fix CI & CUDA forward compatibility - Githubissues

h2oai / h2o4gpu

H2Oai GPU Edition

Apache License 2.0

460 stars 95 forks source link

fix CI & CUDA forward compatibility #709

Closed sh1ng closed 5 years ago

sh1ng commented 5 years ago

The following issues are fixed:

forward compatibility, there was an issue on CUDA 8 + Tesla V at least
https://github.com/h2oai/h2o4gpu/issues/711
CI: triggers, check previous state, etc
adjusted tolerance when comparing with sklearn
disabled tests for lgbm on ppc64le
using a single GPU for xgboost on ppc64le
updated Apache arrow to 0.12.0
Docker-configuration for ppc64le
using the same version of numpy everywhere to speed-up CI
increased number of pytest workers

Are going to be fixed in separate PRs:

lgbm with gpu on ppc64le
xgbbost with NCCL on ppc64le

sh1ng commented 5 years ago

On a machine with V100 and CUDA 8 there's an error

terminate called after throwing an instance of 'thrust::system::system_error'
  what():  function_attributes(): after cudaFuncGetAttributes: invalid device function
Aborted