Nvidia Ampere vs. AMD RDNA 2: Battle of the Architectures

I'm glad that I'm not in the market for a new video card at this time.
Certainly, not being able to get a hold of any of the new video cards is very frustrating.
But so is the choice between AMD and Nvidia this time around.
Nvidia's cards are better in almost every way, and have more software available so as to be more versatile.
But AMD's cards have more memory; even the lowest-end 6800 has more memory than any of the Nvidia cards except the very expensive 3090.
And, while GDDR6x on the 3080 and 3090 give more memory bandwidth, and Infinity Cache seems like a substitute that might not always be adequate, when the larger internal cache does make a difference, it would yield higher speeds.
 
Definitely enjoyed this article. 'Tried to read it all thoroughly, but would liked to have seen some theoretical best practices for optimising for each set of approaches (read architectural optimisations), and where the responsibility lays for accommodating said optimisations.

Like most software, you can just write code, compile it and it will run - but what about those optimisations, like putting BVH data into Infinity Cache? Can you force that, and does it make a difference?

Most end users have a very simplistic outlook on things and only care about what they can see with their eyes right now, the moment they plug in their video card, load the drivers and fire up their favourite game, and that pretty much favours Nvidia right now.
 
Back