Linux iGPU GPGPU driver for Core Ultra 9 285H

[Mino1289]

Hi, I'm using a Lenovo laptop with an Intel Core Ultra 9 285H running Ubuntu 24.04.

On Intel's CPU Page, there are no listed driver for the iGPU for linux, but I saw another post where the vram of the iGPU couldn't be changed with the linux driver.
I tried to follow this guide: https://dgpu-docs.intel.com/driver/client/overview.html#installing-client-gpus-on-ubuntu-desktop without sucess.

So my question is: How can I install the GPGPU driver for 285H, if it's already out ?

Otherwise, when it'll be out ? (And how the other guy managed to install a driver for linux ? here: Please enable at least 75% ram as vram for graphic driver for 285H, 285K, 288V cpu.! )

[RandyT_Intel]

Hi @Mino1289,

This should be working "out of the box" without the need to run additional commands. Just install ubuntu and graphics should be working well right away. Most Linux-based* distributions already include Intel® Graphics Drivers. These drivers are provided and maintained by the Linux* distribution vendors and not by Intel, we recommend contacting the Linux* Operating System Vendors.

Regards,

Randy T.

Intel Customer Support Technician

[Mino1289]

Hi, Thanks for the fast reply.

The display driver is working, my display is working at expected resolution.

However when i want to install pytorch for xpu (ie, the iGPU), I followed the guide here (https://docs.pytorch.org/docs/2.7/notes/get_start_xpu.html) that says to install the gpgpu driver, and then install pytorch with the command provided.

After installing, I check if xpu device is available, but I get the error that I don't have the driver: 

>>> import torch
>>> torch.xpu.is_available()

~/miniconda3/envs/torch2.7+xpu/lib/python3.11/site-packages/torch/xpu/__init__.py:60: UserWarning: Failed to initialize XPU devices. The driver may not be installed, installed incorrectly, or incompatible with the current setup. Please refer to the guideline (https://github.com/pytorch/pytorch?tab=readme-ov-file#intel-gpu-support) for proper installation and configuration. (Triggered internally at /pytorch/c10/xpu/XPUFunctions.cpp:109.)
return torch._C._xpu_getDeviceCount()
False

This is the problem I'm facing.

Thanks for your help

[RandyT_Intel]

Hi @Mino1289,

Let me further investigate the issue. In the meantime, please help by answering the following questions and providing the requested information:

1. Kindly generate and share the SSU (System Support Utility) log report so I can review your system’s configuration in detail.
2. Is this your first time setting up this operating system on your device? If not, what OS or kernel version were you using previously?
3. Have you contacted the Linux or OS vendor support to inquire about the graphics driver compatibility for your integrated GPU?
4. To help us replicate the issue on our end, could you provide a screen recording showing how the issue occurs? Please also include a screenshot of any specific error messages you encounter.

I look forward to your response and am committed to resolving your issue promptly. 

Regards,

Randy T.

Intel Customer Support Technician

[Mino1289]

Hi @RandyT_Intel,

Thanks for the fast reply,
Here are the answers to your questions:

1. The output file of the SSU is attached (txt file). 

2. It's my first time on Linux on this device, I installed Windows 11 Pro, then added Ubuntu 24.04.2 as a Dual Boot.

3. No, I have not contacted Ubuntu Support as I've understood that I needed a GPGPU driver from intel (that i didn't found) for pytorch to use the iGPU as a device.

4. The video is attached (.mp4 file). I tried my best: I showed all the packages I installed with intel, the torch+xpu pip package installed and then the error in python when trying to use xpu device. Tell me if you need more information.

Many thanks for helping me out.

[RandyT_Intel]

Hi @Mino1289,

Kindly allow me some additional time to further investigate this matter, as I need to review it on my end. I will provide you with an update once I have more information. I appreciate your patience while I work on this concern.

Regards,

Randy T.

Intel Customer Support Technician

[RandyT_Intel]

Hi @Mino1289,

I wanted to update you on our efforts to recreate the issue you reported, where Python (pre-installed in Ubuntu) shared an error indicating that the driver was not detected in the virtual environment to run PyTorch. Unfortunately, we were unable to encounter the same issue on our end.

Here are the steps we followed to attempt to recreate the issue:

1. Open a terminal
2. Follow graphics driver installation from - https://dgpu-docs.intel.com/driver/client/overview.html
3. After following the steps above, run the command: sudo apt install python3-pip
4. After the installation above, run the command: sudo apt install python3-venv
5. Create a virtual environment named PyTorch_Test by running the following code: python3 -m venv PyTorch_Test
6. Enter/Activate the virtual environment using the command: source PyTorch_Test/bin/activate
7. Install PyTorch using the following command: pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/xpu
8. Once done, run the following command: python
9. You'll be prompted on run commands in the python interpreter - run the following command:

• import torch
• torch.xpu.is_available()

Regarding the support for PyTorch, I am continuing to check internally. In the meantime, could you please try the steps we've mentioned to ensure that the graphics driver is loaded in the virtual environment you are using?

Once you've completed these steps, could you provide an updated SSU scan of your system? This will help us further analyze the situation and provide more targeted assistance.

Regards,

Randy T.

Intel Customer Support Technician

[Mino1289]

Hi @RandyT_Intel,

Many thanks for the follow up!

I tried the steps you provided, and it worked! However, it's using the native python and venv (which I don't like) instead of conda (miniconda in my case).

I also noted while trying other things, that in my conda env, I was unable to use the NPU and GPU with openvino. But here, with the venv, everything is showing up and should be working just fine.

Do you have any idea why python from a conda environnement could not access those device (primarely the GPU) ?

I did the SSU scan, attached below.

Thanks again for the fast and efficient support.

Regards,

[RandyT_Intel]

Hi @Mino1289,

I'm glad to hear that the initial issue was resolved using the steps provided! Regarding your inquiry about the conda environment and device access, I'll need to look into this further. I'll report back once I have more information.

In the meantime, can you confirm if everything is working as expected with the Python and venv setup? Your feedback will help ensure that the initial problem is fully addressed.

Regards,

Randy T.

Intel Customer Support Technician

[Mino1289]

Hi @RandyT_Intel ,

I confirm that everything is working as expected with the Python and venv setup. Thanks again for your help.

However, with pytorch using torch.compile, I got this error/warning:
"W0702 15:27:54.446000 40833 torch/_inductor/utils.py:1250] [0/1] Not enough SMs to use max_autotune_gemm mode
sycl_arch not recognized: 310379520
sycl_arch not recognized: 310379520
W0702 15:28:02.657000 40833 torch/utils/_sympy/interp.py:176] [0/1] failed while executing pow_by_natural([VR[1, int_oo], VR[-1, -1]])

....
sycl_arch not recognized: 310379520

WARNING:root:Wall time is used instead of elapsed_time because of the bug ('negative timings'). Should be fixed in DLE 2025.1. The timing measurements could be innacurate. "

And it's slower (30x) than not using torch.compile().
Here is the code I used:

import torch
import torchvision.models as models

device = torch.device("xpu" if torch.xpu.is_available() else "cpu")
print(f"Using device: {device}")

model = models.resnet152(num_classes=10).to(device)
model.compile()

with torch.no_grad():
    # Dummy input for profiling
    dummy = torch.randn(1, 3, 224, 224).to(device)
    
    out = model(dummy)
print(f"Output shape: {out.shape}")
print(f"Output: {out}")

Regards,

[RandyT_Intel]

Hi @Mino1289,

Thank you for the updates. I'm currently reviewing and exploring possible options regarding your inquiry. Kindly allow me some additional time to look into this further. I truly appreciate your patience as I work to resolve this matter.

Regards,

Randy T.

Intel Customer Support Technician

[RandyT_Intel]

Hi @Mino1289,

We understand the importance of having the right drivers to ensure optimal performance and compatibility.

Intel generally provides open-source drivers for Linux, which are integrated into the Linux kernel. These drivers are frequently updated through kernel updates and distributions like Ubuntu. You can download the drivers from the following link: Intel Linux Drivers.

If you continue to experience issues after using the link provided, we recommend contacting Lenovo to verify whether your system can fully support Linux. The system specifications indicate that this computer is designed to support Windows® 11 Home, and Lenovo can provide further guidance on compatibility.

Please confirm with Lenovo regarding your system's support for Linux. We will be closing this inquiry,Please note that it will no longer be monitored. However, you may submit a new ticket if you need further assistance in the future.

For additional information, you may find the following articles helpful:

• Where to Obtain an Intel® Graphics Driver for an OEM (Original Equipment Manufacturer) Product?
• How to Identify & Find Graphics Drivers for Linux*
• Where Can I Find Linux Drivers for Intel® Arc™ Graphics?

Regards,

Randy T.

Intel Customer Support Technician