System Build Advice

[hinterlandgrower]

Just stating to put together a dedicated frigate build to handle 40 cameras at mostly 1080p, maybe 2-5 cameras being a bit higher rez. Wondering if I'm over looking anything or going a bit over kill.... the video storage will be on a dedicated NAS.

CPU
Intel Core Ultra 5 245K 4.2 GHz 14-Core Processor
CPU Cooler
be quiet! Pure Rock Pro 3 LX 61.8 CFM CPU Cooler
Motherboard
Asus TUF GAMING B860-PLUS WIFI ATX LGA1851 Motherboard
Memory
Crucial Pro 64 GB (2 x 32 GB) DDR5-5600 CL46 Memory
Storage
TEAMGROUP T-FORCE G50 512 GB M.2-2280 PCIe 4.0 X4 NVME Solid State Drive
Video Card
Asus PRIME GeForce RTX 5060 Ti 16 GB Video Card
Case
Montech X3 Mesh ATX Mid Tower Case
Power Supply
Thermaltake Smart BM3 850 W 80+ Bronze Certified Semi-modular ATX Power Supply
Wired Network Adapter
TP-Link TX201 2.5 Gb/s Ethernet PCIe x1 Network Adapter

[inkeep-ai-assistant[bot]]

Your build looks solid for a 40-camera Frigate installation. Here are some observations based on the planning documentation:

What looks good:

• 64GB RAM: This exceeds the recommended 16GB and provides excellent headroom for 40 cameras with enrichments(1)
• RTX 5060 Ti: Nvidia GPUs work well with Frigate using TensorRT for object detection(2)(3)
• 512GB NVMe SSD: Good for /tmp/cache and system operations(4)
• Network storage on NAS: Acceptable, though ensure you have robust Gigabit Ethernet at minimum for the recording load(1)

Considerations:

GPU Concurrent Stream Limit: For commercial installations, it's important to verify the number of supported concurrent streams on your GPU - many consumer GPUs max out at around 20 concurrent camera streams(1). With 40 cameras, you may hit this limitation. Check the RTX 5060 Ti specifications for its decode session limit.

CPU with NPU: The Intel Core Ultra 5 245K includes an NPU which can provide more efficient detection starting in Frigate 0.17+, though NPUs have seen limited success in community deployments and remain officially unsupported(2)

Detector Configuration: You'll likely want to configure multiple detector instances to handle 40 cameras efficiently(3)

Storage Calculator: Use an IP camera storage calculator like IPConfigure Storage Calculator to ensure your NAS has adequate capacity(1)

Network: The 2.5Gb/s network adapter is good - make sure your NAS connection supports this speed for the recording bandwidth(1)

Overall, this is a capable build, but verify the GPU's concurrent decode session limit for your camera count.

📚 Sources:

1. Planning a New Installation
2. Recommended hardware
3. Supported Hardware
4. Installation

---

Have another question? Just tag @inkeep.

[bdelima]

IMHO, it might overkill. Core Ultra 5 is pretty potent all on it's own, especially the series 2 with it's updated ARC iGPU (50 concurrent streams) and NPU. You may not need a dedicated GPU at all. I would also be pretty surprised if you used up all 64GB of ram, especially with mostly 1080p cameras. It never hurts to overprovision, but given the obscene cost of DDR5 ram right now, I'd start with 32GB. You can always add a dedicated GPU and more RAM later if warranted.

I think it's fine to back up to NAS, but I'd be concerned with UI performance if clips, recordings, and snapshots are all directly stored on network storage, even with a 2.5gb LAN. There is always greater latency associated associated with network storage vs local nvme storage. You also have to consider intent. Continous recording will obviously eat a ton of space, and may not be practical for local storage on say, a mini pc. Recording only detections and a few days worth of motion doesn't take up that much space and should easily fit on a 2 or 4TB nvme drive. 1080p also takes up much less space than 4K, and if your cameras support x265, space requirements will be even less.

As always, YMMV. Good luck!

[hinterlandgrower]

Thanks for taking the time. I don't have any experience with frigate so its good to here from other's experience. Yes DDR5 is insane right now, and I'm now thinking to start with 32GB and see what happens. You really think it will work with out the dedicated GPU? I thought the ultra 5 245k could only handle 20 streams?

[bdelima]

Intel does not set a hard limit on streams. 1080p is far easier on resources than 4k. So it depends. There are multiple things to consider. Stream decoding/encoding (ffmpeg) uses the iGPU. Detection can use either iGPU or the NPU. Enrichments can use either iGPU or NPU. These are all part of the core ultra package, and you have levers to shift workload around to make the most use of the platform. It also matters what kind of cameras you have and whether they have viable sub streams suitable for detection tasks. Common guidance is the sweet spot for detection is 720p. Recordings typically use full res streams and basically just pass through, and don't really affect resource usage all that much. Detection is what takes horsepower. If your cameras provide native 720p sub streams for detection, I think you'd be amazed at what this platform can handle. I use a core 5 ultra 125 with 7 detection streams ranging from 720p to 1920p. My system hums along at around 5% CPU usage. So try it and see what happens!