[Beta Support]: Battlemage GPU

[tassadar81]

Checklist

• I have updated to the latest available Frigate version.
• I have cleared the cache of my browser.
• I have tried a different browser to see if it is related to my browser.
• I have tried reproducing the issue in incognito mode to rule out problems with any third party extensions or plugins I have installed.
• I have asked the AI at https://docs.frigate.video about my issue.

Describe the problem you are having

Segmentation fault (core dumped) occurs when attempting to use hardware acceleration on the new Intel Arc B580 (Battlemage / Xe2 architecture).

The issue stems from a mismatch between the hardware requirements and the media stack provided within the container. The Arc B580 (Device ID e20b) requires Intel's Unified Driver version 25.x or newer for stable operation. The current Frigate image (based on Debian) provides Intel iHD driver 24.3.4, which results in a fatal memory access error (Segfault) when initializing HEVC decoding surfaces for Battlemage GPUs.

Attempts to map host libraries (Ubuntu 24.04 / Noble) into the container to provide the 25.x driver failed due to GLIBC_2.38 version requirements not being met by the container's base OS.

Steps to reproduce

1. Run Frigate on a host with an Intel Arc B580 (Battlemage).

2. Configure hwaccel using vaapi or qsv pointing to the discrete GPU (/dev/dri/renderD128).

3. Observe the crash or run the following test command inside the container:
   /usr/lib/ffmpeg/7.0/bin/ffmpeg -hide_banner -hwaccel vaapi -hwaccel_device /dev/dri/renderD128 -hwaccel_output_format vaapi -reorder_queue_size 1024 -i rtsp://[USER]:[PASS]@[CAMERA_IP] -c:v hevc_vaapi -f null -

4. FFmpeg will log Segmentation fault (core dumped) immediately after stream mapping.

Version

0.17.0-beta2-tensorrt

In which browser(s) are you experiencing the issue with?

No response

Frigate config file

mqtt:
  host: [REDACTED]
  user: [REDACTED]
  password: [REDACTED]

detectors:
  onnx:
    type: onnx
    device: gpu # Nvidia 5060 Ti

model:
  path: /config/model_cache/yolov9-e-640.onnx
  width: 640
  height: 640

ffmpeg:
  path: /usr/lib/ffmpeg/7.0
  hwaccel_args:
    - -hwaccel
    - vaapi
    - -hwaccel_device
    - /dev/dri/renderD128
    - -hwaccel_output_format
    - vaapi
  input_args:
    - -fflags
    - +genpts+discardcorrupt
    - -reorder_queue_size
    - "1024"
    - -rtsp_transport
    - tcp

cameras:
  camera_test:
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/stream_test
          roles: [detect, record]
    detect:
      width: 3840
      height: 2160
      fps: 5

docker-compose file or Docker CLI command

podman run -d \
  --name frigate \
  --memory=48g \
  --shm-size=30g \
  --security-opt label=disable \
  --device nvidia.com/gpu=all \
  --device /dev/dri/renderD128:/dev/dri/renderD128 \
  -v /home/podman/frigate/config.yml:/config/config.yml \
  -v /home/podman/frigate/model_cache:/config/model_cache \
  -v /mnt/frigate/db:/config/db \
  -v /mnt/frigate/recordings:/media/frigate \
  -e LIBVA_DRIVER_NAME=iHD \
  ghcr.io/blakeblackshear/frigate:0.17.0-beta2-tensorrt

Relevant Frigate log output

[hevc_vaapi @ 0x5a1cbebabdc0] No quality level set; using default (25).
Segmentation fault (core dumped)

Relevant go2rtc log output

Output of vainfo inside the container:

libva info: VA-API version 1.22.0
libva info: User environment variable requested driver 'iHD'
libva info: Trying to open /usr/lib/x86_64-linux-gnu/dri/iHD_drv_video.so
libva info: Found init function __vaDriverInit_1_22
libva info: va_openDriver() returns 0
vainfo: VA-API version: 1.22 (libva 2.12.0)
vainfo: Driver version: Intel iHD driver for Intel(R) Gen Graphics - 24.3.4 ()

Operating system

Other Linux

Install method

Docker CLI

Network connection

Wired

Camera make and model

reolink

Screenshots of the Frigate UI's System metrics pages

irrilevant, the problem is in the image as it have been built

Any other information that may be helpful

on host, ubuntu 24.04, i installed intel software following the ufficial steps described here:

https://dgpu-docs.intel.com/driver/client/overview.html

[inkeep-ai-assistant[bot]]

---

The Intel Arc B580 (Battlemage/Xe2) requires newer drivers than what's currently included in the Frigate container image. The container ships with Intel iHD driver 24.3.4(1), but Battlemage GPUs need version 25.x or newer for stable operation.

For newer Intel GPUs (gen13+ and Intel Arc), the recommended preset is preset-intel-qsv-*(1). However, since you're experiencing segmentation faults with the current driver version, you have a few options:

Option 1: Use FFmpeg 7.0 (Already Configured)

You're already using FFmpeg 7.0 in your config, which is good. The documentation notes that upgrading to FFmpeg 7 can resolve certain hardware acceleration issues(1).

Option 2: Try QSV Instead of VAAPI

For Intel Arc GPUs, switch from VAAPI to QSV presets(1):

ffmpeg:
  path: "7.0"
  hwaccel_args: preset-intel-qsv-h265

(1), (2)

Option 3: Provide Custom FFmpeg Build

The documentation mentions you can provide your own FFmpeg build that works with your driver version, though this is unsupported(1). You would need to:

1. Build FFmpeg with the newer Intel drivers
2. Mount it into the container using the ffmpeg.path configuration option(3)

Option 4: Wait for Container Update

Since this is a driver compatibility issue with very new hardware (Arc B580), you may need to wait for the Frigate container image to be updated with newer Intel drivers that support Battlemage architecture.

Temporary Workaround

If none of the above work, you could disable hardware acceleration temporarily and use CPU decoding until driver support is added to the container.

📚 Sources:

1. Video Decoding
2. FFmpeg presets
3. Full Reference Config

---

Have another question? Just tag @inkeep.

[hawkeye217]

I'd suggest using QSV, not VAAPI. Use the preset rather than manual args, eg:

hwaccel_args: preset-intel-qsv-h264

As a side note, your config is also invalid and suboptimal in a number of ways:

1. Your model section is missing a number of required keys. https://docs.frigate.video/configuration/object_detectors/#supported-models-2
2. Your detect resolution is extremely high and will cause excessive resource (CPU and GPU) usage. https://docs.frigate.video/frigate/camera_setup/#choosing-a-detect-resolution

[tassadar81]

I tried qsv but I had problems as well, guess for old drivers again.
Anyawy is this the dockerfile to build the image of beta2 with tensorrt?

https://github.com/blakeblackshear/frigate/blob/v0.17.0-beta2/docker/tensorrt/Dockerfile.amd64

I'd try to update to more up to date base version if you can suggest what I've to consider to do the task.
BTW why you're still on debian12 as trixie is stable since a while? Why use debían as mostly Ubuntu is more supported by hardware vendors?

[vcdx71]

Nice, thanks for the link. I'll do that as well.

[tassadar81]

Hello Guys,

still testing but it seems that i got it working building the image with this Containerfile:

FROM ghcr.io/blakeblackshear/frigate:0.17.0-beta2-tensorrt

ENV DEBIAN_FRONTEND=noninteractive
USER root

RUN echo "deb http://deb.debian.org/debian testing main non-free non-free-firmware" > /etc/apt/sources.list.d/testing.list

RUN printf '%s\n'
'# Default: preferisci stable'
'Package: *'
'Pin: release a=stable'
'Pin-Priority: 900'
''
'Package: *'
'Pin: release a=testing'
'Pin-Priority: 100'
''
'# Pacchetti Intel VAAPI/media da testing'
'Package: intel-media-va-driver-non-free libva2 libva-drm2 libva-wayland2 libva-x11-2 libva-glx2 vainfo'
'Pin: release a=testing'
'Pin-Priority: 990'
''
'# Librerie MFX/VPL per encoding/decoding'
'Package: libmfx1 libmfx-gen1.2 libvpl2'
'Pin: release a=testing'
'Pin-Priority: 990'
''
'# Dipendenze che potrebbero essere necessarie da testing'
'Package: libigdgmm12 libzstd1'
'Pin: release a=testing'
'Pin-Priority: 990'
''
'# libc6 e libstdc++ - aggiorna se necessario'
'Package: libc6 libstdc++6'
'Pin: release a=testing'
'Pin-Priority: 500'
> /etc/apt/preferences.d/intel-testing

RUN apt-get update &&
apt-get remove -y libmfxgen1 2>/dev/null || true &&
apt-get install -y --no-install-recommends -t testing
intel-media-va-driver-non-free
libva2
libva-drm2
libva-glx2
vainfo &&
apt-get clean &&
rm -rf /var/lib/apt/lists/* /tmp/* /var/tmp/*

ENV LIBVA_DRIVER_NAME=iHD
ENV ONEVPL_PRIORITY_PATH=/usr/lib/x86_64-linux-gnu

EXPOSE 5000 8554 8555 8971

[tassadar81]

unfortunately it seems that something is not properly workiing on dmesg i'm getting a neverening

[ 634.392494] xe 0000:06:00.0: [drm] GT1: trying reset from guc_exec_queue_timedout_job [xe]
[ 634.392543] xe 0000:06:00.0: [drm] GT1: trying reset from guc_exec_queue_timedout_job [xe]

[fl8s]

@tassadar81 I went through this just days ago trying to use my B580 for a different ML workload, and the segfault along with the guc_exec_queue_timedout_job that xe is throwing looks very familiar. Does clinfo cause the same segfault? If so, it's probably the firmware on the card. To confirm this, things to check are the following, most of which you've already checked, and in the end, it wound up being the firmware on the card which needed to be updated, and the only way that I was able to resolve that was to update it using a windows machine.

I researched other methods which you can find here, and intel's open source firmware updating tool is here but I deemed it too risky and complicated to do so, and in the end wound up using a live windows 11 environment and the intel driver installation tool, rather than figuring out how igsc worked. Here is the order I did things to confirm that it was in fact the firmware on the card, since I know nobody wants to go through all that for nothing.

• 1. Kernel version (6.11 or better)
     uname -r
• 2. GPU detection
     lspci -nn | grep -i "VGA|Display"
• 3. Kernel driver (xe confirmed in your case)
     lsmod | grep -E "^xe |^i915 "
• 4. Render devices
     ls -la /dev/dri/
• 5. PCIe link status (check for x8 vs x1, mine said it was at 1x which was suspect)
     GPU_BUS=$(lspci -nn | grep -i intel | grep -i "VGA|Display" | head -1 | cut -d' ' -f1)
     sudo lspci -vvv -s "$GPU_BUS" 2>/dev/null | grep -E "LnkCap:|LnkSta:"
• 6. On the host or from a container with the card passed in, does clinfo cause the same error, or a segfault?
     if so, then update the firmware.

Once the firmware is updated, the link status will be 8x, and clinfo will return with the correct information. hope that helps.
The driver which updates the firmware is here

[coledeck]

@fl8s I'm still running into this issue after updating the the latest firmware from linux with igsc (FW 1174, which is the same found in the 101.8425 driver you linked). It also shows link rate 1x:

LnkCap: Port #0, Speed 2.5GT/s, Width x1, ASPM L0s L1, Exit Latency L0s <64ns, L1 <1us
LnkSta: Speed 2.5GT/s, Width x1

My server only supports pcie gen 3 but it should get a Gen3x8 link. clinfo shows GPU info, no crash or segfault. Resize bar enabled and working. Any ideas?