[HW Accel Support]: AMD rx 7900 xtx GPU:100%

[Chreece]

Describe the problem you are having

The GPU is always on 100% with 150W power consumption, a lot of hot air comming from, as long as I start frigate. I have also whisper and ollama loaded on GPU but the activity on GPU is 0% if they don't be used.

Tried removing:

ffmpeg:
  hwaccel_args: preset-vaapi

detect, face_recognition, semantic_search.

Rocm and amd_gpu also reinstalled.

Ollama, whisper which are also using rocm are working fine.

Version

v16.0

Frigate config file

mqtt:
  host: ip
  port: port
  topic_prefix: user
  client_id: id
  user: user
  password: password
  stats_interval: 60
birdseye:
  enabled: true
  height: 1920
  mode: objects
  quality: 8
  width: 2560
ffmpeg:
  hwaccel_args: preset-vaapi
go2rtc:
  streams:
    eingang:
      - ffmpeg:rtsp://username:password@ip:554/stream1#video=copy#audio=copy#audio=aac
      - tapo://email:password@ip
  webrtc:
    candidates:
      - stun:8555
  api:
    origin: '*'
cameras:
  eingang:
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/eingang?video=copy&audio=aac
          input_args: preset-rtsp-restream
          roles:
            - record
            - audio
            - detect
    live:
      streams:
        eingang: eingang
    audio:
      enabled: true
    onvif:
      host: ***
      port: ***
      user: ***
      password: ***
    objects:
      track:
        - person
    detect:
      fps: 5
      stationary:
        interval: 10
        threshold: 50
        max_frames:
          default: 3000
          objects:
            person: 1000
    motion:
      threshold: 30
      contour_area: 40
      delta_alpha: 0.2
      frame_alpha: 0.2
      frame_height: 180
      improve_contrast: true
    snapshots:
      enabled: true
      timestamp: true
      bounding_box: false
      crop: false
      retain:
        default: 10
      height: 1920
      quality: 100
    mqtt:
      bounding_box: false
      enabled: true
      quality: 100
      timestamp: false
    record:
      enabled: true
      retain:
        days: 7
        mode: motion
      alerts:
        retain:
          days: 7
      detections:
        retain:
          days: 7
    face_recognition:
      enabled: true
      min_area: 500
detectors:
  onnx:
    type: onnx
model:
  model_type: rfdetr
  width: 320
  height: 320
  input_tensor: nchw
  input_dtype: float
  path: /config/rfdetr-Medium.onnx
semantic_search:
  enabled: true
  reindex: false
  model_size: large
face_recognition:
  enabled: true
  model_size: large
detect:
  enabled: true
lpr:
  enabled: false
classification:
  bird:
    enabled: false
version: 0.16-0

docker-compose file or Docker CLI command

services:
  frigate:
    container_name: frigate
    privileged: true
    restart: unless-stopped
    image: ghcr.io/blakeblackshear/frigate:0.16.0-rocm
    shm_size: "140mb"
    devices:
      - /dev/kfd
      - /dev/dri
    volumes:
      - /etc/localtime:/etc/localtime:ro
      - /home/chreece/AI/frigate/config:/config
      - /home/chreece/Backups/localusb/frigate/storage:/media/frigate
      - type: tmpfs
        target: /tmp/cache
        tmpfs:
          size: 1000000000
    ports:
      - "5001:5000"
      - "8554:8554"
      - "8555:8555/tcp"
      - "8555:8555/udp"
    environment:
      - LIBVA_DRIVER_NAME=radeonsi
      - HSA_OVERRIDE_GFX_VERSION=11.0.0

Relevant Frigate log output

docker exec -it frigate /opt/rocm/bin/rocminfo
ROCk module version 6.12.12 is loaded
=====================
HSA System Attributes
=====================
Runtime Version:         1.14
Runtime Ext Version:     1.6
System Timestamp Freq.:  1000.000000MHz
Sig. Max Wait Duration:  18446744073709551615 (0xFFFFFFFFFFFFFFFF) (timestamp count)
Machine Model:           LARGE
System Endianness:       LITTLE
Mwaitx:                  DISABLED
DMAbuf Support:          YES

==========
HSA Agents
==========
*******
Agent 1
*******
  Name:                    11th Gen Intel(R) Core(TM) i9-11900K @ 3.50GHz
  Uuid:                    CPU-XX
  Marketing Name:          11th Gen Intel(R) Core(TM) i9-11900K @ 3.50GHz
  Vendor Name:             CPU
  Feature:                 None specified
  Profile:                 FULL_PROFILE
  Float Round Mode:        NEAR
  Max Queue Number:        0(0x0)
  Queue Min Size:          0(0x0)
  Queue Max Size:          0(0x0)
  Queue Type:              MULTI
  Node:                    0
  Device Type:             CPU
  Cache Info:
    L1:                      49152(0xc000) KB
  Chip ID:                 0(0x0)
  ASIC Revision:           0(0x0)
  Cacheline Size:          64(0x40)
  Max Clock Freq. (MHz):   5100
  BDFID:                   0
  Internal Node ID:        0
  Compute Unit:            16
  SIMDs per CU:            0
  Shader Engines:          0
  Shader Arrs. per Eng.:   0
  WatchPts on Addr. Ranges:1
  Memory Properties:
  Features:                None
  Pool Info:
    Pool 1
      Segment:                 GLOBAL; FLAGS: FINE GRAINED
      Size:                    32713424(0x1f32ad0) KB
      Allocatable:             TRUE
      Alloc Granule:           4KB
      Alloc Recommended Granule:4KB
      Alloc Alignment:         4KB
      Accessible by all:       TRUE
    Pool 2
      Segment:                 GLOBAL; FLAGS: EXTENDED FINE GRAINED
      Size:                    32713424(0x1f32ad0) KB
      Allocatable:             TRUE
      Alloc Granule:           4KB
      Alloc Recommended Granule:4KB
      Alloc Alignment:         4KB
      Accessible by all:       TRUE
    Pool 3
      Segment:                 GLOBAL; FLAGS: KERNARG, FINE GRAINED
      Size:                    32713424(0x1f32ad0) KB
      Allocatable:             TRUE
      Alloc Granule:           4KB
      Alloc Recommended Granule:4KB
      Alloc Alignment:         4KB
      Accessible by all:       TRUE
    Pool 4
      Segment:                 GLOBAL; FLAGS: COARSE GRAINED
      Size:                    32713424(0x1f32ad0) KB
      Allocatable:             TRUE
      Alloc Granule:           4KB
      Alloc Recommended Granule:4KB
      Alloc Alignment:         4KB
      Accessible by all:       TRUE
  ISA Info:
*******
Agent 2
*******
  Name:                    gfx1100
  Uuid:                    GPU-73a8e305e19435a4
  Marketing Name:          AMD Radeon Graphics
  Vendor Name:             AMD
  Feature:                 KERNEL_DISPATCH
  Profile:                 BASE_PROFILE
  Float Round Mode:        NEAR
  Max Queue Number:        128(0x80)
  Queue Min Size:          64(0x40)
  Queue Max Size:          131072(0x20000)
  Queue Type:              MULTI
  Node:                    1
  Device Type:             GPU
  Cache Info:
    L1:                      32(0x20) KB
    L2:                      6144(0x1800) KB
    L3:                      98304(0x18000) KB
  Chip ID:                 29772(0x744c)
  ASIC Revision:           0(0x0)
  Cacheline Size:          128(0x80)
  Max Clock Freq. (MHz):   2304
  BDFID:                   768
  Internal Node ID:        1
  Compute Unit:            96
  SIMDs per CU:            2
  Shader Engines:          6
  Shader Arrs. per Eng.:   2
  WatchPts on Addr. Ranges:4
  Coherent Host Access:    FALSE
  Memory Properties:
  Features:                KERNEL_DISPATCH
  Fast F16 Operation:      TRUE
  Wavefront Size:          32(0x20)
  Workgroup Max Size:      1024(0x400)
  Workgroup Max Size per Dimension:
    x                        1024(0x400)
    y                        1024(0x400)
    z                        1024(0x400)
  Max Waves Per CU:        32(0x20)
  Max Work-item Per CU:    1024(0x400)
  Grid Max Size:           4294967295(0xffffffff)
  Grid Max Size per Dimension:
    x                        4294967295(0xffffffff)
    y                        4294967295(0xffffffff)
    z                        4294967295(0xffffffff)
  Max fbarriers/Workgrp:   32
  Packet Processor uCode:: 542
  SDMA engine uCode::      24
  IOMMU Support::          None
  Pool Info:
    Pool 1
      Segment:                 GLOBAL; FLAGS: COARSE GRAINED
      Size:                    25149440(0x17fc000) KB
      Allocatable:             TRUE
      Alloc Granule:           4KB
      Alloc Recommended Granule:2048KB
      Alloc Alignment:         4KB
      Accessible by all:       FALSE
    Pool 2
      Segment:                 GLOBAL; FLAGS: EXTENDED FINE GRAINED
      Size:                    25149440(0x17fc000) KB
      Allocatable:             TRUE
      Alloc Granule:           4KB
      Alloc Recommended Granule:2048KB
      Alloc Alignment:         4KB
      Accessible by all:       FALSE
    Pool 3
      Segment:                 GROUP
      Size:                    64(0x40) KB
      Allocatable:             FALSE
      Alloc Granule:           0KB
      Alloc Recommended Granule:0KB
      Alloc Alignment:         0KB
      Accessible by all:       FALSE
  ISA Info:
    ISA 1
      Name:                    amdgcn-amd-amdhsa--gfx1100
      Machine Models:          HSA_MACHINE_MODEL_LARGE
      Profiles:                HSA_PROFILE_BASE
      Default Rounding Mode:   NEAR
      Default Rounding Mode:   NEAR
      Fast f16:                TRUE
      Workgroup Max Size:      1024(0x400)
      Workgroup Max Size per Dimension:
        x                        1024(0x400)
        y                        1024(0x400)
        z                        1024(0x400)
      Grid Max Size:           4294967295(0xffffffff)
      Grid Max Size per Dimension:
        x                        4294967295(0xffffffff)
        y                        4294967295(0xffffffff)
        z                        4294967295(0xffffffff)
      FBarrier Max Size:       32
*** Done ***

Relevant go2rtc log output

frigate  | 2025-08-19 16:54:20.827531052  [2025-08-19 16:54:20] frigate.app                    INFO    : go2rtc process pid: 127

FFprobe output from your camera

bash: ffprobe: command not found

Install method

Docker Compose

Object Detector

Other

Network connection

Wireless

Camera make and model

TAPO c210

Screenshots of the Frigate UI's System metrics pages

Any other information that may be helpful

No response

[inkeep-ai-assistant[bot]]

(1), (2), (3), (4)

Your AMD RX 7900 XTX GPU is running at 100% utilization because you have multiple GPU-accelerated features enabled simultaneously. Your configuration shows several GPU-intensive components running on ROCm:

1. ONNX detector with RF-DETR model - This is a complex DETR-based model that's computationally expensive(1)
2. Semantic search with large model - This runs a large AI model locally and provides significant performance increase with a GPU(3)
3. Face recognition with large model - Another AI model running on the GPU(4)

The RF-DETR model you're using is particularly resource-intensive. The documentation warns that "Due to the size and complexity of the RF-DETR model, it is only recommended to be run with discrete Arc Graphics Cards"(1), though it should work on your RX 7900 XTX.

Your ROCm setup appears correct - the rocminfo output shows your GPU is properly detected as gfx1100 and your HSA_OVERRIDE_GFX_VERSION=11.0.0 environment variable matches the automatic detection for gfx1103 -> 11.0.0(2).

To reduce GPU usage, try these approaches:

1. Switch to a lighter model: Replace RF-DETR with YOLO-NAS or YOLOv9, which are less computationally demanding(1)

2. Reduce semantic search model size: Change from large to small in your semantic search configuration(3):

semantic_search:
  enabled: true
  model_size: small

(3)

3. Reduce face recognition model size: Change from large to small(4):

face_recognition:
  model_size: small

(4)

The combination of these three GPU-accelerated AI models running simultaneously is likely causing the sustained 100% GPU utilization you're experiencing.

📚 Sources:

1. Supported Hardware
2. Supported Hardware
3. Semantic Search
4. Full Reference Config

---

Have another question? Just tag @inkeep.

[Chreece]

@inkeep i have disabled semantic_search, face_recognition etc. still GPU on 100%

[NickM-27]

I don't think the AMD GPU supports rfdetr

[NickM-27]

I would suggest starting with the recommended yolov9 tiny

[Chreece]

Thank you for your suggestion, I'm here to try everything (literally)
I switched to yolov9-t infernce speed is still high at 165ms when yolov9-e was 210+ and rf-detr under 50 if i recall right.
But still GPU is stuck on 100%

[Chreece]

Just tested the tiny model again, after something was detected. Inference speed drops to 7ms but GPU is still 100%

[NickM-27]

some users have reported a bug in radeontop that reports 100% GPU usage even though it isn't actually, that could also be an issue

[Chreece]

I'm using rocm-smi and glances, both report 100% gpu. Also power consumption is 150W when 15W is idle (when ollama and whisper have loaded the models in gpu and frigate isn't running). Also the case temperature goes high and the air comming out is much warmer.
I thought also it was just a false report but everything seems to go up when frigate is up

[inkeep-ai-assistant[bot]]

(1), (2), (3), (4)

The issue is likely with your RF-DETR model configuration. Based on the available documentation, there are several potential causes for continued 100% GPU usage:

1. FFmpeg hardware acceleration: You still have preset-vaapi enabled in your ffmpeg configuration (1). This preset is designed for Intel & AMD VAAPI and requires checking that the correct driver is chosen (1). Try temporarily removing this:

ffmpeg:
  # Remove or comment out the hwaccel_args line
  # hwaccel_args: preset-vaapi

(1)

2. ONNX detector configuration: Your detector is set to use ONNX with a custom RF-DETR model (2). The detector configuration shows:

detectors:
  onnx:
    type: onnx

(2)

Try switching to CPU detection temporarily to isolate the issue:

detectors:
  cpu:
    type: cpu

(2)

3. Model configuration: Your model settings specify a custom path and configuration (2):

model:
  model_type: rfdetr
  width: 320
  height: 320
  input_tensor: nchw
  input_dtype: float
  path: /config/rfdetr-Medium.onnx

(2)

The documentation shows the default model configuration uses different settings (2). Try using the default EdgeTPU model temporarily to see if the GPU usage drops.

To troubleshoot systematically, disable hardware acceleration components one by one:

1. Remove hwaccel_args: preset-vaapi from ffmpeg (1)
2. Switch detector from onnx to cpu (2)
3. Use default model settings instead of the custom RF-DETR model (2)

This will help identify which component is causing the sustained GPU load.

📚 Sources:

1. FFmpeg presets
2. Full Reference Config
3. Frequently Asked Questions
4. Full Reference Config

---

Have another question? Just tag @inkeep.