Meatsack summary / highlights:

The 5000 series has dropped support for 32-bit CUDA, causing e.g., 2010s-era games that implement PhysX to have to execute PhysX on the CPU, which can lead to atrocious performance (like a GTX 580 outperforming an RTX 5080) under bad in-game conditions – even on a high-end current-gen system – if you leave PhysX enabled and have a 5000-series GPU. Games designed with PhysX in mind are “experientially different” without PhysX and so disabling it is not desirable.

It’s not the biggest deal in the world that games from 12, 15 years ago don’t run on these modern GPUs - that’s not really the core complaint here. […] The real problem comes from everything that’s sort of attached to it. And the concern we have is the broader issue which is that Nvidia has a habit of trying to come up with some sort of exclusive graphics tech for games - and it’s the “exclusive” part that’s key there.

[…] This drags the industry along in Nvidia’s wake, and then when they get bored of the technology and abandon it, the customer’s left holding the bag.

This feels like a grim omen for any Nvidia special feature that’s integrated into a game, especially if it’s core to how that game feels.

[…] This is one of those where 10, 15 years in the future, we don’t know if features like Reflex, MFG, again even ray tracing will work the way they do today on the hardware that will be current in the future vs the stuff from the era they were developed. That’s just sort of the nature of hardware (compatibility mode has been a thing for ages) - so we’re trying to keep perspective on how just big of an issue is this, but it does come back to Nvidia and how much they push these specific technologies.