Maximum performance can be obtained by choosing an AXI4-Stream TDATA width that is greater than the width of the widest channel when converted to a vector. If the TDATA width is smaller than the AXI4-MM channel, the transfer must be converted into multiple AXI4-S transfers.