When attempting to start the Kinetica services, a Cuda runtime error may be encountered.
2018-09-21 18:36:24.206 ERROR (36064,36064,gpudb_cluster_c ) <hostname> Utils/GaiaCudaUtils.cpp:137 - CUDA error: 30 'unknown error', running: 'cudaRuntimeGetVersion(&GaiaGpuDeviceInfo::cuda_version)', exiting.
Config : Error
Error starting host manager with /opt/gpudb/core/bin/gpudb as user: gpudb, error code 1
This may be caused by one of the following reasons:
(I) A DKMS (Dynamic Kernel Module Support) is not installed and the kernel has been upgraded. GPU card drivers will need to be completely uninstalled and reinstalled.
(II) A GPU card that is not compatible with the CUDA version that Kinetica needs for the installed build version - Cuda80 or Cuda91
- Example: K80's have issues on Cuda91 and the Cuda80 version would need to be installed
For information on Nvidia Drivers Installation:
- Nvidia Documentation - Nvidia Installation Guide
- Kinetica Documentation - Install Nvidia Drivers on Debian/Ubuntu
- Kinetica Documentation - Install Nvidia Drivers on RHEL
- Affects Version(s):
- All Cuda Builds