Interested in running frigate on my server as well as home assistant with voice support. Can I accelerate these by adding a Jetson nano to the network? Trying to maintain low power consumption and avoid paying a ton for a graphics card I won't use otherwise. I apologize in advance if this question is out there. Thanks!

HA VM on unriad? Or HA dockers?
*Recomend VM route due to chagnes....

Can it be done yes... I don't sere why not...

Executive Summary

• Frigate: Yes, Jetson Nano works very well for Frigate object detection via NVIDIA hardware acceleration.

• Home Assistant Voice: Partially — Jetson can help, but it’s not a turnkey offload the way Frigate is.

• Unraid integration: The Jetson will not be part of Unraid itself; it runs as a separate network node.

• Power efficiency: Jetson Nano is far more power-efficient than a desktop GPU.

• Best overall low-power option (honestly): Google Coral TPU for Frigate + CPU for HA voice.

Architecture Reality Check (Important)

A Jetson Nano:

• ❌ cannot be passed through to Unraid like a PCIe GPU

• ❌ cannot be used as a remote CUDA device transparently

• ✅ must run Frigate itself (or another inference service) and communicate over the network...

Think of it as a smart coprocessor box, not an accelerator card.

How it works

You run Frigate directly on the Jetson Nano:

• Jetson OS (JetPack / Ubuntu)

• Frigate Docker using:

  • NVIDIA NVDEC (video decode)

  • TensorRT / CUDA (object detection)

Your Unraid server:

• Hosts cameras / storage

• Runs Home Assistant

• Talks to Frigate via MQTT / RTSP / HTTP

Why this is a good idea

• Excellent inference per watt

• No $300–$600 GPU sitting mostly idle

• Nano idles very low power (~5–10W)

• Handles multiple 1080p streams comfortably (model dependent)

Practical note

You will not run Frigate as an Unraid Docker in this setup!
It runs on the Jetson, period.

---

Home Assistant + Voice (This Is the Tricky Part )

Core Home Assistant

You can run Home Assistant on Unraid either way:

Option A — HA Docker (Recommended for Unraid users)

• Lightweight

• Easy backups

• Works well with Frigate over MQTT

Option B — HA VM

• Required only if you need:

  • Supervisor add-ons

  • Full HA OS experience

• More overhead, more complexity

*Docker is usually the right call unless you need add-ons.

---

Voice Acceleration Reality

Voice in HA usually involves:

• Wake word detection

• Speech-to-Text (STT)

• Text-to-Speech (TTS)

Jetson Nano:

• ❌ Not directly supported as a remote accelerator for HA voice

• ⚠️ Can run Whisper / Piper / OpenWakeWord locally — but:

  • Integration is manual

  • Networking adds latency

  • Maintenance burden increases

Most people find:

• CPU STT on a modern server is “good enough”

• Or use cloud STT selectively

• Or use ESP32-S3 voice satellites

---

Power-Efficient Alternatives (Strongly Consider These)

Google Coral TPU (USB or M.2)

Entity: Google Coral

Why it’s often better than Jetson:

• 2–3W power draw

• Directly supported by Frigate

• Plug-and-play with Unraid

• No extra node to maintain

This is the sweet spot if your primary concern is:

“I don’t want a GPU I won’t otherwise use”

Intel iGPU (if available)

If your Unraid CPU has:

• Intel iGPU + Quick Sync

You get:

• Hardware video decode

• Decent inference (OpenVINO)

• Zero extra hardware

---

Recommended Setup (Low-Power, Low-Pain)

Best balance for most people(as seen on the forum...):

Unraid
├── Home Assistant (Docker)
├── MQTT
├── Storage
└── Frigate (Docker)
     └── Google Coral TPU

If you already have a Jetson Nano:

Jetson Nano
└── Frigate (Docker, TensorRT)

Unraid
├── Home Assistant (Docker)
├── MQTT
└── Media / Storage

---

Final Verdict

• Yes, Jetson Nano works great for Frigate

• Not ideal as a general HA voice accelerator

• It will never be “part of” Unraid

• Coral TPU is usually the better low-power answer

• HA Docker on Unraid is the sane default

Edited January 24Jan 24 by bmartino1