Seeed Studio Coral M.2 Accelerator with Dual Edge TPU

not working at all

Works wonders!

TPU détecté, mais pour ce qui est du contenu du colis, il n'y aura que le tpu en lui-même. Pas de dissipateur thermique, manuel d'utilisation comme promis dans la description! J'ai posé la question au vendeur, mais pas de retour! Cordialement.

Picked up the Coral M.2 Dual Edge TPU specifically to offload Frigate NVR object detection from the host CPU and it has transformed the setup completely. Running Frigate on Proxmox with multiple camera streams and the Coral handles the inferencing load without breaking a sweat — detection latency dropped dramatically and CPU utilization on the Proxmox host went from uncomfortably high during active detection to nearly negligible. The dual Edge TPU is the right spec for a multi-camera Frigate deployment. Single TPU units hit their limits when you're running enough camera streams with active object detection — the dual TPU handles the parallel inferencing load that a serious NVR setup generates without queuing delays that cause missed detections on fast moving objects. Passthrough to the Frigate LXC or VM on Proxmox works correctly once the PCIe passthrough is configured properly. Driver installation on the container side is straightforward following the Coral documentation and Frigate recognizes the device cleanly on startup without any manual configuration beyond pointing the detector config at the right device path. Detection accuracy and speed on the Coral versus pure CPU inference is not a subtle difference — it's a completely different experience. Person, vehicle, and animal detection on multiple simultaneous streams with sub-100ms latency is what a proper NVR setup should deliver and the Coral makes that possible without dedicated GPU hardware. For anyone running Frigate on Proxmox or similar homelab NVR setups, the Coral M.2 Dual Edge TPU is the upgrade that makes everything else click into place. Highly recommended without reservation.

Shoved this in my qnap to speed up the QuMagie AI. Works great shows up as 2 TPU's but had to buy a special multiplexing adapter to an m.2 slot and then also get a m.2 to pci slot card. But that's as much to do with the qnap as the card itself.