[Issue]: I can't use Pytorch on Windows 11 on 9070 XT no matter what I do.

[skillmaker-dev]

Problem Description

Hey, I'm trying to follow this tutorial AMD Pytorch installation, however, I get no output from commands run after this step:
1- Verify if Pytorch is installed and detecting the GPU compute device.
python -c "import torch" 2>nul && echo Success || echo Failure

I tried removing all Python installations and tried again with UV, but still the same issue:

Anas ❯ uv venv
Using CPython 3.12.13
Creating virtual environment at: .venv
Activate with: .venv\Scripts\activate
                                                                                                             
Anas ❯ .venv\Scripts\activate
                                                                                                
Anas ❯ uv pip install --no-cache-dir `
>     https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/rocm_sdk_core-7.2.1-py3-none-win_amd64.whl `
>     https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/rocm_sdk_devel-7.2.1-py3-none-win_amd64.whl `
>     https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/rocm_sdk_libraries_custom-7.2.1-py3-none-win_amd64.whl `
>     https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/rocm-7.2.1.tar.gz
Resolved 4 packages in 860ms
      Built rocm @ https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/rocm-7.2.1.tar.gz
Prepared 4 packages in 1m 15s
Installed 4 packages in 1.16s
 + rocm==7.2.1 (from https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/rocm-7.2.1.tar.gz)
 + rocm-sdk-core==7.2.1 (from https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/rocm_sdk_core-7.2.1-py3-none-win_amd64.whl)
 + rocm-sdk-devel==7.2.1 (from https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/rocm_sdk_devel-7.2.1-py3-none-win_amd64.whl)
 + rocm-sdk-libraries-custom==7.2.1 (from https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/rocm_sdk_libraries_custom-7.2.1-py3-none-win_amd64.whl)
                                                                                                 pytorch  pytorch 
1m 19.262s Anas ❯ uv pip install --no-cache-dir `
>     https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/torch-2.9.1%2Brocm7.2.1-cp312-cp312-win_amd64.whl `
>     https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/torchaudio-2.9.1%2Brocm7.2.1-cp312-cp312-win_amd64.whl `
>     https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/torchvision-0.24.1%2Brocm7.2.1-cp312-cp312-win_amd64.whl
Resolved 17 packages in 1.90s
Prepared 14 packages in 1m 18s
Installed 14 packages in 6.88s
 + filelock==3.29.0
 + fsspec==2026.4.0
 + jinja2==3.1.6
 + markupsafe==3.0.3
 + mpmath==1.3.0
 + networkx==3.6.1
 + numpy==2.4.4
 + pillow==12.2.0
 + setuptools==82.0.1
 + sympy==1.14.0
 + torch==2.9.1+rocm7.2.1 (from https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/torch-2.9.1%2Brocm7.2.1-cp312-cp312-win_amd64.whl)
 + torchaudio==2.9.1+rocm7.2.1 (from https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/torchaudio-2.9.1%2Brocm7.2.1-cp312-cp312-win_amd64.whl)
 + torchvision==0.24.1+rocm7.2.1 (from https://repo.radeon.com/rocm/windows/rocm-rel-7.2.1/torchvision-0.24.1%2Brocm7.2.1-cp312-cp312-win_amd64.whl)
 + typing-extensions==4.15.0
                                                                                                
Anas ❯ python -c "import torch" 2>nul && echo Success || echo Failure
Success
                                                                                                
Anas ❯ python -c "import torch; print(torch.cuda.is_available())"
                                                                                                
Anas ❯ python -c "import torch; print(f'device name [0]:', torch.cuda.get_device_name(0))"
                                                                                                
Anas ❯ python -m torch.utils.collect_env
<frozen runpy>:128: RuntimeWarning: 'torch.utils.collect_env' found in sys.modules after import of package 'torch.utils', but prior to execution of 'torch.utils.collect_env'; this may result in unpredictable behaviour
Collecting environment information...

Anas ❯ hipinfo
                                      

I'm on version 26.3.1 of the Adrenalin GPU driver.

Operating System

Windows 11

CPU

AMD RYZEN 5 9600x

GPU

RX 9070XT

ROCm Version

7.2

ROCm Component

No response

Steps to Reproduce

No response

(Optional for Linux users) Output of /opt/rocm/bin/rocminfo --support

No response

Additional Information

No response

[Samiii777]

@skillmaker-dev it might be picking up your igpu, from device manager under device adapters check if you have 2 devices and if so disable the integrated graphics and try again

[skillmaker-dev]

@Samiii777 That was it, thanks! Disabling the iGPU made it work, but how can I make my second monitor use the iGPU? Is there a way to make ROCm ignore the iGPU?

[ScottTodd]

Does hipinfo or offload-arch give you a list of GPU targets like gfx1201? I'm curious which the iGPU is. Those ROCm 7.2.1 packages were built for some subset of GPU targets. We're in the process of bringing up new packages (https://github.com/ROCm/TheRock/blob/main/RELEASES.md#installing-multi-arch-rocm-python-packages) and we have a few things that might help (or at least that we'd like to work):

This should fetch all devices we currently build (some targets may not be fully functional, but you might get a different failure mode)

pip install --index-url https://rocm.nightlies.amd.com/whl-staging-multi-arch/ "torch[device-all]"

This should fetch just two devices:

# replace gfx1036 with whatever that iGPU is
pip install --index-url https://rocm.nightlies.amd.com/whl-staging-multi-arch/ "torch[device-gfx1201,device-gfx1036]"

I suspect that if you have packages that include at least baseline support for the dgpu and the igpu then torch should be able to filter on its own via the devices list 🤞

[Samiii777]

you could use env variables to isolate your gpus if needed
set HIP_VISIBLE_DEVICES=0 or 1
https://rocm.docs.amd.com/projects/HIP/en/latest/reference/env_variables.html