Host preparation for GDR

If we would like to use GDR (GPU Direct RDMA) among GPUs, do we need to load any additional kernel module besides what the MLNX_OFED already loads? I recall that in the past we needed to have “gdrcopy” and “nvidia_peermem”.

Do we still need to have these two modules loaded in the kernel in order to reap all GDR benefits?

Thanks