Intel's cheapest CPUs do what Google's discontinued Coral accelerators used to do, and they cost less

Published Jul 26, 2026, 1:30 PM EDT Ayush Pande is a PC hardware and gaming writer. When he's not working on a new article, you can find him with his head stuck inside a PC or tinkering with a server operating system. Besides computing, his interests include spending hours in long RPGs, yelling at his friends in co-op games, and practicing guitar. Pairing network video recorder setups with security cameras is a great way to bolster your home’s security. But due to the privacy implications of storing surveillance footage on cloud providers, I tend to avoid them like the plague. With most manufacturers slapping AI-powered tracking services on their products, I’ve pretty much disconnected my surveillance cameras from their proprietary clouds and switched to the free and open-source Frigate to manage my camera streams. Since Frigate supports most of the AI features like its proprietary cloud-based counterparts, I’d always wanted to build a completely local object detection pipeline for my security cameras. Google’s Coral AI accelerators used to be the no-brainer option for this project, and I’d even planned to pick up the USB variant. But between Google abandoning this accessory and Frigate 0.17 no longer recommending Coral, Intel OpenVINO (and even NPU) has become the ideal method to add AI facilities to self-hosted NVR setups. Coral isn’t quite the NVR beast it used to be in the past Too bad Google abandoned this neat AI accelerator Back in the day, Coral used to be a common method to implement AI features in Frigate who couldn’t slot full-fledged GPUs into their NVR-hosting rig. While it did offer an M.2 variant, the USB-based Coral accelerator was significantly more convenient for typical Frigate setups. For folks running Frigate from a virtual machine on a server node, passing the USB-powered Coral chip to the NVR was easier than configuring M.2 2230 A+E passthrough. But its real utility was with Raspberry Pi-based setups, as you could just slot it into any USB port and get rock-solid object detection capabilities on the pint-sized board. Factor in the low-power consumption of this combo, and you can see why many tinkerers used to opt for a Raspberry Pi + Coral Edge TPU combo. Unfortunately, Google stopped supporting the original Coral TPU lineup a while ago, and most of the libraries required by this accelerator haven’t been updated since 2022. Worse still, the GASKET driver required by the PCI-based Coral devices isn’t even supported on the latest version of Linux, and you’d have to look into workarounds to get it up and running on modern Linux flavors. Sure, Frigate hasn’t removed support for Coral TPU, and if you’ve already got one of these accelerators, you can still use them without any issues. However, the official Frigate documentation has outright stopped recommending Coral accelerators since version 0.17, and has instead started steering GPU-less setups towards Intel OpenVINO. Regular ol’ Intel CPUs can deliver similar results as this dead AI accelerator Yet another win for the humble N100 If you’ve read my server articles here at XDA, you’re probably aware that I adore the Quick Sync functionality in budget Intel processors, which pretty much eliminates the dedicated GPU requirement for video transcoding tasks. OpenVINO performs a similar role for object detection workloads, and since it’s available on most modern Intel CPUs, you don’t need to go out of your way to grab an AI accelerator or a powerful GPU just to harness AI features in Frigate. Better yet, if you’re using a mini-PC with an Intel NPU to self-host Frigate, you can even put the dedicated AI processing chip to good use thanks to OpenVINO. Depending on your processor, you might encounter slightly higher power consumption than a dedicated Coral Edge TPU, but seeing as how you don’t need to spend money on a dedicated accelerator, I call that a win. I’d probably use this setup myself if I wasn’t already using my Raspberry Pi + AI Kit combo. The Raspberry Pi AI Kit can help out with AI inference on RPi SBCs It’s good enough to handle my security cameras I grabbed the Raspberry Pi AI Kit when it first hit the shelves in 2024, but it wasn’t until a few weeks ago that I began using it with my RPI 5. Sure, I could technically run AI detection tasks on my tinkering companion’s processor, but I’d have to lower the stream resolution and configure large masks just to avoid the CPU utilization percentage from hitting the red zone. Tossing the Raspberry Pi AI Kit and its Hailo-8L (and there’s an even faster variant on newer AI Kit models) AI accelerator solved this dilemma for my SBC-based object detection tasks. Of course, I’ve configured some object masks to filter certain regions of my streams, but this pint-sized setup is more than enough to accommodate my surveillance cameras. Personally, I’d go for an OpenVINO Frigate pipeline running off an Intel mini-PC Truth be told, I’m only using my Raspberry Pi-based Frigate setup because I have a spare RPi 5 I bought ages ago. Considering the current pricing trends for (non-Zero) Raspberry Pi systems, I wouldn’t recommend grabbing one of these tiny boards. Instead, I’d suggest moving to an Intel N100 mini-PC and using the OpenVINO to run Yolov9/MobileNet v2 models to bring object detection capabilities to Frigate. Frigate