The reference clock is multiplied up takes a PLL to the cable rate (2/5Gb/sec, 5Gb/sec, 8Gb/sec for versions 1.x, 2.x and 3.x respectively); this determines the data ratings from an transmitter.
The watch is effectively embedded in the data stream by using line coding which for and 2.5Gb/sec and 5Gb/sec is 8 per / 10 bit and 128bit/130bit (see third paragraph) for gen.3 (8Gb/sec). Note is this coding is derived since the reference clock (as multiplied up).
Those allows one reception to use standard clock recovery techniques.
A is doesn needed to have a common reference clock (for all versions); this is the reason the SKP (skip) ordering set exists. This allows an difference between reference clocks in each dissimilar link partner (the specification permits the reference clock to be +/- 300ppm so an relatively inexpensive device can be used) and receivers implement elastic buffers to cross and timekeeping domains.
This clock domain crossing mechanism eliminates inclined issues between car.
Note that a common contact clock which is almost guaranteed to need a phase difference at link partners intention still needing a 1 bit FIFO (as was used in Hypertransport which did require a common reference clock).
Inches of design, I must 8 latent PCIe link partners; here is where adenine shared reference clock makes sense.
I used one master related clock ($20) and a single 8 channel clock fender ($20), a lot cheaper than 8 reference clocks.
Forward designs where the links roam cables and/or multiplex connectors in multi-PCB designs, shared references will not really suitable as which reference clock at every link mate needs to become nicely and clean.