Skip to content

Conversation

@kotauchisunsun
Copy link

Hello,

I have created a pull request for the Windows SYCL backend. The version of the OneAPI base toolkit used in the Windows build was 2025.0.0, which is outdated, so I have updated it to the latest version (2025.1.0).

I conducted a performance comparison between the SYCL version of llama.cpp in this repository and the version built in the ipex-llm repository using the following URL. The results showed that the llama.cpp version in this repository is slower. (The article is in Japanese.)

Performance Comparison Article

Based on these findings, I suspect that the outdated OneAPI base toolkit version is the cause of the slow performance of llama.cpp. Therefore, I have created this pull request.

I attempted to create a pull request on my own fork, but I was unable to build it due to Intel's licensing restrictions. Since my repository does not have the required license files, the build could not proceed. However, I believe that the build server for llama.cpp has the necessary license files, so it should function correctly there.

Build Attempt Log

Could you please review and merge this pull request?
If there are any additional necessary actions, please let me know.

@github-actions github-actions bot added the devops improvements to build systems and github actions label Apr 18, 2025
@qnixsynapse
Copy link
Collaborator

Thank you for your contribution. I think the latest version has some problem on Windows.
Please see: #12696

@Rbiessy Rbiessy self-requested a review April 18, 2025 14:00
@Rbiessy Rbiessy added the SYCL https://en.wikipedia.org/wiki/SYCL - GPU programming language label Apr 18, 2025
@s-Nick
Copy link
Contributor

s-Nick commented Apr 18, 2025

Thank you @kotauchisunsun for your contribution. @qnixsynapse is right, currently due to a cmake issue, oneAPI 2025.1 doesn't configure the project correctly on Windows. We are looking into it and update the configuration as soon as possible.

@NeoZhangJianyu
Copy link
Collaborator

2025.1.0 issue will be fixed in next release, like 2025.1.1.
Let's wait for the 2025.1.1 release.
I suggest pending this PR for next release.

@kotauchisunsun
Copy link
Author

@qnixsynapse @s-Nick @NeoZhangJianyu

Thank you for your reply.

2025.1.0 issue will be fixed in next release, like 2025.1.1.
Let's wait for the 2025.1.1 release.
I suggest pending this PR for next release.

I'm going to close this PR. Thanks for the review!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

devops improvements to build systems and github actions SYCL https://en.wikipedia.org/wiki/SYCL - GPU programming language

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants