How to install Intel GPU drivers

General Discussion & Chat
1

Hello, I use AlmaLinux 10 as my server, and I have Intel i7 1360p with 96 Xe GPU cores, I would like to use these cores for AI.

I created Ollama docker, I installed some models, they run fine, but they run on CPU, so they are little bit slow, I would like to offload them to GPU cores to be little bit faster.

But I cannot found intel opencl package to install, AI said to install these packages but they are not avaiable:

[root@telekom intel-drivers]# sudo dnf install intel-opencl-rt intel-level-zero-gpu intel-media-driver intel-mediasdk
Last metadata expiration check: 0:04:50 ago on Fri Feb 20 15:49:08 2026.
No match for argument: intel-opencl-rt
No match for argument: intel-level-zero-gpu
No match for argument: intel-media-driver
No match for argument: intel-mediasdk

and I cannot find any docs about this in google, please which packages to install so Ollama will see my GPU cores?

If someone is interested this is my docker compose:


services:
  ollama:
    image: ollama/ollama:latest
    container_name: ollama
    # Na Fedore/Alme je privileged: true najistejšia cesta, ako obísť SELinux pre GPU
    privileged: true
    ports:
          - "11434:11434"
    devices:
      - /dev/dri:/dev/dri
    environment:
      - OLLAMA_HOST=0.0.0.0
      - OLLAMA_INTEL_GPU=1
      - OLLAMA_VULKAN=1
      - ONEAPI_DEVICE_SELECTOR=level_zero:0
      # Povieme Ollame, kde má hľadať tie nové Intel knižnice
      - LD_LIBRARY_PATH=/opt/intel/oneapi/redist/lib/intel64:/usr/lib64
    group_add:
      - "39"
      - "105"
    volumes:
      - ./ollama_data:/root/.ollama:Z  # ":Z" povie SELinuxu, že k tomuto volume má mať Docker prístup
      - /usr/lib64:/usr/lib64:ro
      - /usr/share/vulkan:/usr/share/vulkan:ro
      - /etc/OpenCL/vendors:/etc/OpenCL/vendors:ro
      - /opt/intel/oneapi:/opt/intel/oneapi:ro
    restart: unless-stopped

  open-webui:
    image: ghcr.io/open-webui/open-webui:main
    container_name: open-webui
    ports:
      - "3000:8080"
    environment:
      - OLLAMA_BASE_URL=http://ollama:11434
    volumes:
      - ./open_webui_data:/app/backend/data:Z # Opäť ":Z" kvôli SELinuxu
    depends_on:
      - ollama
    restart: unless-stopped
2

HI, I have the same problem, no binaries in repos, every single git release only for debian.. Intels, repos are down / blocked ? 403 or 404; Even conda does not fix the problem; Anyone tried to install gpu drivers from intel for SYCL ?

3

hello

The standard repo for AlmaLinux 10 does not include intel-opencl-rt, but EPEL 10.1 includes intel-compute-runtime, which has intel-opencl / intel-level-zero / intel-ocloc / intel-igc as dependencies. Therefore, if you enable EPEL on AlmaLinux 10 and run dnf install intel-compute-runtime, you should be able to install the required user space runtime set, including dependencies.

4

this helped, thank you, I managed to install intel drivers, but still Ollama is using CPU, and not GPU :confused:

5

Hi, glad intel-compute-runtime installed. Note that this provides OpenCL/Level Zero, but Ollama’s Intel GPU support on Linux is via Vulkan, not OpenCL/Level Zero.

Can you confirm Vulkan works on the host and inside the container?

Host: vulkaninfo --summary (or vulkaninfo | head) shows an Intel GPU device

Container: /dev/dri/renderD* is present and vulkaninfo --summary works

Also check /usr/share/vulkan/icd.d/ has an Intel ICD JSON (Mesa ANV), otherwise Ollama will fall back to CPU.

If you paste the output of ls -l /dev/dri, vulkaninfo --summary, and the Ollama server log (with debug if possible), we can see why it’s not selecting Vulkan.