Chicago
Track Your Order
0
Here's how my PC's GPU helps power my Smart Home
Purpose-built smart home tech is often extraordinarily lightweight, using minimal power and requiring little in the way of performance. Not everything can run on low-power hardware, however. Sometimes, you need a little extra juice, but it might not be immediately obvious what you'd need it for.
A discrete GPU like one found in a workstation or gaming PC can be extremely valuable to a smart home setup, as its plethora of cores lends itself well to parallel processing tasks. If you want to level up your smart home setup, adding a dGPU is the way.
Local AI-powered home assistant
Powerful, extremely useful
It's one thing to use built-in, out-of-the-box solutions like Amazon's Alexa or Google Home to turn off and on your lights, but it's another thing to power your own home assistant, especially if you want to use a local LLM along with it. A GPU can come in handy here immensely, giving you the processing power you need to run something like OpenAI's Whisper, a speech recognition model.
The main reason why you'd want something like this locally is to eliminate the need to connect your devices to the Internet, especially if you want access to an LLM. Combining the ability to ask questions to a model that you run locally with smart home functionality can be a recipe to save you a lot of time and effort.
The caliber of GPU required for something like this isn't too high—anything with 8GB of VRAM or more will be sufficient to run a setup like this, but the more you have, the better the response times will be. I personally run Phi-3 Mini to handle my smart home functionality, because it's lightweight enough to run on an older GPU, like the old 3060 I had lying around.
Security cameras with object recognition
Level up your security
You can opt for one of the many paid, one-size fits all solutions for security cameras, but adding a GPU to your smart home setup allows for a bit more customizability and, of course, privacy. Many of the proprietary solutions require a paid subscription as well as connection to the Internet, but hosting your own cameras allows you to bypass the middleman, and get some really cool functionality on top of it.
Any of the options you choose will need to leverage GPU power to be effective, but there's a pretty wide range of options to choose from. You can go the full DIY route by leveraging YOLO (You Only Look Once) and OpenCV, or alternatively, Frigate, which is better for out-of-the-box privacy and smart home integrations. With these tools you can configure things like face detection, smart object tracking, and much more. Want to know when your package was delivered, and where exactly it was dropped? You can collect that information and process it completely independent of the greater Internet, all on your local machine. I haven't set this up quite yet in my own setup, but I'd opt for Frigate because of its relatively easy setup.
The downsides to GPUs in smart home setups
Power consumption and heat
One of the biggest downsides to adding a GPU to your smart home system is the power consumption. Running AI object detection on your security cameras as well as your own LLM can suck up a lot of juice, especially if you're using an older, less efficient GPU. Most GPUs one might use for such a setup would be a spare one, like your old RTX 30-series or RX 6000 series GPU. These GPUs just aren't as efficient as newer cards for AI workloads, so unless you're willing to buy a brand-new card just for your smart home, you'll be using more power, ultimately leading to a higher bill.
More power also inevitably means more heat, and that heat needs to go somewhere. If you live somewhere that gets scorching hot in the summer like I do, adding a GPU to an already hot system can be like literally adding fuel to the fire. You can mitigate this somewhat by isolating the system that powers your smart home to its own room or closet, but obviously concessions need to be made to keep it from being a literal fire hazard. In the winter, it's not as big of a deal, because the heat coming from my setup actually helps warm the house, albeit really inefficiently. In the summer it's a bit more of an annoyance, but not an unmanageable one.
It may still be cheaper than paying for a proprietary system
Self-hosting is almost always a cheaper option
Factoring in the costs of air conditioning, power consumption, and initial setup, adding a spare GPU to your smart home setup to run these types of applications will likely still be cheaper. If you were to add up the cost of having a premium AI subscription, along with a proprietary security system, you'd be well above $50 a month, and that's a conservative estimate. If you're putting in a spare GPU that was taking up space anyway, you can recoup your initial costs of things like cameras pretty quickly.
GPUs can be a powerful asset to your smart home setup
If you're okay with the added cost of power and perhaps even cooling, leveraging the power of a GPU in your smart home setup can enable some really cool functionality. Local AI-powered security cameras and home assistants can be a game-changer, and that's just the tip of the iceberg—there's so much more that the parallel processing of GPUs allows you to do in your smart home.