[HW Accel Support]: High GPU Memory usage #13293
-
Describe the problem you are havingDespite coral TPU, I have high GPU memory consumption. Is that normal? Version0.14.0 In which browser(s) are you experiencing the issue with?No response Frigate config file# Frigate Configuration file
# Defaults: https://docs.frigate.video/configuration/#full-configuration-reference
# MQTT needed for Smart Home Applications
mqtt:
host: 10.11.1.236 # <---- same mqtt broker that home assistant uses
user: '{FRIGATE_MQTT_USER}'
password: '{FRIGATE_MQTT_PASSWORD}'
#ui:
# Optional: Set the default live mode for cameras in the UI (default: shown below)
# live_mode: webrtc
detectors:
coral:
type: edgetpu
#device: pci
# Hardware acceleration of video conversion
ffmpeg:
# global hwaccel args
hwaccel_args: preset-vaapi
#hwaccel_args: preset-intel-qsv-h265
database:
path: /config/frigate.db
# Overview of all cameras https://frigate.video/images/birdseye-2-optimized.jpg
birdseye:
enabled: true
mode: objects
restream: true
# Snapshots of events
snapshots:
# Enable writing jpg snapshot to /media/frigate/clips
enabled: true
# print a timestamp on the snapshots
timestamp: true
# List of objects to capture https://docs.frigate.video/configuration/objects/
objects:
track:
- car
- person
- cat
- dog
- bicycle
- eye glasses
- backpack
- handbag
go2rtc:
log:
exec: trace
ffmpeg:
input_mulaw: -f mulaw -ar 8000 -i {input}
noise_filter: -af highpass=200,lowpass=3000,afftdn
streams:
carportcam:
- rtsp://{FRIGATE_RTSP_USER}:{FRIGATE_RTSP_PASSWORD}@10.11.1.70:88/videoMain
- ffmpeg:carportcam#audio=aac
carportcam_sub:
- rtsp://{FRIGATE_RTSP_USER}:{FRIGATE_RTSP_PASSWORD}@10.11.1.70:88/videoSub # <- stream which supports video & aac audio.
- ffmpeg:carportcam_sub#audio=aac # <- copy of the stream which transcodes audio to opus for webrtc
haustuercam:
- rtsp://10.11.1.71/mpeg/media.amp#media=video
- ffmpeg:http://10.11.1.71/bha-api/bha-api/audio-receive.cgi#media=audio#backchannel=0#input=input_mulaw#audio=opus#audio=noise_filter
hausseitecam:
- rtsp://{FRIGATE_RTSP_USER}:{FRIGATE_RTSP_PASSWORD}@10.11.1.72:88/videoMain
- ffmpeg:hausseitecam#audio=aac
hausseitecam_sub:
- rtsp://{FRIGATE_RTSP_USER}:{FRIGATE_RTSP_PASSWORD}@10.11.1.72:88/videoSub # <- stream which supports video & aac audio.
- ffmpeg:hausseitecam_sub#audio=aac
werkstattcam:
- rtsp://{FRIGATE_RTSP_USER}:{FRIGATE_RTSP_PASSWORD}@10.11.1.76/stream1
- ffmpeg:werkstattcam#audio=aac
werkstattcam_sub:
- rtsp://{FRIGATE_RTSP_USER}:{FRIGATE_RTSP_PASSWORD}@10.11.1.76/stream1 # <- stream which supports video & aac audio.
- ffmpeg:werkstattcam_sub#audio=aac # <- copy of the stream which transcodes audio to opus for webrtc
terrassecam:
- rtsp://{FRIGATE_RTSP_USER}:{FRIGATE_RTSP_PASSWORD}@10.11.1.75:88/videoMain
- ffmpeg:hausseitecam_sub#audio=aac
terrassecam_sub:
- rtsp://{FRIGATE_RTSP_USER}:{FRIGATE_RTSP_PASSWORD}@10.11.1.75:88/videoSub # <- stream which supports video & aac audio.
- ffmpeg:hausseitecam_sub#audio=aac
babycam:
- rtsp://{FRIGATE_RTSP_USER}:{FRIGATE_RTSP_PASSWORD}@10.11.1.77/live0
- ffmpeg:babycam#audio=opus
babycam_onvif:
- onvif://{FRIGATE_RTSP_USER}:{FRIGATE_RTSP_PASSWORD}@10.11.1.77
babycam_sub:
- rtsp://{FRIGATE_RTSP_USER}:{FRIGATE_RTSP_PASSWORD}@10.11.1.77/live0 # <- stream which supports video & aac audio.
- ffmpeg:babycam_sub#audio=opus # <- copy of the stream which transcodes audio to opus for webrtc
gartencam:
- ffmpeg:rtsp://{FRIGATE_RTSP_USER}:{FRIGATE_RTSP_PASSWORD}@10.11.1.73:554/cam/realmonitor?channel=1&subtype=1
- ffmpeg:gartencam#video=h264#audio=aac
gartencam_sub:
- ffmpeg:rtsp://{FRIGATE_RTSP_USER}:{FRIGATE_RTSP_PASSWORD}@10.11.1.73:554/cam/realmonitor?channel=1&subtype=2
- ffmpeg:gartencam#video=h264#audio=aac
webrtc:
candidates:
- 10.11.1.3:8555
- stun:8555
record:
enabled: true
retain:
days: 7
mode: motion
events:
retain:
default: 14
mode: active_objects
# List of cameras
cameras:
carportcam:
ffmpeg:
inputs:
- path: rtsp://10.11.1.3:8554/carportcam # <--- the name here must match the name of the camera in restream
input_args: preset-rtsp-restream
roles:
- record
- path: rtsp://10.11.1.3:8554/carportcam_sub # <--- the name here must match the name of the camera_sub in restream
input_args: preset-rtsp-restream
roles:
- detect
output_args:
record: preset-record-generic-audio-aac
live:
stream_name: carportcam_sub
detect:
width: 1280
height: 720
fps: 5
snapshots:
required_zones:
- pforte
- auffahrt
- komplett
zones:
auffahrt:
coordinates: 479,195,874,720,1280,720,1280,142,1095,81,974,103,837,82,817,238,644,242
objects:
- car
komplett:
coordinates: 1280,0,1280,720,0,720,42,238,140,38,210,27,213,0
objects:
- person
- cat
- dog
- bicycle
- eye glasses
- backpack
- handbag
motion:
mask:
- 1280,0,936,0,929,64,1089,104,1280,157
- 308,280,21,672,0,350,269,165,254,0,532,0,543,82,382,99,398,192,557,184,552,267
- 508,51,829,54,821,103,778,246,699,130,635,130,505,136
objects:
filters:
person:
mask: 487,143,521,138,519,63,486,68
# review:
# alerts:
# required_zones:
# - pforte
# - auffahrt
# - komplett
haustuercam:
ffmpeg:
output_args:
record: preset-record-generic-audio-aac
inputs:
- path: rtsp://10.11.1.3:8554/haustuercam?video=all&audio=all
input_args: preset-rtsp-restream
roles:
- record
- detect
zones:
komplett:
coordinates: 0,720,1280,720,1280,0,0,0
objects:
- person
- cat
- dog
- bicycle
- eye glasses
- backpack
- handbag
snapshots:
required_zones:
- komplett
live:
stream_name: haustuercam
detect:
width: 1280
height: 720
fps: 5
review:
alerts:
required_zones:
- komplett
hausseitecam:
ffmpeg:
inputs:
- path: rtsp://10.11.1.3:8554/hausseitecam
input_args: preset-rtsp-restream
roles:
- record
- path: rtsp://10.11.1.3:8554/hausseitecam_sub # <--- the name here must match the name of the camera_sub in restream
input_args: preset-rtsp-restream
roles:
- detect
output_args:
record: preset-record-generic-audio-aac
live:
stream_name: hausseitecam_sub
detect:
width: 1280
height: 720
fps: 5
motion:
mask:
- 988,720,1280,720,1280,586,1280,411,1136,271,1055,201,907,107,758,34,715,123
werkstattcam:
ffmpeg:
inputs:
- path: rtsp://10.11.1.3:8554/werkstattcam
input_args: preset-rtsp-restream
roles:
- record
- path: rtsp://10.11.1.3:8554/werkstattcam_sub # <--- the name here must match the name of the camera_sub in restream
input_args: preset-rtsp-restream
roles:
- detect
output_args:
record: preset-record-generic-audio-aac
detect:
width: 1280
height: 720
fps: 5
live:
stream_name: werkstattcam_sub
snapshots:
required_zones:
- autowerkstatt
- komplett
zones:
autowerkstatt:
coordinates: 609,94,612,0,909,0,928,80,911,289,0,202,91,96
objects:
- car
komplett:
coordinates: 1280,0,421,0,405,32,0,37,0,720,1280,720
objects:
- person
- cat
- dog
- bicycle
- eye glasses
- backpack
- handbag
motion:
mask:
- 338,237,0,622,0,94,277,0
review:
alerts:
required_zones:
- autowerkstatt
- komplett
terrassecam:
ffmpeg:
inputs:
- path: rtsp://10.11.1.3:8554/terrassecam?video&audio
# input_args: preset-rtsp-restream
roles:
- record
- path: rtsp://10.11.1.3:8554/terrassecam_sub?video&audio
# input_args: preset-rtsp-restream
roles:
- detect
output_args:
record: preset-record-generic-audio-aac
motion:
mask:
- 338,237,0,622,0,94,277,0
record:
events:
objects:
- person
- cat
- dog
- bicycle
- eye glasses
- backpack
- handbag
detect:
width: 1280
height: 720
fps: 5
live:
stream_name: terrassecam_sub
objects:
filters:
handbag:
threshold: 0.8
person:
threshold: 0.76
gartencam:
ffmpeg:
inputs:
- path: rtsp://10.11.1.3:8554/gartencam?video&audio
# input_args: preset-rtsp-restream
roles:
- record
- path: rtsp://10.11.1.3:8554/gartencam_sub?video&audio
# input_args: preset-rtsp-restream
roles:
- detect
output_args:
record: preset-record-generic-audio-aac
motion:
mask:
- 144,284,152,531,57,720,0,720,0,94,0,0,319,0
snapshots:
required_zones:
- komplett
zones:
komplett:
coordinates: 0,720,1280,720,1280,0,0,0
objects:
- person
- cat
- dog
- bicycle
- eye glasses
- backpack
- handbag
detect:
width: 1280
height: 720
fps: 5
live:
stream_name: gartencam_sub
objects:
filters:
handbag:
threshold: 0.8
person:
threshold: 0.76
review:
alerts:
required_zones:
- komplett
version: 0.14docker-compose file or Docker CLI commandName:
frigate
Beschreibung:
A complete and local NVR designed for Home Assistant with AI object detection. Uses OpenCV and Tensorflow to perform realtime object detection locally for IP cameras.
You cas use a integrated or dedicated GPU (Intel/AMD/Nvidia) to perform the image decoding of the input streams of your cameras.
Optionally (but highly recommended), you can use multiple devices to perform the object detetion, such as a Google Coral Accelerator Nvidia GPU, OpenVINO (Intel GPU) and Intel NCS2 VPU
See the documentation for more details about your specific hardware and needs (https://docs.frigate.video/). This is a general templeate that aims to give a fast deoplyment for every user so check all the options carefully and remove all the configurations that you are not going to use.
Zusätzliche Anforderungen:
<br>- A valid config.yml file must exist in the config directory to startup the container.<br>- If you are using a Coral Detector, you must set the container to privileged mode under advanced view<br>- If you are using a PCI Coral instead of a USB one, you must install first the needed drivers going to the CA app and searching for Coral-Driver (thanks to @ich777)<br>- If you want to use a NVidia card to image decoding and/or detection, you must first install the drivers from CA app (thanks to @ich777), add "--runtime=nvidia" as extra parameter under advanced view and set the "Nvidia Visible Devices" variable with your "GPU UUID" as the value.<br>- If you want to use the NVidia TensorRT Detector, you have to select the correspondig NVIDIA branch.<br>- If you want to use a AMD GPU to image decoding you must change driver to "radeonsi"<br>- With some of the Intel processors, like the J4125, the default driver "iHD" doesn't seem to work correctly for hardware acceleration. You may need to change the driver to "i965"<br>- If you want to use a NCS2 as detector you have to add "--device-cgroup-rule='c 189:\* rmw'" as extra parameter under advanced view and install the "udev rule" in the unraid host.
Quelle:
ghcr.io/blakeblackshear/frigate:stable
Kategorien:
Heimautomation, Sicherheit
Kategorien
Sicherung
Cloud
Kryptowährung
Download-Tools
Treiber
Spiele-Server
Heimautomation
Produktivität
Sicherheit
Werkzeuge
Andere
Medien-App
Medien-App:Video
Medien-App:Musik
Medien-App:Bücher
Medien-App:Fotos
Medien-App:Andere
Medien-Server
Medien-Server:Video
Medien-Server:Musik
Medien-Server:Bücher
Medien-Server:Fotos
Medien-Server:Andere
Netzwerk
Netzwerk:Web
Netzwerk:DNS
Netzwerk:FTP
Netzwerk:Proxy
Netzwerk:VoIP
Netzwerk:Management
Netzwerk:Messenger
Netzwerk:VPN
Netzwerk:Privacy
Netzwerk:Andere
Development Status
Status:Stabil
Status:Beta
...schließen
Hilfe-Forenbeitrag:
https://forums.unraid.net/topic/98064-support-blakeblackshear-frigate/
Projektseite:
https://frigate.video/
Zuerst lesen:
Registry URL:
https://github.com/blakeblackshear/frigate/pkgs/container/frigate
Symbol-URL:
https://raw.githubusercontent.com/yayitazale/unraid-templates/main/frigate.png
WebUI:
http://[IP]:[PORT:5000]
Extra Parameter:
--shm-size=512mb --mount type=tmpfs,target=/tmp/cache,tmpfs-size=10000000000 --restart unless-stopped
Post-Argumente:
CPU Zuweisung:
CPU:
HT:
Netzwerktyp:
Custom : br0
Statische IP-Adresse (Optional):
10.11.1.3
Subnet: 10.11.1.0/24
Konsole:
Shell
Privilegierte Rechte:
An
Config Path:
/mnt/user/appdata/frigate
BearbeitenEntfernen
Container Path: /config
Media path:
/mnt/user/Kameras/
BearbeitenEntfernen
Container Path: /media/frigate
HTTP port:
5000
BearbeitenEntfernen
Container Port: 5000
RTSP port:
8554
BearbeitenEntfernen
Container Port: 8554
Frigate RTSP Password:
7CYCi1MixmnO
BearbeitenEntfernen
Container Variable: FRIGATE_RTSP_PASSWORD
Frigate+ API key:
BearbeitenEntfernen
Container Variable: PLUS_API_KEY
Optional: Enter the frigate+ API key or remove this if you are not using it
go2rtc TCP port:
8555
BearbeitenEntfernen
Container Port: 8555
If you are not going to use MSE/WebRCT you can remove all the WebRTC entrys
go2rtc UDP port:
8555
BearbeitenEntfernen
Container Port: 8555
If you are not going to use MSE/WebRCT you can remove all the WebRTC entrys
go2rtc server port:
1984
BearbeitenEntfernen
Container Port: 1984
For Home Assistant WebRTC integration. Remove this if you are not using it.
Frigate RTSP User:
••••••••••••
BearbeitenEntfernen
Container Variable: FRIGATE_RTSP_USER
Frigate mqtt User:
mqttuser
BearbeitenEntfernen
Container Variable: FRIGATE_MQTT_USER
Frigate matt Password:
••••••••••••
BearbeitenEntfernen
Container Variable: FRIGATE_MQTT_PASSWORD
Coral TPU/NCS2 Mapping:
/dev/apex_0
BearbeitenEntfernen
Container Device:
Use /dev/bus/usb for USB devices and /dev/apex_0 for PCIe devices (you must install the drivers first for PCIe devices). Remove this if you are not using it.
Intel/AMD GPU mapping:
/dev/dri/renderD128
BearbeitenEntfernen
Container Device:
Intel/AMD GPU mapping for image decode (and detection with Intel OpenVINO). Remove this if you are not using it.
Driver for Intel/AMD GPUs:
radeonsi
BearbeitenEntfernen
Container Variable: LIBVA_DRIVER_NAME
(Intel = iHD or i965 and AMD = radeonsi). Remove this if you are not using it.
Mehr Einstellungen ausblenden ...
Localtime:
/etc/localtimeRelevant Frigate log outputsiehe ScreenshotRelevant go2rtc log output...FFprobe output from your camera...Operating systemUNRAID Install methodDocker Compose Network connectionWired Camera make and modelsix cameras Screenshots of the Frigate UI's System metrics pages and any other screenshots pertinent to your issue.
Any other information that may be helpfulReturn Code: 0 Process Output: vainfo: VA-API version: 1.17 (libva 2.10.0) |
Beta Was this translation helpful? Give feedback.
Answered by
NickM-27
Aug 23, 2024
Replies: 3 comments 4 replies
-
|
how much memory does your GPU have allocated? this is not unusual in general with this many cameras since ffmpeg uses the GPU for decoding. If this is an iGPU then you can increase the amount of GPU memory in the BIOS |
Beta Was this translation helpful? Give feedback.
0 replies
Answer selected by
kobelka
-
──────────────────────────────────────────────────────────────────────────┼─────────────────────────────────────────────────────────────────────── |
Beta Was this translation helpful? Give feedback.
4 replies
-
|
yeah looks like very little, you'll want to increase it in the BIOS |
Beta Was this translation helpful? Give feedback.
-
|
How much do you recommend? |
Beta Was this translation helpful? Give feedback.
-
|
2G would be a good starting point but really it depends how much system memory you have |
Beta Was this translation helpful? Give feedback.
-
|
@kobelka I had the same problem.
Everything works fine but with a GPU usage around 88%. The value of 88% came out using Auto.
For the Frigate LXC container I had dedicated 8 GB RAM. Just my 2 cents |
Beta Was this translation helpful? Give feedback.
-
|
I have now changed my vram to 2 gb in the bios and so far it looks good :-)
|
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
how much memory does your GPU have allocated? this is not unusual in general with this many cameras since ffmpeg uses the GPU for decoding. If this is an iGPU then you can increase the amount of GPU memory in the BIOS