Arc A380 not detected in Docker on Linux, fallback to CPU, why?

[PieterV]

Hi, new to OpenVINO, trying to get hardware ML working with my ARC A380 on Proxmox 8 and Docker 29.

I first noticed the problem with the GPU not being detected when I tried out Immich.

To help troubleshoot I am now trying to get OpenVINO to work following these instructions.

I get the same error: Available devices for Open VINO: CPU

$ docker run --rm -it  --device=/dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) -v ${PWD}/model:/opt/model -p 9123:9123 openvino/model_server:latest-gpu --model_path /opt/model --model_name resnet --port 9123 --tar
get_device GPU
[2025-11-30 16:40:39.639][1][serving][info][server.cpp:91] OpenVINO Model Server 2025.3.0.6e2e910de
[2025-11-30 16:40:39.639][1][serving][info][server.cpp:92] OpenVINO backend 2025.3.0.dev20250826
[2025-11-30 16:40:39.639][1][serving][info][pythoninterpretermodule.cpp:37] PythonInterpreterModule starting
Python version:
3.12.3 (main, Aug 14 2025, 17:47:21) [GCC 13.3.0]
Python sys.path output:
['', '/ovms/lib/python', '/ovms/python_deps', '/lib/python312.zip', '/lib/python3.12', '/lib/python3.12/lib-dynload']
[2025-11-30 16:40:39.710][1][serving][info][pythoninterpretermodule.cpp:50] PythonInterpreterModule started
[2025-11-30 16:40:39.980][1][modelmanager][info][modelmanager.cpp:156] Available devices for Open VINO: CPU
[2025-11-30 16:40:39.981][1][serving][info][capimodule.cpp:40] C-APIModule starting
[2025-11-30 16:40:39.981][1][serving][info][capimodule.cpp:42] C-APIModule started
[2025-11-30 16:40:39.981][1][serving][info][grpcservermodule.cpp:102] GRPCServerModule starting
[2025-11-30 16:40:39.982][1][serving][info][grpcservermodule.cpp:179] GRPCServerModule started
[2025-11-30 16:40:39.982][1][serving][info][grpcservermodule.cpp:180] Started gRPC server on port 9123
[2025-11-30 16:40:39.982][1][serving][info][servablemanagermodule.cpp:51] ServableManagerModule starting
[2025-11-30 16:40:39.983][1][serving][info][model.cpp:42] Getting model from /opt/model
[2025-11-30 16:40:39.983][1][serving][info][model.cpp:49] Model downloaded to /opt/model
[2025-11-30 16:40:39.983][1][serving][info][model.cpp:149] Will add model: resnet; version: 1 ...
[2025-11-30 16:40:39.983][1][serving][info][modelversionstatus.cpp:109] STATUS CHANGE: Version 1 of model resnet status change. New status: ( "state": "START", "error_code": "OK" )
[2025-11-30 16:40:39.983][1][serving][info][modelinstance.cpp:1058] Loading model: resnet, version: 1, from path: /opt/model/1, with target device: GPU ...
[2025-11-30 16:40:39.983][1][serving][info][modelversionstatus.cpp:109] STATUS CHANGE: Version 1 of model resnet status change. New status: ( "state": "START", "error_code": "OK" )
[2025-11-30 16:40:39.983][1][serving][info][modelversionstatus.cpp:109] STATUS CHANGE: Version 1 of model resnet status change. New status: ( "state": "LOADING", "error_code": "OK" )
[2025-11-30 16:40:40.254][1][modelmanager][info][modelinstance.cpp:593] Input name: input_1; mapping_name: input_1; shape: (-1,224,224,3); precision: FP32; layout: N...
[2025-11-30 16:40:40.254][1][modelmanager][info][modelinstance.cpp:656] Output name: activation_49; mapping_name: activation_49; shape: (-1,1001); precision: FP32; layout: N...
[2025-11-30 16:40:40.255][1][modelmanager][error][modelinstance.cpp:842] Cannot compile model into target device; error: Exception from src/inference/src/cpp/core.cpp:109:
Check '!m_device_map.empty()' failed at src/plugins/intel_gpu/src/plugin/plugin.cpp:518:
[GPU] Can't get PERFORMANCE_HINT property as no supported devices found or an error happened during devices query.
[GPU] Please check OpenVINO documentation for GPU drivers setup guide.


; model: resnet; version: 1; device: GPU
[2025-11-30 16:40:40.255][1][serving][info][modelversionstatus.cpp:109] STATUS CHANGE: Version 1 of model resnet status change. New status: ( "state": "LOADING", "error_code": "UNKNOWN" )
[2025-11-30 16:40:40.255][1][serving][error][model.cpp:154] Error occurred while loading model: resnet; version: 1; error: Cannot compile model into target device
[2025-11-30 16:40:40.255][1][modelmanager][error][modelmanager.cpp:1483] Error occurred while loading model: resnet versions; error: Cannot compile model into target device
[2025-11-30 16:40:40.255][1][serving][info][model.cpp:197] Will clean up model: resnet; version: 1 ...
[2025-11-30 16:40:40.255][1][serving][info][model.cpp:89] Updating default version for model: resnet, from: 0
[2025-11-30 16:40:40.255][1][serving][info][model.cpp:101] Model: resnet will not have default version since no version is available.
[2025-11-30 16:40:40.255][1][serving][info][modelversionstatus.cpp:109] STATUS CHANGE: Version 1 of model resnet status change. New status: ( "state": "LOADING", "error_code": "UNKNOWN" )
[2025-11-30 16:40:40.275][1][modelmanager][error][modelmanager.cpp:184] Couldn't start model manager
[2025-11-30 16:40:40.275][1][serving][error][servablemanagermodule.cpp:58] ovms::ModelManager::Start() Error: Cannot compile model into target device
[2025-11-30 16:40:40.275][1][serving][info][grpcservermodule.cpp:188] GRPCServerModule shutting down
[2025-11-30 16:40:40.275][1][serving][info][grpcservermodule.cpp:193] Shutdown gRPC server
[2025-11-30 16:40:40.275][1][serving][info][grpcservermodule.cpp:198] GRPCServerModule shutdown
[2025-11-30 16:40:40.275][1][serving][info][servablemanagermodule.cpp:65] ServableManagerModule shutting down
[2025-11-30 16:40:40.275][1][serving][info][modelversionstatus.cpp:109] STATUS CHANGE: Version 1 of model resnet status change. New status: ( "state": "UNLOADING", "error_code": "OK" )
[2025-11-30 16:40:40.276][1][serving][info][modelversionstatus.cpp:109] STATUS CHANGE: Version 1 of model resnet status change. New status: ( "state": "END", "error_code": "OK" )
[2025-11-30 16:40:40.276][1][serving][info][servablemanagermodule.cpp:71] ServableManagerModule shutdown
[2025-11-30 16:40:40.276][1][serving][info][pythoninterpretermodule.cpp:61] PythonInterpreterModule shutting down
[2025-11-30 16:40:40.276][1][serving][info][pythoninterpretermodule.cpp:65] PythonInterpreterModule shutdown
[2025-11-30 16:40:40.280][1][serving][info][capimodule.cpp:50] C-APIModule shutting down
[2025-11-30 16:40:40.280][1][serving][info][capimodule.cpp:52] C-APIModule shutdown

The ARC A380 does work for video transcoding, so it can be used from docker.

The system hardware is a Supermicro H13SSL-N with an AMD Epyc Genoa 9334, 384GB RAM, Sparkle Intel Arc A380 ELF 6GB GDDR6 Single Fan SA380E-6G GPU.

OS is Proxmox 8.4, kernel 6.8.12-16-pve, Docker version 29.0.3, build 511dad6, docker is installed directly on the host (no LXC, no VM).

GPU drivers installed are intel-opencl-icd and intel-gpu-tools.

$ uname -r
6.8.12-16-pve

$ lspci -nn | grep -Ei "3d|display|vga"
43:00.0 VGA compatible controller [0300]: Intel Corporation DG2 [Arc A380] [8086:56a5] (rev 05)
c6:00.0 VGA compatible controller [0300]: ASPEED Technology, Inc. ASPEED Graphics Family [1a03:2000] (rev 52)

$ ls -l /dev/dri
total 0
drwxr-xr-x 2 root root        100 Nov  2 07:38 by-path
crw-rw---- 1 root video  226,   0 Nov  2 07:38 card0
crw-rw---- 1 root video  226,   1 Nov  2 07:38 card1
crw-rw---- 1 root render 226, 128 Nov  2 07:38 renderD128

$ sudo intel_gpu_top
intel-gpu-top: Intel Dg2 (Gen12) @ /dev/dri/card1

Please help me troubleshoot why the GPU is not detected?

[Zulkifli_Intel]

Hi PieterV,

Thank you for reaching out to us.

Can you run these commands and share the output

lspci -nnk -s 43:00.0

lsmod | grep xe

clinfo | grep Device

Regards,

Zul

[PieterV]

Info as requested:

pieter@server-2:~$ sudo lspci -nnk -s 43:00.0
[sudo] password for pieter:
43:00.0 VGA compatible controller [0300]: Intel Corporation DG2 [Arc A380] [8086:56a5] (rev 05)
        Subsystem: Device [172f:3943]
        Kernel driver in use: i915
        Kernel modules: i915, xe
pieter@server-2:~$ sudo lsmod | grep xe
xe                   3567616  0
intel_vsec             24576  1 xe
gpu_sched              65536  1 xe
drm_gpuvm              49152  1 xe
drm_gpusvm_helper      36864  1 xe
drm_ttm_helper         16384  1 xe
drm_exec               12288  2 drm_gpuvm,xe
drm_suballoc_helper    16384  1 xe
drm_buddy              28672  2 xe,i915
ttm                   118784  3 drm_ttm_helper,xe,i915
drm_display_helper    274432  2 xe,i915
cec                    94208  3 drm_display_helper,xe,i915
video                  77824  2 xe,i915
i2c_algo_bit           16384  3 ast,xe,i915
pieter@server-2:~$ sudo clinfo | grep Device
sudo: clinfo: command not found

I also made a discovery; I read some posts claiming that a newer kernel may be required, so I upgraded Proxmox from v8 to v9, and the kernel from 6.8 to 6.17, and now it is working!

pieter@server-2:~$ docker run --rm -it  --device=/dev/dri --group-add=$(stat -c "%g" /dev/dri/render* | head -n 1) -u $(id -u):$(id -g) -v ${PWD}/model:/opt/model -p 9123:9123 openvino/model_server:latest-gpu --model_path /opt/model --model_name resnet --port 9123 --target_device GPU
[2025-12-01 19:55:19.292][1][serving][info][server.cpp:88] OpenVINO Model Server 2025.4.0.8db21b815
[2025-12-01 19:55:19.292][1][serving][info][server.cpp:89] OpenVINO backend 2025.4.0.0rc3
[2025-12-01 19:55:19.292][1][serving][info][pythoninterpretermodule.cpp:37] PythonInterpreterModule starting
Python version:
3.12.3 (main, Nov  6 2025, 13:44:16) [GCC 13.3.0]
Python sys.path output:
['', '/ovms/lib/python', '/ovms/python_deps', '/lib/python312.zip', '/lib/python3.12', '/lib/python3.12/lib-dynload']
[2025-12-01 19:55:19.362][1][serving][info][pythoninterpretermodule.cpp:50] PythonInterpreterModule started
[2025-12-01 19:55:19.617][1][modelmanager][info][modelmanager.cpp:156] Available devices for Open VINO: CPU, GPU
[2025-12-01 19:55:19.618][1][serving][info][capimodule.cpp:40] C-APIModule starting
[2025-12-01 19:55:19.618][1][serving][info][capimodule.cpp:42] C-APIModule started
[2025-12-01 19:55:19.618][1][serving][info][grpcservermodule.cpp:110] GRPCServerModule starting
[2025-12-01 19:55:19.618][1][serving][info][grpcservermodule.cpp:137] Binding gRPC server to address: 0.0.0.0:9123
[2025-12-01 19:55:19.620][1][serving][info][grpcservermodule.cpp:192] GRPCServerModule started
[2025-12-01 19:55:19.620][1][serving][info][grpcservermodule.cpp:193] Started gRPC server on port 9123
[2025-12-01 19:55:19.620][1][serving][info][servablemanagermodule.cpp:51] ServableManagerModule starting
[2025-12-01 19:55:19.620][1][serving][info][model.cpp:42] Getting model from /opt/model
[2025-12-01 19:55:19.620][1][serving][info][model.cpp:49] Model downloaded to /opt/model
[2025-12-01 19:55:19.620][1][serving][info][model.cpp:149] Will add model: resnet; version: 1 ...
[2025-12-01 19:55:19.620][1][serving][info][modelversionstatus.cpp:109] STATUS CHANGE: Version 1 of model resnet status change. New status: ( "state": "START", "error_code": "OK" )
[2025-12-01 19:55:19.620][1][serving][info][modelinstance.cpp:1058] Loading model: resnet, version: 1, from path: /opt/model/1, with target device: GPU ...
[2025-12-01 19:55:19.620][1][serving][info][modelversionstatus.cpp:109] STATUS CHANGE: Version 1 of model resnet status change. New status: ( "state": "START", "error_code": "OK" )
[2025-12-01 19:55:19.620][1][serving][info][modelversionstatus.cpp:109] STATUS CHANGE: Version 1 of model resnet status change. New status: ( "state": "LOADING", "error_code": "OK" )
[2025-12-01 19:55:20.170][1][modelmanager][info][modelinstance.cpp:593] Input name: input_1; mapping_name: input_1; shape: (-1,224,224,3); precision: FP32; layout: N...
[2025-12-01 19:55:20.170][1][modelmanager][info][modelinstance.cpp:656] Output name: activation_49; mapping_name: activation_49; shape: (-1,1001); precision: FP32; layout: N...
[2025-12-01 19:55:21.387][1][modelmanager][info][modelinstance.cpp:1363] Number of OpenVINO streams: 1
[2025-12-01 19:55:21.387][1][modelmanager][info][modelinstance.cpp:863] Plugin config for device: GPU
[2025-12-01 19:55:21.387][1][modelmanager][info][modelinstance.cpp:867] OVMS set plugin settings key: PERFORMANCE_HINT; value: LATENCY;
[2025-12-01 19:55:21.387][1][serving][info][modelinstance.cpp:930] Loaded model resnet; version: 1; batch size: -1; No of InferRequests: 1
[2025-12-01 19:55:21.387][1][serving][info][modelversionstatus.cpp:109] STATUS CHANGE: Version 1 of model resnet status change. New status: ( "state": "AVAILABLE", "error_code": "OK" )
[2025-12-01 19:55:21.387][1][serving][info][model.cpp:89] Updating default version for model: resnet, from: 0
[2025-12-01 19:55:21.387][1][serving][info][model.cpp:99] Updated default version for model: resnet, to: 1
[2025-12-01 19:55:21.387][187][modelmanager][info][modelmanager.cpp:1200] Started model manager thread
[2025-12-01 19:55:21.387][1][serving][info][servablemanagermodule.cpp:55] ServableManagerModule started
[2025-12-01 19:55:21.387][188][modelmanager][info][modelmanager.cpp:1219] Started cleaner thread

Is there a minimum kernel that is required?

[PieterV · Accepted solution]

Also working for Immich with the new kernel, so my kernel was too old.

[PieterV]

Or if not the kernel version, it could be that the OpenVino version in the test changed from "2025.3.0.dev20250826" to "2025.4.0.0rc3" between now and a day ago?

[Zulkifli_Intel]

Hi PieterV,

Thank you for confirming your issue has been resolved and for sharing the resolution with us. The previous configuration (Proxmox 8 and Kernel 6.8) does not work could be due to the Intel GPU driver (xe) being incomplete or unstable. Proxmox 9 and Kernel 6.17 have major fixes and include significant updates to Intel's open-source graphics drivers, including the xe driver.

Regards,

Zul

[Zulkifli_Intel]

This thread will no longer be monitored since this issue has been resolved. If you need any additional information from Intel, please submit a new question.