Is it possible to install xgboost 3.0.4 for R with GPU support on Windows (without WSL2)?

[marboe123]

I have tried to install XGBoost 3.0.4 for R with GPU support on Windows with these steps:

Installing:

1. R 4.5
2. Visual Studio 2022
3. CUDA Toolkit 12.9
4. Rtools44
5. xgboost-src-3.0.4.tar.gz

Although I do not succeed.

Would it be possible at all,
or is it only possible to install XGBoost 3.x for R with GPU support on Windows in WSL2?

Thanks a lot!

[trivialfis]

Unfortunately, no. The Visual Studio C++ compiler had some issues with the R headers and we dropped the binary support on Windows. I haven't done any checks with the latest VC yet.

[marboe123]

Thank you for your answer. I suppose it is a good workaround to do my coding in the XGBoost CPU version within Rstudio Desktop and train models in XGBoost GPU version in WSL2. It would be nice if the VC issues would be resolved and binary support on Windows could be continued. Thank you in any case for all the new features and development of XGBoost!

[trivialfis]

It would be nice if the VC issues would be resolved and binary support on Windows could be continued.

We will need to discuss this with the R development team or the Microsoft VC team to make it happen. So far, I don't think the R core team works with MSVC since there's no test for CRAN packages using MSVC, I doubt that they will add it now.

[marboe123]

Thank you for you explanation.