[HW Accel Support]: Cheap GPU which supports H265 video decoding?

[Johnyb62]

Describe the problem you are having

System:
Unraid server running several dockers (Dell 3020 - Intel® Core™ i5-4570 CPU @ 3.20GHz / 8Gb RAM)
PCI Coral running for detector
Camera: Annke C800 (

I'd like to add a GFX card with hardware decoding to take the burden from the CPU. The CCTV cameras output H265 so needs to manage this. Looking for a cheap 2nd hand card from ebay, recommendations welcomed. Thanks.

Version

0.13.2-6476f8a

Frigate config file

mqtt:
  #enabled: false
  host: 192.168.0.63
  user: xxxxxx
  password: xxxxx
  
cameras:
  Front:  # <------ Name the camera
    ffmpeg:
      inputs:
        - path: rtsp://xxxxx:[email protected]:554/Streaming/Channels/102/?transportmode=unicast.sdp # <----- The stream you want to use for detection
          roles:
            - detect
            
        - path: rtsp://xxxxx:[email protected]:554/Streaming/Channels/101/?transportmode=unicast.sdp # <----- The stream you want to use for detection
          roles:
            - record
      #hwaccel_args: preset-vaapi
      
    detect:
      enabled: True # <---- disable detection until you have a working camera feed
      width: 640 # <---- update for your camera's resolution
      height: 480 # <---- update for your camera's resolution
      fps: 5
      
    objects:
        track:
        - person
        - dog
        - cat
 
    motion:
        mask:
        - 243,0,252,480,0,480,0,0
        - 640,0,522,147,372,99,240,157,212,0
        - 640,23,640,480,529,480,396,480,448,298,610,39
    
    zones:
        Driveway:
            coordinates: 262,480,388,480,432,294,527,167,402,87,188,175
    
    
    record:
        enabled: True
        events:
            retain:
                default: 45
                mode: active_objects
            required_zones:
              - Driveway
                
    snapshots: # <----- Enable snapshots
      enabled: True
      required_zones:
        - Driveway

  Back_1:  # <------ Name the camera
    ffmpeg:
      output_args:
        record: preset-record-generic-audio-aac
      inputs:
        - path: rtsp://xxxxx:[email protected]:554/Streaming/Channels/302/?transportmode=unicast.sdp # <----- The stream you want to use for detection
          roles:
            - detect
            - audio
            
        - path: rtsp://xxxxx:[email protected]:554/Streaming/Channels/301/?transportmode=unicast.sdp # <----- The stream you want to use for detection
          roles:
            - record
      
    detect:
      enabled: True # <---- disable detection until you have a working camera feed
      width: 640 # <---- update for your camera's resolution
      height: 480 # <---- update for your camera's resolution
      fps: 5
      
    objects:
        track:
        - person
        - dog
        - cat

    motion:
        mask:      
        - 448,0,640,0,640,154,640,275,518,291,473,215,405,168,388,92,414,48
        - 0,0,0,231,104,28,189,27,187,0
    
    zones:
        Garden_1:
            coordinates: 545,480,640,361,640,255,582,173,545,285,382,135,450,25,431,0,182,0,0,480
    
    
    record:
        enabled: True
        events:
            retain:
                default: 45
                mode: active_objects
            required_zones:
              - Garden_1
                
    snapshots: # <----- Enable snapshots
      enabled: True
      required_zones:
        - Garden_1
    
    audio:
      enabled: True
      min_volume: 400
      max_not_heard: 30
      listen:
        - bark
      filters:
        bark:
          threshold: 0.5
          
detectors:
  coral_pci:
    type: edgetpu
    device: pci 
#ffmpeg:
#  hwaccel_args: preset-vaapi

docker-compose file or Docker CLI command

n/a

Relevant log output

n/a

FFprobe output from your camera

[{"return_code":0,"stderr":"","stdout":{"programs":[],"streams":[{"avg_frame_rate":"0/0","codec_long_name":"H.265/HEVC(HighEfficiencyVideoCoding)","display_aspect_ratio":"4:3","height":480,"width":640},{"avg_frame_rate":"0/0","bit_rate":"64000","codec_long_name":"PCMmu-law/G.711mu-law"}]}},{"return_code":0,"stderr":"","stdout":{"programs":[],"streams":[{"avg_frame_rate":"0/0","codec_long_name":"H.265/HEVC(HighEfficiencyVideoCoding)","display_aspect_ratio":"16:9","height":2160,"width":3840},{"avg_frame_rate":"0/0","bit_rate":"64000","codec_long_name":"PCMmu-law/G.711mu-law"}]}}]

Operating system

UNRAID

Install method

Docker Compose

Network connection

Wired

Camera make and model

Annke C800

Any other information that may be helpful

No response

[NickM-27]

most spec sheets will show if the GPU supports decoding and for what codecs

[Johnyb62]

most spec sheets will show if the GPU supports decoding and for what codecs

Ok but when looking at the NVIDIA table for example, it shows HEVC 4:2:0 and 4:4:4 @ 8/10/12bit
How do I know when one is required for my camera? ( I guess email the manufacturer?)

[NickM-27]

yeah, though in general cameras use the most basic profiles for hevc

[ElectronicBattle]

I bought a secondhand GT1030 off eBay a month ago for about 25 quid. You'll have seen the Nvidia developer page here where you can type in "1030" and see what's available.

I went for the 1030 on account of it being fanless and reduced height to suit my computer. It does ndec (Nvidia decoding) but not nvenc (encoding) so it's not suitable for full transcoding. My cameras also generate hvec and I am able to use the GPU to offload the burden from my CPU.

On another box I have a relatively expensive Radeon RX580 which can do full transcoding and it works well but I have formed the impression (based on not much) that Nvidia is a better bet for compatibility and open-source driver availability.

I expected the Nvidia to be running full-tilt but it doesn't; about 15% core and I can't remember the GPU memory load (it's not working at the moment). The Radeon has a 3-4% GPU load and 50-60% RAM use.