Another problem is the version of NCCL - the 1<sup>st</sup> version is [available on github](https://github.com/NVIDIA/nccl/releases) and we can integrate it easily to the CI. But the latest release links to CUDA 8.0 wich requires a separate docker build, since the usual builds are against 9.1 currently. The 2<sup>nd</sup> version is available only behind a registration: https://developer.nvidia.com/nccl
[ Full content available at: https://github.com/apache/incubator-mxnet/pull/12477 ] This message was relayed via gitbox.apache.org for [email protected]
