Hardware for 16 4k cameras

[chrisb123]

Hi,

Asking for advice on building a 16 x 4k home camera setup.
I currently have 8 Google Nest cameras I want to replace them with something like Amcrest 4k cameras on a separate POE LAN
16 is probably too many cameras, 8 is what I currently use.
Dont need much storage, a week max worth of activity. So dont need many TB of storage
For the server will a mid level mini PC 16GB 1TB 7640HS and a M2 Coral be sufficient?
How many cameras could that kind of device handle?

I currently have a proof of concept rpi3 running frigate and a single usb camera, will connect it to HA shortly

[inkeep-ai-assistant[bot]]

For a 16-camera 4K setup, the hardware you're considering should work, but there are important factors to consider(1)(2).

Object Detection Hardware

A single M.2 Coral should handle 8-16 cameras with moderate simultaneous activity(1). The Coral's performance depends on inference speed - with a typical 10ms inference time, a single Coral can process approximately 100 frames per second(2). For 8-16 cameras with occasional motion, one Coral is usually sufficient(1). However, if many cameras frequently capture activity simultaneously, you may need additional detection hardware(1).

CPU and Video Decoding

The AMD 7640HS with integrated graphics is a solid choice(2). Hardware acceleration for video decoding is highly recommended, especially for multiple 4K cameras(1). The integrated GPU can offload video decoding from the CPU, freeing up resources(1).

For AMD GPUs, you would configure hardware acceleration using VAAPI(3):

ffmpeg:
  hwaccel_args: preset-vaapi

(3)

You'll also need to set the environment variable LIBVA_DRIVER_NAME=radeonsi(3).

Storage Considerations

For storage estimation, use an IP camera storage calculator like the IPConfigure Storage Calculator(1). Storage needs depend on:

• Number of cameras
• Resolution and framerate - 4K streams consume significantly more storage than lower resolutions(1)
• Retention mode (continuous vs motion-only)(4)

SSDs are recommended for Frigate, offering high speed and responsiveness(1). Modern SSDs have significantly higher longevity than older models and will likely outlast your NVR hardware(1).

Camera Recommendations

For 4K cameras, Dahua and Hikvision are recommended, with Amcrest as a fallback (they rebrand Dahua cameras)(2). Cameras that output H.264 video offer the most compatibility(2). Note that larger sensors are better than higher resolutions, especially at night(2).

📚 Sources:

1. Planning a New Installation
2. Recommended hardware
3. Video Decoding
4. Recording

---

Have another question? Just tag @inkeep.

[chrisb123]

Object detection - may occasionally spike to a few cameras but living in a quiet street activity is relatively low. This should be fine. I don't get that much activity.

CPU video decoding - would assume that means a modern mini PC based CPU is capable of detecting motion in 16 cameras

Storage - According to that website 2TB will be fine for what I use it for

I understand the Coral TPU is discontinued.
How would a 8845HS based mini PC handle this setup?

[evanjrowley]

Eventually the M5Stack AI-8850 may be suitable for use with Frigate.

The specs indicate it could be a strong contender for hardware accellerated video and detection:

SoC – Axera AX8850

• CPU – Octa-core Cortex‑A55 processor at 1.7 GHz
• NPU – 24 TOPS @ INT8
• VPU
  • Video Encoder – 8K @ 30 fps H.264/H.265 encoding, supports scaling / cropping
  • Video Decoder – 8K @ 60 fps H.264/H.265 decoding, supports 16 channels 1080p parallel decoding, supports scaling / cropping

Typically you will be using low-resolution detect video streams the vast majoirty of the time. 16 detect steams at 1080p is overkill but can be balanced with lower resolutions and a 4K record stream no problem.

M.2 card:
https://www.cnx-software.com/2025/10/03/m5stack-llm-8850-card-an-m-2-m-key-ai-accelerator-module-based-on-axera-ax8850-24-tops-soc/

Raspberry Pi 5 HAT:
https://www.cnx-software.com/2026/01/31/24-tops-m5stack-ai-8850-llm-accelerator-m-2-kit-offers-an-alternative-to-raspberry-pi-ai-hat-2/

Previously discussed here: #20936

[chrisb123]

That module sounds interesting.
I did some more planning and purchased the following

4x8MP and 5x12MP POE cameras all with main and sub feeds, H264/5, IP67.
A mini PC with 32GB RAM 8845HS CPU 780M iGPU Dual 2.5GB NIC. It does have a spare m2 slot and oculink if needed
Switch 250W 16port 100MB POE+ with 2x2.5GB uplink

This should be enough with room to expand
I don't know if I'll get more cameras, 9 covers all the outside areas and the undercover areas.

I have frigate and HA integrated with MQTT even though Im currently using only motion detection on a rpi3 and usb camera. Its very responsive scrubbing through the footage, Im impressed such a low power device can handle it so well