[HW Accel Support]: vendor-agnostic *_gpu_top tool

[Arol450]

Describe the problem you are having

People seem to be having problems with intel_gpu_top.
I myself could not get it to work within a (podman) container. Rootless, systemd-started, system-user-mapped.
I tried all permutations of Capabilities, AmbientCapabilities, added capabilities, etc.

frigate.util.services ERROR : Unable to poll intel GPU stats: Failed to initialize PMU! (Operation not permitted)

I can get a different tool to work, a hardware agnostic version of intel_gpu_top.
https://gitlab.freedesktop.org/drm/igt-gpu-tools
https://gitlab.freedesktop.org/drm/igt-gpu-tools/-/blob/master/NEWS : ' ...- New vendor agnostic gputop tool. (Tvrtko Ursulin) ...'

DRM minor 128
PID MEM RSS render copy video compute NAME
1861 166M 161M | 0.0% || 0.0% || 0.0% || 0.0% || 60.7% █▉ | frigate.detecto
2280 1G 603M | 0.0% || 0.0% || 5.7% ▎ || 0.0% || 0.0% | ffmpeg
2330 1G 648M | 0.0% | 0.0% || 5.0% ▎ || 0.0% || 0.0% | ffmpeg
2176 587M 331M | 0.0% || 0.0% || 3.1% ▏ || 0.0% || 0.0% | ffmpeg
2177 556M 302M | 0.0% || 0.0% || 2.9% ▏ || 0.0% || 0.0% | ffmpeg
2171 556M 302M | 0.0% || 0.0% || 2.9% ▏ || 0.0% || 0.0% | ffmpeg
2184 559M 301M | 0.0% || 0.0% || 2.9% ▏ || 0.0% || 0.0% | ffmpeg
2154 557M 307M | 0.0% || 0.0% || 2.6% ▏ || 0.0% || 0.0% | ffmpeg
2231 552M 361M | 0.0% || 0.0% || 2.4% ▏ || 0.0% || 0.0% | ffmpeg
2573 451M 246M | 0.0% || 0.0% || 2.1% ▏ || 0.0% || 0.0% | ffmpeg

I get pretty believable statistics for video acceleration and detector-compute utilization.
I thought i would mention this.

This is on Intel Meteorlake hardware, a medium yolo_nas model on OpenVino, and transparently trans-coding the main stream to AV1 video format, saving a ton of space on the drive, h264_qsv -> av1_qsv
CPU is bored, 90%idle with 18 4MP cameras.

Version

0.15.0-a68c7f4

Frigate config file

detectors:
  ov:
    type: openvino
    device: AUTO

model:
  model_type: yolonas
  width: 640
  height: 640
  input_tensor: nchw
  input_pixel_format: bgr
  path: /config/yolo_nas_m.onnx
  labelmap_path: /labelmap/coco-80.txt

ffmpeg:
  hwaccel_args: preset-intel-qsv-h264

go2rtc:
  streams:
    Cars_live:
      - ffmpeg:rtsp://****:****@192.168.2.44/Streaming/Channels/1

cameras:
  Cars:
    ffmpeg:
      output_args:
        record: -f segment -segment_time 10 -segment_format mp4 -reset_timestamps 1 -strftime 1 -c:v av1_qsv -global_quality:v 31 -an
      inputs:
        - path: rtsp://localhost:8554/Cars_live  
          input_args: -loglevel error -hwaccel qsv -qsv_device /dev/dri/renderD128 -hwaccel_output_format qsv -c:v h264_qsv
          roles:
            - record
        - path: rtsp://****:****@192.168.2.44/Streaming/Channels/102
          roles:
            - detect
    detect:
      enabled: true
      width: 640
      height: 480
    motion:
      threshold: 30
      contour_area: 10
      improve_contrast: true
    live:
      stream_name: Cars_live

docker-compose file or Docker CLI command

systemd.service:

[Unit]
Wants=network-online.target
After=network-online.target
Description=Run frigate-vpl in a container

[Service]
User=frigate
#Group=users
Type=notify
Restart=always
KillMode=control-group
TimeoutStartSec=180
TimeoutStopSec=360

ExecStart=/usr/bin/podman run -d \
    --name=frigate-vpl \
    --cidfile=/run/user/967/frigate-vpl.cid \
    --replace --rm --cgroups=split \
    --sdnotify=conmon \
    -p 8971:8971 \
    --device /dev/dri:/dev/dri:rwm \
    --device /dev/accel/accel0:/dev/accel/accel0:rwm \
    -v /mnt/frigate/config:/config \
    -v /mnt/frigate/storage:/media/frigate \
    -v /home/frigate/ssl/cert.crt:/etc/letsencrypt/live/frigate/fullchain.pem:ro \
    -v /home/frigate/ssl/cert.key:/etc/letsencrypt/live/frigate/privkey.pem:ro \
    --mount type=tmpfs,tmpfs-size=1G,destination=/tmp/cache \
    --shm-size=1G \
    --stop-timeout=320 \
#    --privileged \
#    --cap-add CAP_PERFMON \
#    --cap-add PERFMON \
#    --cap-add CAP_SYS_ADMIN \
#    --cap-add SYS_ADMIN \
    ghcr.io/blakeblackshear/frigate:a68c7f4
ExecStop=/usr/bin/podman rm -v -f -i  --cidfile=/run/user/967/frigate-vpl.cid
Delegate=yes

NotifyAccess=all
SyslogIdentifier=%N

# an attempt of hardening

#MemoryDenyWriteExecute=yes
#SystemCallFilter=~memfd_create
#CapabilityBoundingSet=CAP_SETGID CAP_SETUID CAP_SYS_PACCT CAP_KILL CAP_WAKE_ALARM CAP_LINUX_IMMUTABLE CAP_IPC_LOCK CAP_SYS_MODULE CAP_SYS_BOOT CAP_LEASE CAP_NET_BIND_SERVICE CAP_NET_BROADCAST CAP_NET_RAW CAP_SYS_NICE CAP_SYS_RESOURCE CAP_NET_ADMIN CAP_SYS_PTRACE CAP_SYS_RAWIO CAP_SYS_ADMIN CAP_SYSLOG CAP_SYS_TIME CAP_MKNOD  CAP_BLOCK_SUSPEND CAP_CHOWN CAP_FSETID CAP_SETFCAP

#AmbientCapabilities=CAP_PERFMON CAP_SYS_ADMIN

# restrict IP address ranges to 192.168.2.1-192.168.2.128 and
IPAddressDeny=any
IPAddressAllow=localhost @frigate_group

[Install]
WantedBy=multi-user.target

Relevant Frigate log output

frigate.util.services          ERROR   : Unable to poll intel GPU stats: Failed to initialize PMU! (Operation not permitted)

Relevant go2rtc log output

none

FFprobe output from your camera

cameras are not the problem

Install method

Docker CLI

Object Detector

OpenVino

Network connection

Wired

Camera make and model

Hitosino something something

Screenshots of the Frigate UI's System metrics pages

Any other information that may be helpful

No response