Best Hardware for Frigate: Mini PCs, GPUs, Coral and Storage for a Local AI NVR

Posted on by

Frigate can run on surprisingly modest hardware. The trick is matching each workload to the right hardware instead of buying the fastest computer you can afford.

Video decoding, object detection, AI enrichments and storage are different jobs. The best Frigate server handles those jobs efficiently without wasting power or money.

For many home installations, that makes a small Intel system more interesting than a large gaming PC.

My Short Answer

If I were building a new Frigate system today, I would start by looking at an Intel-based mini PC or small desktop with a supported integrated GPU, hardware video decoding and enough storage connectivity for the retention I want.

Frigate supports OpenVINO on Intel CPUs, integrated GPUs, Arc GPUs and Intel NPUs. Its current getting-started guidance says Intel integrated graphics can provide enough detector performance for many typical Frigate setups when the GPU is exposed correctly and hardware acceleration is configured.

That is why an inexpensive Intel N100- or N150-class mini PC can be much more capable for Frigate than a simple CPU benchmark suggests.

What Frigate Hardware Actually Has to Do

I find it easier to size Frigate when I separate the workload into four jobs:

  • Decode camera video without making the CPU do all the work
  • Run object detection efficiently
  • Run optional enrichments such as semantic search, face recognition and license-plate recognition
  • Write and retain recordings reliably

Those jobs do not have to land on the same hardware. The device decoding H.264 or H.265 may not be the best detector, and storage can become a bigger design constraint than compute once you retain weeks of video.

Why Intel Is Such a Good Fit for Frigate

Intel has a useful combination for Frigate: integrated graphics for video decoding and OpenVINO support for object detection and other AI workloads.

Frigate can use most Intel integrated GPUs and Arc GPUs for hardware-accelerated video decoding, with VAAPI or Quick Sync presets depending on the platform and stream. OpenVINO can then use supported Intel CPU, GPU or NPU resources for object detection.

That combination is why small Intel systems come up so often in Frigate builds. You can get low idle power, efficient video decoding and local AI without installing a large discrete GPU.

Tier 1: Reuse Hardware You Already Own

Before buying anything, I would test Frigate on hardware already in the house.

If you have a reasonably modern Intel desktop, mini PC, laptop or Unraid server with an Intel iGPU available to Docker, it may already be enough for several cameras.

The key is not to judge the machine only by CPU utilization before hardware acceleration is configured. Frigate’s documentation explicitly recommends hardware-accelerated video decoding, and the difference can be dramatic.

This is also where I would start if I were learning Frigate. Connect one camera, enable hardware acceleration, configure a proper detection stream and see what the system actually uses before buying another box.

Tier 2: Intel N100 or N150 Mini PC

For a small dedicated Frigate server, Intel N100- and N150-class mini PCs are extremely interesting. If you are shopping for one, you can compare Intel N100 mini PCs on Amazon.

Frigate’s current hardware table lists the Intel N100 at roughly 15 ms for MobileNetV2, about 30 ms for YOLOv9 small at 320 pixels and around 25 ms for YOLO-NAS 320. Frigate also notes that the N100 is limited to one detector instance. Those are reference measurements rather than a guarantee for every installation, but they show why these inexpensive chips can be useful Frigate hosts.

I would consider this class of machine for a modest home system where low power use matters and the detector can keep up with the actual workload.

What I Would Look For

  • Intel N100, N150 or newer low-power Intel processor
  • At least 8 GB of RAM; 16 GB gives more breathing room
  • NVMe storage for the operating system and Frigate database
  • Gigabit Ethernet or better
  • Good Linux support
  • Enough USB, SATA or network-storage options for recordings

I would not buy one solely because it is cheap. Cooling, Ethernet reliability, storage expansion and vendor support still matter.

Tier 3: Core i3/i5 Mini PC or Small Desktop

If I expected more cameras, heavier AI models, semantic search, face recognition or more headroom for other services, I would move up to a Core i3 or Core i5 system with modern Intel integrated graphics.

This is the tier I like when I want room to grow. The extra CPU capacity helps with the surrounding services while the iGPU can still handle video work efficiently.

A small business-class mini PC can be especially attractive because used Dell OptiPlex Micro, HP EliteDesk Mini and Lenovo Tiny systems are widely available and often have excellent Linux support.

Tier 4: Intel Arc or a Discrete GPU

A discrete GPU starts to make more sense when you want additional detector capacity, larger models or heavier AI enrichments. For a low-power Intel option, you can check current Intel Arc A310 options on Amazon.

Frigate supports Intel Arc GPUs with OpenVINO, Nvidia GPUs through its ONNX/TensorRT path and supported AMD discrete GPUs through ROCm/ONNX. Frigate’s current hardware table lists an Intel Arc A310 at roughly 5 ms for MobileNetV2 and shows strong performance across several larger model options.

I would look at a discrete GPU when:

  • You have many cameras
  • You want to run larger object-detection models
  • You are using semantic search, face recognition or license-plate recognition heavily
  • Your existing GPU is already available in the server
  • You need multiple detector instances

I would not add a large GPU to a small Frigate system just because I could. Idle power, heat and complexity all matter in a server that runs 24 hours a day.

What About Google Coral?

Google Coral has been closely associated with Frigate for years because the Edge TPU made efficient object detection practical on low-power hardware.

It can still be useful, especially in an existing installation. But it is no longer the only obvious path to efficient Frigate detection. Modern OpenVINO support on Intel hardware, plus GPU and NPU options, gives builders much more flexibility than they had a few years ago.

If I already owned a working Coral, I would keep using it unless I had a reason to change. If I were buying a completely new Frigate server, I would choose the host platform first and then decide whether I needed a separate accelerator at all.

Hailo, NPUs and the New Accelerator Options

Frigate also supports newer accelerator options, including Hailo8 and Hailo8L, while Intel Core Ultra systems can use their NPU for OpenVINO workloads.

Frigate’s documentation specifically suggests that on systems with both an Intel NPU and GPU, the NPU can handle object detection while the GPU is reserved for enrichments such as semantic search and face recognition.

That points toward a useful pattern for home AI servers: specialized accelerators doing different jobs instead of one oversized GPU doing everything.

How Much RAM Does Frigate Need?

Frigate itself usually is not the reason to put 64 GB of memory in a home NVR.

For a dedicated small system, I would personally be comfortable starting at 8 GB and would usually choose 16 GB if the cost difference is small. That is my sizing preference rather than a Frigate requirement. More memory becomes useful when Frigate shares the machine with Home Assistant, databases, containers, dashboards or other homelab services.

If the machine is also an Unraid server running a large Docker stack, memory planning should be based on the whole server rather than Frigate alone.

Storage Matters More Than Most People Expect

Compute gets most of the attention, but storage can become the bigger hardware decision.

Four cameras averaging 4 Mbps can generate roughly 173 GB per day. Thirty days of continuous recording can exceed 5 TB before snapshots, database overhead, additional cameras or higher bitrates enter the picture.

That means I would separate two questions:

  • What hardware should run Frigate?
  • Where should the recordings live?

Those do not have to be the same machine.

Local Drive

A dedicated SATA hard drive is simple and efficient for a small Frigate box. I would keep the OS, database and application data on SSD or NVMe and use a larger hard drive for recordings.

Unraid or NAS Storage

If you already have a storage server, Frigate can fit naturally into that architecture. The compute node can stay small while recordings land on larger storage infrastructure.

I would pay attention to network reliability and mount behavior, though. A camera system needs storage that stays available predictably.

Where Unraid Fits

Unraid can be an excellent Frigate host when the server already has suitable hardware and runs 24/7.

The advantages are obvious: Docker is already there, storage is already there, backups are easier to integrate and a modern Intel iGPU can often be passed through for video acceleration and OpenVINO.

The tradeoff is that Frigate becomes one workload among many. If the Unraid server is older, lacks a useful iGPU or spends a lot of power idling, a separate low-power mini PC may still be the better Frigate compute node.

My Preferred Architecture

If I were building a larger home Frigate system from scratch, I would strongly consider separating compute from bulk storage.

A small Intel machine would run Frigate, decode video and handle object detection. Recordings could stay local for a small system or move to larger storage infrastructure if retention requirements grew.

That gives me the efficiency of a purpose-built compute node without forcing all of my storage decisions into a tiny mini PC chassis.

Hardware I Would Avoid

CPU-Only Detection as the Long-Term Plan

Frigate can run detection on a CPU, but its documentation recommends hardware-accelerated detector types instead. If I were building a permanent system, I would plan around OpenVINO, an Edge TPU, a supported GPU, NPU or another accelerator.

A Huge Gaming GPU for Two Cameras

It will work, but that does not make it efficient. A Frigate server runs continuously, so idle power and heat are part of the hardware cost.

A Raspberry Pi Just Because It Is Small

Frigate supports hardware-accelerated decoding on some Raspberry Pi hardware, and there are valid small deployments. But if I were choosing a fresh general-purpose Frigate server, an inexpensive Intel mini PC gives me a much more flexible path for decoding, OpenVINO and expansion.

Buying Hardware Before Testing Camera Streams

Detection workload depends heavily on camera count, detection resolution and frame rate. I would not size the whole server around the cameras’ headline resolution numbers.

How I Would Size a System

I would not size Frigate by camera count alone. Detection resolution, detect frame rate, model choice, scene activity, codec, recording settings and AI enrichments can matter as much as the number of cameras. Frigate also recommends starting most object-detection models at 320×320 and watching the system metrics to make sure the detector keeps up.

Small, Efficient Build

For a few cameras and a straightforward object-detection workload, I would start by evaluating a low-power Intel mini PC such as an N100/N150-class system, using the Intel iGPU for hardware decoding and OpenVINO.

More Headroom

If I expected more simultaneous activity, heavier models, AI enrichments or several other services on the same host, I would move toward a newer Core i3/i5-class Intel system.

Large or AI-Heavy Build

For a larger installation or heavier AI workload, I would start considering a stronger Intel platform, Intel Arc, Nvidia, Hailo or another supported accelerator, along with more deliberate storage and network design.

The important test is not whether a machine fits an arbitrary camera-count bracket. It is whether decoding stays hardware-accelerated and the detector keeps up without skipped detections.

Don’t Forget the Cameras

The best Frigate server cannot fix a poor camera architecture. Stable PoE cameras, useful substreams, H.264 support, RTSP and ONVIF make the whole system easier to run.

I covered that side of the build in my Best Cameras for Frigate guide.

If you are still deciding whether Frigate is the right platform, start with my complete guide to Frigate and Home Assistant. I also compared Frigate vs. Blue Iris vs. UniFi Protect if you are choosing an NVR architecture.

Frequently Asked Questions

Do I need a Coral for Frigate?

No. Coral is still useful, but Frigate now supports several efficient detector options, including OpenVINO on Intel hardware, Nvidia, AMD, Hailo and other accelerators.

Is an Intel N100 good enough for Frigate?

It can be. Frigate’s current hardware documentation includes N100 inference benchmarks that make it a practical option for modest systems, particularly when Intel hardware decoding and OpenVINO are configured correctly.

Does Frigate need a GPU?

Not necessarily a discrete GPU. An Intel integrated GPU can accelerate video decoding and can also be used by OpenVINO for object detection.

How much RAM should I use?

For a small dedicated system, 8 GB can be workable. I would usually choose 16 GB for additional headroom, especially if the machine will run other services.

Can Frigate run on Unraid?

Yes. Unraid is a common Frigate platform because it already provides Docker and storage. Hardware acceleration depends on the host hardware and how the GPU or iGPU is exposed to the container.

Should recordings go on an SSD?

I would use SSD or NVMe for the operating system, Frigate configuration and database, but bulk recordings usually make more economic sense on large hard drives or existing NAS storage.

What I Would Build

For a first dedicated Frigate server, I would start with a modern Intel mini PC, 16 GB of RAM and NVMe storage, then use the Intel iGPU for hardware decoding and OpenVINO.

I would connect one camera, measure the actual load, and expand from there.

If the system grew into many cameras, heavier AI models or multiple enrichments, I would move up to a stronger Intel platform or add a supported accelerator rather than replacing the whole architecture.

Frigate does not need a giant server to be good. It needs the right hardware doing the right jobs, with enough headroom for the workload you actually run.