As per the PIPE 3.0 specification, for every 128 bits that are moved across the Gen3 PCS, the PHY must transmit 130 bits of data. Intel uses the pipe_tx_data_valid signal every 16 blocks of data to transmit the built-up backlog of 32 bits of data.
The 130-bit block is received as follows in the 32-bit data path: 34 (32+2-bit sync header), 32, 32, 32. During the first cycle, the gearbox converts the 34-bit input data to 32-bit data. During the next three clock cycles, the gearbox merges bits from adjacent cycles. For the gearbox to work correctly, a gap must be provided in the data for every 16 shifts because each shift contains two extra bits for converting the initial 34 bits to 32 bits in the gearbox. After 16 shifts, the gearbox has an extra 32 bits of data that are transmitted out. This requires a gap in the input data stream, which is achieved by driving pipe_tx_data_valid low for one cycle after every 16 blocks of data.