Trying to configure Frigate to use Nvidia GPU. SparkBox AI replied: That confirms Frigate isn’t getting the GPU: it’s using Docker’s runc runtime, with no GPU device request. Installing the toolkit and registering the NVIDIA runtime didn’t attach it to this container. SparkBox’s documented NVIDIA setting covers Jellyfin/media and certain other apps, but doesn’t specify Frigate’s GPU wiring. I won’t guess an override or tell you to restart it—that won’t add the missing GPU access. I don’t have Frigate’s SparkBox setup documented; use Share to demox so Tom can confirm the supported configuration.
11 replies
Chris wrote:
Got it — this is an automatic acknowledgment from Chris, your AI support representative. Your post is in the support queue, and we’ll get to it as soon as we can. Legend members receive priority replies. If you’d like faster support while helping fund the project’s goals and continued development, consider becoming a Legend backer. <!-- chris-support-queue-ack:v1 --
Chris wrote:
Thanks for reporting this: behavior. SparkBox's handling of that path was wrong, and it's fixed in the latest update. Your box will pick it up during its nightly update, or you can run sparkbox update now.
Chris wrote:
Hi justinuky, Chris here (SparkBox's support assistant, an AI). First, a correction to the note above: the command should have been sudo sparkbox upgrade, not sparkbox update. update only refreshes the apps; upgrade installs the new SparkBox. Sorry about that. What was wrong You did the NVIDIA side right. SparkBox only knew how to give Frigate Intel graphics. The Hardware Transcoding = NVIDIA setting handed your card to Jellyfin and the other media apps but never to Frigate, which is why it stayed on the normal runtime with no GPU. SparkBox 1.6.938 (out now) gives Frigate the card with that same setting. Steps, in the terminal on the box 1. sudo sparkbox upgrade 2. If Hardware Transcoding in Settings is not already nvidia: sudo sparkbox set-env JELLYFINHWACCEL nvidia (Jellyfin and the other media apps will use the card too) 3. sudo sparkbox restart frigate One check sudo docker inspect -f '{{.HostConfig.Runtime}}' sb-frigate should now print nvidia. Frigate then switches to NVIDIA decoding by itself, and its log says "Automatically detected nvidia hwaccel for video decoding". One exception: if Frigate's config has an hwaccelargs: line under ffmpeg: that says something other than auto or preset-nvidia (for example preset-vaapi), change it to preset-nvidia and restart Frigate once more. This covers decoding your camera streams. Running object detection on the card needs a different Frigate edition, which SparkBox does not ship. What our assistant knows has been corrected as well, so the next person who asks about Frigate and NVIDIA gets these steps instead of "not documented".
justinuky wrote:
Running the update now. Can you tell me why my legend status doesn't seem to be reflected in demox?
Chris wrote:
Sorry—your Legend status isn’t showing as expected. I can’t confirm the cause from the public thread, and I don’t want to guess at your account details. Please open a private ticket at https://demox.world/support using the same Demox account; Anthony can check why the entitlement isn’t appearing. Please keep purchase details and any licence information out of this public thread.
justinuky wrote:
after following the steps above, docker inspect on frigate still prints runc
Chris wrote:
Sorry those steps still haven't got Frigate onto the NVIDIA runtime. Frigate's NVIDIA support is included starting in SparkBox 1.6.938; what SparkBox version does your dashboard show now? That will tell us the right next step without guessing.
justinuky wrote:
I realized after my last reply that when i upgraded sparkbox, it upgraded to 1.6.937. That is the newest version available to me.
Chris wrote:
Thanks—that clarifies it. The Frigate/NVIDIA fix is in SparkBox 1.6.938, so 1.6.937 does not include it yet. Since your last upgrade only reached 1.6.937, please try sudo sparkbox upgrade once more; if it still leaves you on 1.6.937, reply with that version and we’ll keep this open rather than repeat the Frigate checks.
justinuky wrote:
Here is what i get when attempting the update: This box is in the group still on v1.6.937 while v1.6.938 rolls out gradually — upgrading to v1.6.937. [OK] SparkBox is already on the latest release (v1.6.937).
Chris wrote:
Sorry—I sent you back to retry the upgrade when your output already showed that 1.6.938 was not available to your box yet. The Frigate/NVIDIA change is in the published 1.6.938 release, but your updater is still offering 1.6.937, so another retry will not get you that fix. Please check the SparkBox version shown in your dashboard later; once it shows 1.6.938, let us know whether Frigate still reports runc using the same check you already ran. What our assistant knows has been corrected so it does not send the next person into the same retry loop.