General rule is single threaded performance (and presuming sufficiently wide resources) trumps multi threaded performance (when in excess) for gaming, and people tend to look at the benchmarks for this metric.
At this point, AMD chips can mostly match Intel on these benchmarks, so people will look at current Ryzen scores, and think it should be the same. Yet even with equal single threaded scores, Intel chips tend to preform better (and in some cases a LOT better) because of the fact that current consumer Intel chips have very low latency communication between cores vs. the relatively high latency of the communication between different CCXs in the Ryzen die.
Take for example HUBs more recent video regarding deciding which platform to choose to benchmark new Nvidia GPUs.
Here we see that despite similar single threaded performance, the i9-10900k (at stock) beats the 3900x by about 5% at 1080p (not GPU limited) and 2% at 1440p (GPU limited); it’s worth noting that the 10900k has some overclocking headroom whereas the 3900x won’t really improve much from any overclock. Therefore these differences will be even more pronounced. It’s also worth noting the games selected here are likely biased towards omitting the larger performance differences (as much as 20%) in favor of Intel as shown here:
https://www.gamersnexus.net/hwreviews/3587-intel-core-i9-10900k-cpu-review-benchmarks
They would be doing a disservice to choose the 3900x, which clearly bottlenecks a 2080ti in more current titles, so test new GPUs, at least at 1080p and 1440p. That said, HUB is somewhat AMD biased and believe they’ll choose the Ryzen CPU because they know more people will enjoy it because of recent Ryzen popularity, despite the empirical evidence.
With as much as a 20% difference, they should use the 10900k. We’ll see what they do.
Hopefully, Zen 3 will help close this gap and potentially surpass Intel (see here: https://www.notebookcheck.net/Latest-Zen-3-rumor-further-reiterates-10-15-IPC-gains-per-thread-and-32-MB-of-shared-L3-cache-per-CCX-Zen-4-to-feature-1-MB-L2-cache-and-AVX-512-support.459845.0.html#:~:text=Zen%203%20will%20feature%20up,cache%20and%20AVX%20512%20support.
)
Zen 3 will feature up to eight cores per CCX and a 32 MB L3 cache that can potentially improve latency in gaming. Also being confirmed is that Zen 4 will require a new socket and feature a 1 MB L2 cache and AVX 512 support.
As far as PCIE3 vs 4? Well, 2080 ti barely completely saturates PCIE2 (for gaming, anyway), and PCIE3 is double the bandwidth of PCIE2. Given the 3090 is rumored to be between 40-80% faster than the 2080 ti, it’s unlikely the GPU will need MORE than double the bandwidth of PCIE3 (most of which is unused by current 2080 ti).