Can't compile working version of xgboot on Jetson Nano

Hi all

I am trying to use my Jetson Nano’s GPU for XGBoost training. According to the instructions here, I have downloaded the source and compiled successfully. According to these instructions, I have configured the build like this:

cmake .. -DUSE_CUDA=ON -DUSE_OPENMP=ON -DGPU_COMPUTE_VER=53

After installation, I can use the library with OpenMP, but not with CUDA. I’m getting the following error message:

terminate called after throwing an instance of ‘thrust::system::system_error’
what(): parallel_for failed: cudaErrorLaunchFailure: unspecified launch failure
Aborted (core dumped)

I’m wondering if I’m building xgboost the right way?

My system is based on an unmodified JP46 image.

Any help would be appreciated!

please refer to this post :

Hmmm, not sure what in the linked article applies. Are you saying that the issues with JetPack 4.2 from two years ago are still present in 4.6 and I need to use the open source version of thrust?

Hi Matthias,

Yes , the problem was solved in the post above. The thrust issue was verified and fixed on the newer jetpacks but it is still worth a try. I will find out more info