i literally just posted a test of pcie 3.0 with the 1080 up about 5 posts!! so that question is answered now
a long time ago when 3.0 was coming out, i guessed that what would drive bandwidth needs would not be gpu's but rather storage. it did not take long for that to prove true. we see pcie storage quickly filling up lanes while gpu's don't even push a small portion of the bandwidth. (previous tests took titan x cards to x4 mode and showed almost no performance loss).
i will say the same thing about pcie 4.0 gpu's will probably never catch up to bandwidth available but storage options will catch up with it the same as it did with 3.0, as quickly as they can make drives fast enough to use the bandwidth. the nvlink is not really the same thing. 2 cards sharing data over pcie will be slowed by pcie. nvlink is a way to connect the cards directly so they can share data faster, much faster. not the same thing really as x16 lanes not being enough for a single card. we won't see nvlink on consumer cards anytime soon as it is just not needed. it's used for those super cards used for AI and science research and such.