Intel AI Boost NPU is available and working from Windows with OpenVINO, but it is not exposed to WSL2 or Docker Desktop Linux containers in a way that Linux OpenVINO can use.
On this machine, Windows OpenVINO can see and run inference on NPU, but inside WSL2 / Docker Desktop:
/dev/dxg exists/usr/lib/wsl/drivers contains Intel NPU driver files and OpenVINO wrapper files/dev/accel/accel0 does not exist/dev/dri does not existmodprobe ivpu fails because the module is not presentCONFIG_DRM_ACCEL is not enabled in the WSL kernel- Docker direct device mapping fails:
--device=/dev/accel/accel0 -> no such file or directory
- Linux OpenVINO 2026.2.1 inside Docker with
/dev/dxg and /usr/lib/wsl mounted reports only:
ov.Core().available_devices == ["CPU"]
This prevents containerized OpenVINO workloads from using the Intel NPU.
Describe the solution you'd like
Please add supported Intel AI Boost NPU exposure for WSL2 and Docker Desktop Linux containers.
Ideally WSL2 would expose the NPU through the native Linux accelerator device path:
Expected container-visible device path:
/dev/accel/accel0
Expected Docker Compose mapping:
devices:
- "/dev/accel/accel0:/dev/accel/accel0"
Expected OpenVINO result inside WSL2 or Docker container:
import openvino as ov
print(ov.Core().available_devices)
Expected: ["CPU", "NPU"] or similar
Describe alternatives you've considered
I built a Windows host-side OpenVINO bridge that Docker containers can call through host.docker.internal. This works, but it adds host/container boundary complexity and is not ideal for fully containerized AI workloads.
I also tested Docker --device=/dev/dxg with /usr/lib/wsl mounted, but Linux OpenVINO still reports only CPU.
Running the workload on native Linux may work, but the goal is to support Windows AI PCs using WSL2 and Docker Desktop.
Additional context
Use case: local AI / OpenVINO workloads in Docker containers using Intel AI Boost NPU for embeddings, perception, classifiers, and background inference.
Environment observed:
Windows 11
WSL2 kernel: 6.6.87.2-microsoft-standard-WSL2
Docker Desktop backend: WSL2
OpenVINO: 2026.2.1
Hardware: Intel AI Boost NPU
This would be valuable for AI PC developers who want reproducible Linux container workflows while using the onboard NPU.
Intel AI Boost NPU is available and working from Windows with OpenVINO, but it is not exposed to WSL2 or Docker Desktop Linux containers in a way that Linux OpenVINO can use.
On this machine, Windows OpenVINO can see and run inference on
NPU, but inside WSL2 / Docker Desktop:/dev/dxgexists/usr/lib/wsl/driverscontains Intel NPU driver files and OpenVINO wrapper files/dev/accel/accel0does not exist/dev/dridoes not existmodprobe ivpufails because the module is not presentCONFIG_DRM_ACCELis not enabled in the WSL kernel--device=/dev/accel/accel0->no such file or directory/dev/dxgand/usr/lib/wslmounted reports only:ov.Core().available_devices == ["CPU"]This prevents containerized OpenVINO workloads from using the Intel NPU.
Describe the solution you'd like
Please add supported Intel AI Boost NPU exposure for WSL2 and Docker Desktop Linux containers.
Ideally WSL2 would expose the NPU through the native Linux accelerator device path:
Expected container-visible device path:
/dev/accel/accel0
Expected Docker Compose mapping:
devices:
Expected OpenVINO result inside WSL2 or Docker container:
import openvino as ov
print(ov.Core().available_devices)
Expected: ["CPU", "NPU"] or similar
Describe alternatives you've considered
I built a Windows host-side OpenVINO bridge that Docker containers can call through host.docker.internal. This works, but it adds host/container boundary complexity and is not ideal for fully containerized AI workloads.
I also tested Docker --device=/dev/dxg with /usr/lib/wsl mounted, but Linux OpenVINO still reports only CPU.
Running the workload on native Linux may work, but the goal is to support Windows AI PCs using WSL2 and Docker Desktop.
Additional context
Use case: local AI / OpenVINO workloads in Docker containers using Intel AI Boost NPU for embeddings, perception, classifiers, and background inference.
Environment observed:
Windows 11
WSL2 kernel: 6.6.87.2-microsoft-standard-WSL2
Docker Desktop backend: WSL2
OpenVINO: 2026.2.1
Hardware: Intel AI Boost NPU
This would be valuable for AI PC developers who want reproducible Linux container workflows while using the onboard NPU.