Introducing AMD GPU EP for Generative AI with Amuse
Oct 06, 2025
Overview
We’re excited to announce the technology preview of the AMD GPU Execution Provider (EP) for ONNX-based generative AI workflows using Amuse—an AI based image generation application. This release brings AMD ROCm™ based GPU acceleration to both integrated (iGPU) and discrete GPUs (dGPU) for Windows platforms, enabling creators and developers to harness the full power of AMD Radeon™ graphics for AI image generation.
Key Highlights
- ONNX-based AMD GPU EP backend is now available with Amuse.
- Supports AMD optimized models from the AMD Hugging Face Collection. For more detail on these models check our earlier blog post
- Built on AMD ROCm technologies, including MIGraphX, rocMLIR, MIOpen, HIPBlasLT, Composable Kernels, and KernelLib.
- Preview support for Windows platforms powered by AMD Radeon™ graphics.
Prerequisites
- Platform: System with AMD RDNA3™+ Graphics (GPUs)
- Driver: AMD Software: Adrenalin Edition™
- Amuse with AMD GPU EP technology preview
Getting Started with Amuse + AMD GPU EP Technology Preview
- Launch Amuse and switch to Expert Mode
2. Navigate to Settings and set Execution Device to AMDGPU (Preview).
3. Go back to the Ez mode, Start generating the images by clicking “Generate Images”
4. Accept the license agreement and optimization warning
5. Model download and optimization will start
Once downloaded, models are automatically optimized.
Note: The AMDGPU EP execution provider currently requires models to be optimized before they can run. This optimization step is automatically triggered after the model is downloaded and may consume significant time and system resources, depending on your hardware.
The optimization is performed once per model, and subsequent runs will use the cached optimized version from disk—unless model settings are changed.
If the system does not meet the memory requirements for a given model, the automatic optimization process may fail.
6. At the end of the process the images will be generated:
To run any specific models downloaded from Hugging Face, go back to the “Expert Mode”:
- Go to Model Manager, and select a model to download
Manual optimization is available if needed
- Load the optimized model and start generating images.
Prompting Tips for Stable Diffusion 3x
- Structure prompts by describing the image type → structure → details.
- Use consistent settings (seed, steps, sampler) for reproducibility.
- Iterate prompts to refine visual structure.
- Use negative prompts sparingly to avoid quality degradation.
Sample Prompts
Bengal Tiger
"Close up, award winning wildlife photography, vibrant and exotic face of a Bengal Tiger against a colorful background, vibrant color, best shot, 8k, photography, high res"
Steps: 20 | Seed: 169776499 | Backend: AMD GPU EP (Preview)
GPU EP Soda Pop
"Macro photography, advertising commercial for the brand GPU EP. A GPU EP soda pop in bright colors with only big letters 'GPU EP' logo text. The background has sliced fruit and water splashing around the can. Studio setting, deep depth of field, professional photography, highly detailed."
Steps: 22 | Seed: 474739515 | Backend: AMD GPU EP (Preview)
GPU EP Jersey
"Wide and low angle, cinematic, fashion photography for the brand GPU EP. Woman wearing a GPU EP jersey with big letters 'GPU EP' logo text, and brown chinos. The background is a gradient, red, pink and orange, studio setting."
Steps: 22 | Seed: 1544622413 | Backend: AMD GPU EP (Preview)
Under the Hood: AMD GPU EP Architecture
AMD GPU EP is built on ONNX Runtime and powered by ROCm technologies.
Key components of AMD GPU EP includes:
- MIGraphX: ML inference engine and graph compiler for efficient execution.
- rocMLIR: Dynamic kernel generator for convolutions and GEMMs.
- MIOpen & HIPBlasLT: Provide finely tuned kernels for high-performance AI workloads.
- Composable Kernels & KernelLib: Enable scalable, tile-based programming and hand-optimized operators.
- KernelLib: highly optimized ML fused operators reused from DirectX based ML implementation
Known Opens(s)
- If “AMDGPU EP” does not appear in the “Execution Device” dropdown menu, ensure that any previous version of Amuse is uninstalled before installing the technology preview.
- Model optimization may take a significant amount of time, depending on system configuration and available resources.
- Intermittent failures may occur during the optimization phase on some systems. If encountered, restart the application and resume the process.
Conclusion
The AMD GPU EP preview for Amuse marks a significant step in democratizing generative AI on AMD platforms. With ROCm technologies and AMD optimized models, creators can now experience high-performance image generation directly on AMD Radeon graphics.
References
https://gpuopen.com/learn/accelerating_generative_ai_on_amd_radeon_gpus/
https://stability.ai/news/stable-diffusion-now-optimized-for-amd-radeon-gpus
https://www.amd.com/en/products/graphics/radeon-ai.html
https://www.amd.com/en/blogs/2025/worlds-first-bf16-sd3-medium-npu-model.html
Footnotes
DISCLAIMERS AND FOOTNOTES
*= Image generation using SD 3 Medium is regulated by the Stability AI Community License and is free for personal use and for SMEs under $1 million in annual revenue. Licensing requirements may change at the sole discretion of the third party. For complete licensing requirements please refer to: LICENSE.md · stabilityai/stable-diffusion-3-medium at main. Internet connection is required to download the model and other configuration files.
Please note the Amuse technology preview release is an experimental software. AMD GPU EP preview is still under active development and is subject to frequent changes. Models tested on various AMD platforms: Stable Diffusion 3.0 medium, Stable Diffusion 3.5 medium, Stable Diffusion 3.5 Turbo, Stable Diffusion 1.5, Stable Diffusion XL, Stable Diffusion XL Turbo. It is being provided on an “as is” basis. As such, there may be bugs and the user experience may be unstable or cause unintended behavior.
Not all ROCm building blocks referenced in the article are enabled in the current technology preview release. These components will be progressively enabled and deployed in future updates.
Links to third-party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites, and no endorsement is implied. GD-98
GD-97 - Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.
RX-1143: Based on specifications of AMD RDNA 4 architecture compared to AMD RDNA 3 architecture as of December 2024.
Attributions
The information presented in this document is for informational purposes only and may contain technical inaccuracies, omissions, and typographical errors. The information contained herein is subject to change and may be rendered inaccurate for many reasons, including but not limited to product and roadmap changes, component and motherboard version changes, new model and/or product releases, product differences between differing manufacturers, software changes, BIOS flashes, firmware upgrades, or the like. Any computer system has risks of security vulnerabilities that cannot be completely prevented or mitigated. AMD assumes no obligation to update or otherwise correct or revise this information. However, AMD reserves the right to revise this information and to make changes from time to time to the content hereof without obligation of AMD to notify any person of such revisions.
THIS INFORMATION IS PROVIDED ‘AS IS.” AMD MAKES NO REPRESENTATIONS OR WARRANTIES WITH RESPECT TO THE CONTENTS HEREOF AND ASSUMES NO RESPONSIBILITY FOR ANY INACCURACIES, ERRORS, OR OMISSIONS THAT MAY APPEAR IN THIS INFORMATION. AMD SPECIFICALLY DISCLAIMS ANY IMPLIED WARRANTIES OF NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR ANY PARTICULAR PURPOSE. IN NO EVENT WILL AMD BE LIABLE TO ANY PERSON FOR ANY RELIANCE, DIRECT, INDIRECT, SPECIAL, OR OTHER CONSEQUENTIAL DAMAGES ARISING FROM THE USE OF ANY INFORMATION CONTAINED HEREIN, EVEN IF AMD IS EXPRESSLY ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.
© 2025 Advanced Micro Devices, Inc. All rights reserved. AMD, the AMD Arrow logo, Radeon, Ryzen, and combinations thereof are trademarks of Advanced Micro Devices, Inc. Microsoft is a registered trademark of Microsoft Corporation in the US and/or other countries. Other product names used in this publication are for identification purposes only and may be trademarks of their respective owners. Certain AMD technologies may require third-party enablement or activation. Supported features may vary by operating system. Please confirm with the system manufacturer for specific features. No technology or product can be completely secure.
Footnotes
DISCLAIMERS AND FOOTNOTES
*= Image generation using SD 3 Medium is regulated by the Stability AI Community License and is free for personal use and for SMEs under $1 million in annual revenue. Licensing requirements may change at the sole discretion of the third party. For complete licensing requirements please refer to: LICENSE.md · stabilityai/stable-diffusion-3-medium at main. Internet connection is required to download the model and other configuration files.
Please note the Amuse technology preview release is an experimental software. AMD GPU EP preview is still under active development and is subject to frequent changes. Models tested on various AMD platforms: Stable Diffusion 3.0 medium, Stable Diffusion 3.5 medium, Stable Diffusion 3.5 Turbo, Stable Diffusion 1.5, Stable Diffusion XL, Stable Diffusion XL Turbo. It is being provided on an “as is” basis. As such, there may be bugs and the user experience may be unstable or cause unintended behavior.
Not all ROCm building blocks referenced in the article are enabled in the current technology preview release. These components will be progressively enabled and deployed in future updates.
Links to third-party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites, and no endorsement is implied. GD-98
GD-97 - Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.
RX-1143: Based on specifications of AMD RDNA 4 architecture compared to AMD RDNA 3 architecture as of December 2024.
Attributions
The information presented in this document is for informational purposes only and may contain technical inaccuracies, omissions, and typographical errors. The information contained herein is subject to change and may be rendered inaccurate for many reasons, including but not limited to product and roadmap changes, component and motherboard version changes, new model and/or product releases, product differences between differing manufacturers, software changes, BIOS flashes, firmware upgrades, or the like. Any computer system has risks of security vulnerabilities that cannot be completely prevented or mitigated. AMD assumes no obligation to update or otherwise correct or revise this information. However, AMD reserves the right to revise this information and to make changes from time to time to the content hereof without obligation of AMD to notify any person of such revisions.
THIS INFORMATION IS PROVIDED ‘AS IS.” AMD MAKES NO REPRESENTATIONS OR WARRANTIES WITH RESPECT TO THE CONTENTS HEREOF AND ASSUMES NO RESPONSIBILITY FOR ANY INACCURACIES, ERRORS, OR OMISSIONS THAT MAY APPEAR IN THIS INFORMATION. AMD SPECIFICALLY DISCLAIMS ANY IMPLIED WARRANTIES OF NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR ANY PARTICULAR PURPOSE. IN NO EVENT WILL AMD BE LIABLE TO ANY PERSON FOR ANY RELIANCE, DIRECT, INDIRECT, SPECIAL, OR OTHER CONSEQUENTIAL DAMAGES ARISING FROM THE USE OF ANY INFORMATION CONTAINED HEREIN, EVEN IF AMD IS EXPRESSLY ADVISED OF THE POSSIBILITY OF SUCH DAMAGES.
© 2025 Advanced Micro Devices, Inc. All rights reserved. AMD, the AMD Arrow logo, Radeon, Ryzen, and combinations thereof are trademarks of Advanced Micro Devices, Inc. Microsoft is a registered trademark of Microsoft Corporation in the US and/or other countries. Other product names used in this publication are for identification purposes only and may be trademarks of their respective owners. Certain AMD technologies may require third-party enablement or activation. Supported features may vary by operating system. Please confirm with the system manufacturer for specific features. No technology or product can be completely secure.
Related Blogs
-
-
AMD Instinct MI355X GPU Supports DeepSeek-V3.2-Exp
AMD ROCm software ensures developers unlock fast adoption and strong performance for DeepSeek 3.2 model.
October 03, 2025
-
-
The Road to ROCm on Radeon for Windows and Linux
PyTorch on Windows and Linux is now available for public preview. Developers can run AI inference workloads directly on AMD hardware in the Windows environment.
September 24, 2025
-
Windows ML + AMD: Unlocking AI for Every Windows Developer
AMD and Microsoft are empowering a new era of intelligent applications, frictionless local inference, and easily deployable AI experiences across Windows.
September 23, 2025
-
AMD Leads with Results in the AI Era
Learn how Enterprise customers are using AMD processors to put AI to work in their business on servers in their data centers.
September 23, 2025
-
India’s 5G Breakthrough: How Jio Is Redefining Telecom with AMD
Reliance Jio is revolutionizing telecom with the world's largest 5G Standalone network, powered by AMD's advanced compute platforms.
September 17, 2025
