Intel, are you OK?

[yizhangliu]

We need you!

[Dosperado74]

Looks like they stopped developing IPEX-LLM...Good bye AI.

[erwinzierler]

I guess that was the last Intel PC I bought for AI work...

[matheus2308]

Are there any alternatives for LLM with Intel GPUs that is being maintained?

[ysaric]

Have had success in Windows with the Vulkan version here: https://github.com/ggml-org/llama.cpp

[matheus2308]

Also just saw that Vulkan can now be enabled with the offical Ollama. It worked with my GPU (Intel B50), although is a lot slower than ipex-llm (2-3x slower tokens/s), details here: https://github.com/ollama/ollama/releases/tag/v0.12.11

[ChrisPowers666]

Also the new LM Studio Beta works with Vulkan and Intel GPUs: https://lmstudio.ai/beta-releases

[charlescng]

Also just saw that Vulkan can now be enabled with the offical Ollama. It worked with my GPU (Intel B50), although is a lot slower than ipex-llm (2-3x slower tokens/s), details here: https://github.com/ollama/ollama/releases/tag/v0.12.11

On my Arc B580, running gemma3:12b it went from 35t/s with ipex to 15t/s with Vulkan.

[hoiber]

seems strange considering the B50 and B60...

[erwinzierler]

I have been playing with OpenVINO (https://github.com/openvinotoolkit/openvino) lately. It needs a bit more effort to get it running than LMstudio or Ollama but it really does make use of Intels CPU, GPU and NPU.