Size a Frigate NVR by Bitrate, Not Megapixels

By LK Wood IV · Published 2026-09-07 · St. Louis County, MO

Reolink’s RLC-520A is a 5MP camera. Its RLC-811A is 4K — 3840×2160 against the 520A’s 2560×1920, about seventy percent more pixels. Reolink’s own spec pages give both the same mainstream default: 6144 kbps.

The shared default is a useful starting point, but it does not establish identical file sizes. Average recorded bitrate and retention determine footage capacity; resolution alone cannot tell you which drive fits.

This page is about picking the right input. For which machine to buy once you have the number, see best hardware for Frigate NVR.

Frigate copies the stream, so the camera sets the size

Frigate’s recording documentation states that recordings are “written directly from your camera stream without re-encoding.” Use the recorded streams’ average bitrate for footage estimates, then allow for audio, container overhead, other stored data and free-space reserve.

That collapses the whole storage question to one multiplication. One megabit per second of continuous recording is 86,400 megabits a day, which is 10.8 GB/day. Everything else — resolution, frame rate, codec — matters only because it moves the bitrate.

Here is the Reolink mainstream range, converted:

Bitrate ranges from the Reolink RLC-520A and RLC-811A specification pages, read 2026-09-07. Calculations assume those rates are sustained averages, continuous recording and decimal gigabytes, before overhead. Configured limits are not measurements of those averages.

The calculated footage spans a factor of eight because the assumed bitrate does. Two different resolutions can use the same storage when their actual average recorded bitrates match; identical maximum settings alone do not ensure that.

Reolink’s substreams are capped at 512 kbps against the mainstream’s 8192 kbps, per its bandwidth support article — a sixteen-fold gap that matters later.

Put your own cameras’ numbers into the Frigate NVR storage calculator. It takes a known bitrate directly.

What megapixels actually cost: the detect stream

Resolution is not free. It just does not bill you where people look.

Frigate’s camera setup guide describes the detect stream as “the only stream that Frigate will decode for processing,” and the record stream as “the resolution you wish to store for reference.” Two different streams, two different budgets. The 4K stream goes to disk untouched. The substream is what gets decoded, buffered and inspected.

You can see the split in shared memory. Frigate’s current installation calculator, checked October 16, 2026, uses (width × height × 1.5 + 270480) / 1048576 × 25 per camera, computed on the detect resolution. It also budgets room for two added cameras and 50 MB for logs and cache. The documentation labels the result MB; the divisor is 2²⁰ bytes. The table follows that calculator and rounds the combined allowance upward:

Add another 8 MB if Birdseye restream is enabled. The calculator targets 25 frames, above the 20-frame warning minimum; Frigate can retain up to 50 frames when space allows. This shared-memory budget is separate from total machine RAM. Use the live calculator for mixed resolutions or a larger frame allowance.

A higher detect resolution raises this frame-buffer budget substantially. A separate 4K record stream does not change the detect-resolution term. A 1280×720 detect substream is a starting point to test against your scene, not a guarantee that distant objects remain recognizable. Frigate’s guide asks that detect and record streams share an aspect ratio for its UI.

The detector budget is frames, not cameras

The third budget is inference, and Frigate publishes the arithmetic for it. From the recommended hardware page: “With an inference speed of 10, your Coral will top out at 1000/10=100, or 100 frames per second.”

That inverse-latency example is specifically about Coral. The same hardware documentation says a pipelined MemryX MX3’s throughput cannot be calculated that way. Nor does 100 inferences per second guarantee twenty cameras at 5 fps: Frigate’s video pipeline analyzes motion regions, so camera frames and model invocations are not necessarily one-to-one.

Frigate’s camera guide recommends starting at 5 detect fps, with higher rates for fast-moving objects when needed. Size the accelerator using its supported model, observed inference workload and performance during busy scenes. Check CPU decoding and motion processing as well; buying a faster detector does not remove those separate workloads.

Detect frame rate and scene activity affect this budget; record resolution alone is not a detector-capacity estimate.

Frigate’s object detector docs note that “when using many cameras one detector may not be enough to keep up” and that “multiple detectors can be defined assuming GPU resources are available.” If your arithmetic lands close to the ceiling, that is the escape hatch — not a bigger drive.

The fourth budget is watts

Cameras draw power, and the megapixel number predicts that no better than it predicts storage.

Reolink lists the fixed 4K RLC-811A at “<12W” on “IEEE 802.3af, 48V active.” The 4K RLC-823A PTZ is listed at “DC 12.0V⎓2A, <24W” and requires “IEEE 802.3at.” Identical resolution, identical 6144 kbps default, double the power class.

Six fixed cameras plus two PTZs is roughly 120 W of rated device draw before cable loss and headroom. Plenty of eight-port PoE switches do not supply that, because “eight PoE ports” describes connectors rather than a power budget. Add the rated watts, add the run length, add a margin, and buy on the total — the PoE budget calculator does exactly that.

Getting your real number off the camera

Reolink’s support article says to open Settings → Stream and click Maximum Bitrate. Use that configured limit as an initial planning input, then measure average recorded bitrate and storage growth for your camera and scene.

What the number means depends on the control mode. Axis’s bitrate control white paper explains that variable bitrate prioritizes image quality while allowing demand to vary with the scene. Maximum bitrate instead constrains demand by increasing compression when needed, potentially reducing image quality.

Use a maximum bitrate as a planning assumption, then measure. Axis explains that actual output can be below the limit and can temporarily overshoot it. A cap helps constrain demand, but it neither guarantees exact file sizes nor proves that busy-scene image quality is adequate. Include reserve and check storage growth over representative day and night activity.

Codec can change the bitrate needed for acceptable quality. Catellier and Pinson’s subjective study for NTIA/ITS investigates H.265/HEVC’s coding-efficiency improvement over H.264/AVC. The RLC-811A lists H.264 and H.265; the older RLC-520A lists H.264 only. That codec difference and their shared default do not establish a quality ranking between the two cameras. Compare actual footage from your scene: equal measured average bitrates give equal footage-size estimates, while quality depends on the encoder, scene and settings.

Axis makes the same point commercially with Zipstream, which it says “lowers bandwidth and storage requirements by an average of 50% or more when compared to standard compression” by adapting “based on scene motion, scene content, ambient light level, and settings.” Note what is missing from that list: resolution. Axis ships it on cameras as ordinary as the 1080p P3265-LVE.

Where a bitrate table stops working

Equal bitrate is not equal quality. In the same NTIA study, “for all 20 scenes coded with H.265/HEVC at 4 Mbps mean opinion scores span 38% of the subjective scale, which indicates the importance of scene selection.” Nearly two fifths of the quality range, at one fixed bitrate, purely from what was in front of the lens.

Equal actual average recorded bitrates give equal footage estimates, while equal maximum settings may not. A quiet hallway and a wind-blown tree line can also look very different at the same average bitrate. Record representative activity, inspect moving objects and night footage, then adjust each scene and remeasure its storage growth.

Leave capacity for overhead and changes in bitrate

Four cameras averaging 4096 kbps during continuous recording for 30 days produce about 5.31 TB of footage. At an actual average of 6144 kbps, that becomes 7.96 TB, which leaves no practical reserve on an 8TB drive.

Those figures assume the stated bitrate is sustained throughout continuous recording. A bitrate cap is not a measurement of the actual average. Keep the recording mode and retention settings fixed while measuring storage growth; Frigate’s recording settings determine which segments are retained. Filesystem space and other files also reduce the capacity available to recordings.

Here is a planning example that reserves 20% of nominal drive capacity for overhead and margin. That reserve is an assumption for this comparison, not a Frigate requirement. Divide the footage estimate by 0.8 to obtain the nominal capacity target:

An 8TB drive clears the first example’s arithmetic. A 10TB drive clears the second, but four cameras at 6144 kbps already use almost its entire planned footage allowance. A fifth camera at that bitrate needs a larger drive or a change to retention or recording mode. Do not count the full nominal label as free recording space.

The August feed also contained a model-named 8TB SkyHawk row, ST8000VX004, at $514.10 . That historical comparison does not establish today’s best buy. Choose capacity from your measured recording needs, then compare current offers for an exact model, condition and warranty. Motion-only retention can use less storage, but no fixed savings percentage or retention guarantee follows from that label.

I may earn a commission if you buy through these affiliate links. As an Amazon Associate I earn from qualifying purchases.

Those prices come from a feed deposited 2026-08-17 and drive pricing moves weekly. Open the link before you spend anything.

Size each workload before buying a bigger NVR. Measure average recorded bitrate for storage, use detect resolution for the shared-memory calculation, and check decoding and inference performance under your actual scene load. A higher-resolution record stream is not automatically a reason to replace the host, but its network and disk demands still need room. Compare hosts in best hardware for Frigate NVR and drive classes in best NAS hard drives.

The four numbers, in order

1. Average recorded bitrate per camera. Multiply by 10.8 GB per Mbps per day, then by retention; add overhead and reserve. A configured cap is an assumption until measured. → storage calculator
2. Detect substream resolution, tested against the objects you need to recognize. Run Frigate’s current shared-memory calculator on it.
3. Observed detection workload, starting with the documented 5 fps guidance. Check the specific accelerator and model during busy scenes; inverse latency is not a universal capacity rule.
4. Rated watts per camera, summed, plus cable loss and headroom. → PoE budget calculator

Keep record resolution, detect resolution and their separate resource budgets distinct.

Sources

• Frigate documentation, opened 2026-09-07: recording (streams written without re-encoding), camera setup (detect is the only decoded stream), installation (shm formula and worked example), recommended hardware (inference-speed arithmetic), planning a new installation, motion detection, object detectors
• Reolink specification pages, opened 2026-09-07: RLC-520A, RLC-811A, RLC-823A, and the bandwidth and finding your bitrate support articles
• Axis Communications white papers: bitrate control for IP video, Zipstream technology; product specifications for the AXIS P3265-LVE
• A. A. Catellier and M. H. Pinson, Characterization of the HEVC Coding Efficiency Advance Using 20 Scenes, ITU-T Rec. P.913 Compliant Subjective Methods, VQM, and PSNR, NTIA/ITS, December 2015
• Drive prices: Newegg product feed via Rakuten (MID 44583), deposited 2026-08-17, lowest in-stock USD row per listing