The Exigent Duality
In the Dust - 15:36 CDT, 9/01/20 (Sniper)
I've been writing a lot lately about the concept of the "GPU of forever"-- such as here, here, and here. But I didn't realize we'd be hitting that point this month.

For those who haven't seen the presentation yet, these Ampere cards are stunning, offering nearly double the performance of Turing, at equivalent prices-- a claim which Richard Leadbetter has independently verified.

We're talking about "better image quality than native" reconstructed 2160p, at a locked sixty frames per second, across the board. Other than to support future API features, it's not apparently where to even go from there, for people like myself who aren't particularly sensitive to refresh rates.

I also wrote about the new consoles in this post, and how they would soon be left in the dust-- we're already at that point, and they haven't even launched yet: I wasn't expecting Ampere to be this big of a jump, yet here we are. And as ray-tracing loads increase in future games, the distance between Ampere and even the Series X is going to grow into a gulf.

For me, the most interesting specific technical detail from the presentation is that Ampere uses the Tensor cores to do ray tracing de-noising. It makes me wonder in how many other, not-yet-conceived-of ways this RTX architecture's pieces could feed off of each other.

I had been planning to hold pat on any PC upgrades until probably 2022, but Ampere has completely caught me off guard: I may just order a 3080 in these coming weeks-- then my current 2080 will go to my wife, her 1080 will go to my daughter, and my son will stay on his 2060. My daughter's current 1070 will go into a box as a backup GPU.