When [RT]XUSRCLK2 operates at half the frequency of [RT]XUSRCLK (i.e., separate BUFG_GTs with divide by 1 and divide by 2), a tight skew requirement exists between the [RT]XUSRCLK/[RT]XUSRCLK2 pair at each GT*CHANNEL of a GT interface. To meet the skew requirement, GT*CHANNELs can be a maximum of 2 clock regions above or below the master channel that generates the [RT]XUSRCLK/[RT]XUSRCLK2 pair. In addition, the placer tightly controls skew as follows:
- Assigns the BUFG_GT pairs to the upper or lower 12 BUFG_GTs in a Quad
- Assigns the clock root for both clocks close to the clock region containing the BUFG_GTs
Recommended: To avoid skew violations, AMD
highly recommends following this clocking topology when [RT]XUSRCLK2 operates at half
the frequency of [RT]XUSRCLK.