Software Engineer: On the SSD side of things (referring to PCIE 5.0/6.0), the pure bandwidth improvement doesn't seem very useful right now, you want to focus more on average bandwidth, or in particular latency and general efficiency rather than just raw bandwidth and to some degree brute forcing things can help but something that's come up, on the PS5 they've talked about this thing on the GPU that clears the caches on loading stuff...and that's basically something you have to do when your doing DMA and your modern CPU will have efficient support for DMA, but that's being added to GPU's now and they need better optimisation for that...
I actually looked through RDNA 2 instruction set and I noticed they had things for clearing the whole cache but I couldn't see anything for selective cache clearing so that does seem to be PS5 specific...and that specific example, they might just copy it from PS5...and the reason you need to clear the cache is if you've sneakily transferred something from the background into main memory and you have a copy of the old data in cache that's now stale, you need to clear it from the cache if it is in the cache and if you don't know it's in the cache, the only thing you can safely do is to clear the cache or accept the fact that you're going to have graphics glitches. I'm not going to promise which way AMD will implement it into RDNA 3 but one of those two will be very likely