Building XGBoost 3.2 fails when CUDA < 12.9

[abdouskamel]

Hello,

I am trying to build XGBoost 3.2 but I get the error:

CMake Error at CMakeLists.txt:236 (message):
  CUDA version must be at least 12.9!

I do not have access to CUDA >= 12.9. Is there a way around this error or do I require CUDA >= 12.9?

Many thanks for your time.

[trivialfis]

Hi, do you have access to the user space of newer CUDA? Meaning, no need to update the system driver, we just need the compiler and libraries.

One can obtain user space programs through many means, I use conda to fetch the toolchain for development: https://github.com/trivialfis/dxgb_bench/blob/master/dev/xgboost_dev.yml

[trivialfis]

One can even fetch the CUDA toolkit through pip from pypi now: https://pypi.org/project/nvidia-cuda-nvcc/

[abdouskamel]

Thank you for your quick reply. I will give the pypi package a try.