Problem: I got the error No CUDA host compiler for the selected platform and CUDA Toolkit version 10.0 when trying to build catboost for GPU
Build command:
../../../ya make -r -DUSE_ARCADIA_PYTHON=no -DOS_SDK=local -DPYTHON_CONFIG=python3-config -DCUDA_ROOT=/usr/local/cuda
catboost version: Commit 56c0928
Operating System: Ubuntu 18.04
CPU: Ryzen Threadripper 1920x
GPU: RTX 2080 Ti
~ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2018 NVIDIA Corporation
Built on Sat_Aug_25_21:08:01_CDT_2018
Cuda compilation tools, release 10.0, V10.0.130
~ nvidia-smi
Thu Nov 1 16:50:03 2018
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 410.73 Driver Version: 410.73 CUDA Version: 10.0 |
|-------------------------------+----------------------+----------------------+
Edit: Add build command and nvcc --version output
@taurandat How did you solve this issue?
@taurandat I have the same problem. How did you solve this issue?
Hi @taurandat, @ei-grad, @diaper151,
The build problem is solved by adding -DCUDA_HOST_COMPILER='' flag on ya make ... build line.
Hi @taurandat, @ei-grad, @diaper151,
The build problem is solved by adding-DCUDA_HOST_COMPILER=''flag onya make ...build line.
That works for me, thanks @dbakshee
Most helpful comment
Hi @taurandat, @ei-grad, @diaper151,
The build problem is solved by adding
-DCUDA_HOST_COMPILER=''flag onya make ...build line.