There's a sliding sweet spot of graphics interface speed and memory size that is a couple of years behind the maximums in hardware. BUT by the time that sweet spot catches up with the interface speed and mem size, you've got it on a GPU that's a couple of years out of date relative to the current ones. So if you bought the next gen speed and RAM size, you can sorta play the games 2 years down the road, but you're only doing as well as the budget GPUs that are then current. Maybe when considering retro, if you buy the top end RAM and IFspeed, you get a hypothetical year more range upward, but it ceases to be a good experience, and lacks newer DX features, but for the "core" range of the architecture, it makes tiny percentages of difference vs the middle/standard RAM and common interface speeds. There's also a play-off where the more RAM you've got the less you need the high interface speed, and the higher interface speed, the less you need the huge RAM.
It's a bit of a crock really, continuing since AGP 1x, where there was a collective "Oh noes, AGP is really slow feeding those textures into my 8MB card every time a new texture needed." so AGP 2x development started and by the time it came out, RAM prices halved and cards had 16/32MB holding way more textures at once. However, the speed is eventually needed, and should there be huge disruptions in RAM supply or there's some physical brick wall to process shrinking, then at least there's a couple of years slack in the interface speeds to cushion the performance shock. Meanwhile the only cards that "benefit" in their current era are the budget cards with bare minimum RAM yet fastest interface, and by "benefit" I mean it makes them a little less awful.
Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.