Hello all,
I’m trying to install frigate on HA. But I notice that when I try this my Intel NUC becomes very, very slow. However, it worked a few days ago, but now suddenly my NUC becomes very slow. What am I doing wrong, and how can I fix this?
Hello all,
I’m trying to install frigate on HA. But I notice that when I try this my Intel NUC becomes very, very slow. However, it worked a few days ago, but now suddenly my NUC becomes very slow. What am I doing wrong, and how can I fix this?
Hard to help without more data. How many cameras do you have? What values do you see in the Stats page in frigate? Are you running any detector?
Maybe start checking frigate logs (and potentially increase the log level) to see if something pops up.
How many cameras?
Are you using coral? This is highly recommended.
And the spec of the nuc, and what else it is running.
Thank you very much for your response.
I have 16 cameras but really only want to use 1 to 5 (if possible). And no, I don’t use Google coral. Does using this really make that much of a difference? and is it difficult to integrate into HA?
Thanks in advance for your comment.
And here are my MC details
Now I have also tried DOODS, I have the same problem there??
Without a TPU (i.e. Google Coral) switch OFF object detection for your cameras:
detect:
enabled: False
Now check the load on your NUC again. A Celeron J4025 should be able to handle 16 cameras. But without object detection only!
Yes, absolutely and without doubt.
Not at all. Just plug the Coral into an USB3 port on the NUC, enable it through frigate.yaml
and you are good to go.
PS. Also make sure to configure GPU hardware acceleration within frigate.yam
l.
Pretty much without a coral you are going to need top end CPU to manage object detection. even then it is not recommended. With coral it will manage fine.
I run 9 cameras on an old HP T630 with 16gb ram. Without coral it just does not run, with coral it does hit high CPU load sometimes but has been running for a few years now.
Thank you for your response. But the intention is: “object detection”!!
So if I understand correctly, with a google coral I could certainly integrate 5 cameras into HA and use object and person detection?
With a google coral for object detection and hardware acceleration enabled for decoding the video streams with that Intel NUC you can even integrate all of your cameras. But you definitely should read (and understand) how to fine tune Frigate! The learning curve is steep but once you got the grip on it’s workings it is not that hard anymore.
Thank you. I try a lot with HA. The strange thing is that I got both frigate and Doods running a few days ago and now it won’t work anymore because HA completely freezes when I even start frigate or Doods. I had to completely reset my NUC 3 times.
And by the way, what do you mean by: “and hardware acceleration enabled”. I’m not very familiar with everything in HA yet and am trying to find out as much as possible.
You need to read and understand (as much as possible) all the sections in the docs.
Hardware acceleration has it’s own section. Frigate is a great application, but it really has a steep learning curve, and if you are just learning your way around HA, it may be advisable to hang back a bit on Frigate.
You may well have got frigate up and running at one point but as soon as you start adding to the config for Frigate it will soon grind to a halt.
Hola, Sorry, but this is all a bit much for me.
Then I will have to learn a lot. Is it normal that I also have that problem with Doods? My CPU usage increases from 40% to more than 96% and HA is no longer accessible!!
If using a TPU like Coral you don’t need DOODS in conjunction with Frigate. Frigate has it’s own Object Detection Service already build in.
Anyway, without a TPU DOODS will also use your NUC’s CPU (and GPU) which most probably brings your system down sooner than later. I suspect you activated DOODS and have Object Detection activated with Frigate. Little are the chances that your NUC will not stall caused by a too heavy load with this combination!
If you insist on using object detection with Frigate but without a TPU disable DOODS and activate object detection within Frigate only. It might work with 2 or 3 cameras although expect a high load on your CPU and a not very accurate object detection.
But as said: When using Frigate DOODS is unnecessary.
Please read at least this paragraph to understand the difference between the workings of a CPU, GPU and TPU when desiring object detection
Thanks, sorry for the many questions. What works best: Frigate or DOODS. My goal is to slowly move towards facial detection. And if I understand it all correctly, this really doesn’t work well without Google coral, is that true?
Face recognition is a whole new ball game. I really can’t see that being possible even with a coral TPU. You are going to need a lot more power for that. Admittedly I have not played with Face recognition in frigate, but judging by the resources needed for object detection, faces are going to challenge your system.
I have not tried Doods, as frigate works so well. Before I had a coral TPU I used motioneye with detection on 2 cameras, which was passable but not a patch on Frigate.
No problem.
If you mean for object detection I can not answer you which piece of software works better. But both need a TPU to efficiently do object detection. Otherwise it will put a heavy load on your CPU (besides the CPUs coming with NUCs are not made for doing object detection).
I have never used DOODS but as far as I can tell it is a standalone image detection service. While Frigate has a detection service already build in.
I doubt that face recognition will work reliably anyway at this hardware and software level. Neither with DOODS nor with Frigate and also not in conjunction with a Coral. It’s ment for detecting whether an object is a person, a car, motorcycle, bird, cat, dog and many more. Look HERE for more information on detectable objects.