Feature Request: Add Support for NPUs on Intel, AMD, and Qualcomm

[DaNitroNinja]

Intel Core Ultra, AMD and Qualcomm's Snapdragon X Plus/Elite chips have NPUs and they should have support in LM Studio as it would severely improve efficiency.

Intel (I believe) has OpenVINO and surely could somehow be used to allow for NPUs to be accessed by LM Studio instead of just CPU / GPU.

[johntsams]

Unfortunately it is not so simple or else it would be implemented already. It would significantly improve energy efficiency running models, and if implemented well, probably speed it up too.

LM Studio uses the llama.cpp backend that can handle a ton of different CPU and GPU vendors and configurations. It is made to handle GGUF model formats primarily.

The Snapdragon NPU can use ONNX model formats. However, these will not run on LM Studio.

If the llama backend can be written to handle Snapdragon NPUs then this would work pretty seamlessly. But that is a very challenging and intensive process to integrate that functionality. Some company will likely have to support that effort.

[DaNitroNinja]

So it could be possible in the future, especially with companies like Intel as they have OpenVino.

[johntsams]

Check this out. Someone got ONNX formatted models running in LM studio on the NPU

https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/cross-platform-edge-ai-made-easy-with-onnx-runtime/4303521

[taksheels]

It looks like llama.cpp has added support (experimental) for Qualcomm Hexagon NPUs.

https://github.com/ggml-org/llama.cpp/releases/tag/b6822

[behroozbc]

I think this is possible in experimental mode if anybody wants it

[athuljayaram]

i have ryzen ai 7 350. i want it to run. how can I?

[perrypink]

i have ryzen ai 7 350. i want it to run. how can I?

Try this build, sir 🙏🏾
https://lmstudio.ai/ryzenai

[athuljayaram]

Does it support 20b parameter and more models?

i have ryzen ai 7 350. i want it to run. how can I?

Try this build, sir 🙏🏾 https://lmstudio.ai/ryzenai

[DaNitroNinja]

i have ryzen ai 7 350. i want it to run. how can I?

Try this build, sir 🙏🏾 https://lmstudio.ai/ryzenai

Do you know if there is an alternative for Intel AI Boost? I have Lunar Lake with 140V

[athuljayaram]

i have ryzen ai 7 350. i want it to run. how can I?

Try this build, sir 🙏🏾 https://lmstudio.ai/ryzenai

great, I can see its working

how to give it full continous utilization

[behroozbc]

is any way to test qualcomm?

[perrypink]

Do you know if there is an alternative for Intel AI Boost?

OpenVINO + llama.cpp
Intel AI Playground 3 🙏🏾

[perrypink]

Snapdragon Laptops

Nexa AI SDK
AnythingLLM 🙏🏾👍🏾
https://www.qualcomm.com/developer/blog/2026/03/run-nexa-ai-agents-locally-on-snapdragon-pc-with-hexagon-npu