MPICH_G2G_PIPELINE default value

[jgphpc]

• In cuda_mpi.pdf/slide 13:

MPICH_G2G_PIPELINE
Sets the maximum number of 512 kB message chunks that can be in flight (default 16)

but in /opt/cray/pe/mpt/7.6.0/gni/man/mpich/man3/intro_mpi.3:

If MPICH_RDMA_ENABLED_CUDA is enabled but MPICH_G2G_PIPELINE is 0, 
the default value is set to 8.

=> 8 or 16 ? can you confirm please ?

• kernels.pdf/slide10:
  can run 2028 threads
  I'd say 2048 ?