. Everything above is that arithmetic on published numbers, not a rule of thumb.
The 20% headroom is our addition, and it is deliberate. Running a detector at its ceiling means any burst of activity drops frames, and dropped frames are missed events.
The detector is not the whole story
This is the part that catches people out. Frigate uses the CPU to decode camera streams and watch for motion, and only hands the interesting frames to the detector. An accelerator does nothing for decoding. Raising a camera from 1080p to 4K, or lifting its frame rate, increases CPU and motion-detection work regardless of what detector is fitted.
That is why hardware video acceleration still matters on a Frigate box even when a Coral or a Hailo module is installed, and why setting sensible resolutions on the camera itself beats fixing it in software. Ourtranscode calculator covers which chips decode which codecs in hardware.
The Coral is no longer the default answer
For years the Google Coral was the automatic recommendation. Frigate's own documentation has moved on: it no longer recommends the Coral for new installations except where power draw is critical or the hardware cannot use another accelerator. Modern Intel integrated graphics running OpenVINO, and the NPU in Core Ultra parts, are competitive or faster and need no extra module.
One more constraint worth knowing before you buy: Frigate documents that the N100 can only run a single detector instance, so you cannot scale it by adding a second process. It remains a fine small-deployment box, but that is its ceiling.
Tune before you spend
If the calculator says you are close to the limit, Frigate's advice is to fix the configuration before buying hardware. A camera pointed at a road or a tree generates constant motion, and every one of those frames consumes a detection that should have gone to your driveway. Motion masks are free and frequently reclaim more capacity than an upgrade would.
Picking the box
See our Frigate mini PC picks, theHome Assistant and Frigate combined guide, and theFrigate setup walkthrough. A camera network is also the classic case for a second network port, covered in ourdual NIC guide.
How these figures are sourced
Every inference time comes from Frigate's recommended-hardware documentation, quoted for the exact hardware and model pairing shown in the dropdown. Where Frigate publishes a range, the slower end is used, because sizing on the optimistic number is how deployments end up short. We have not benchmarked these accelerators ourselves. Real inference speed varies with model resolution, driver version and what else the machine is doing, so treat the camera count as a planning figure and confirm against the inference speed Frigate reports on your own hardware.