Sylvain Jeaugey

I think CUDA 7 is quite old already ; having a solution which may not work in some cases with CUDA 6 and earlier doesn't seem like a serious problem...

Libnvml should be installed together with the driver. Those stubs are only there in case you want to compile on a machine that has no CUDA card but only the...

I'm assuming we're talking about containers where only a single GPU is visible -- please correct me if I'm wrong. I see two rather large steps. 1. CUDA. Last time...

@khj94 Sorry for asking @PerkzZheng to ask again -- my bad, I just realized the second log had the right `NCCL_DEBUG_SUBSYS` set and I'm not seeing anything because the error...

The bash script doesn't initialize MPI at all, so maybe an intermediate step would be to run an MPI hello world program (to get rid of the DL framework, Horovod...