In other words upgrade your rig from 2008 that's on life support. Stick at least a 40 series in there, Windows 11, your flavor of modern AMD/Intel, and make something decent
That's not it at all. It means that if you want to keep running old software, you'll need old hardware (where 40 series GeForce is "old"). It's like how Windows recently dropped support to run 16-bit applications natively. If you have some ancient 16-bit DOS or Windows 3.1 application, you'll just have to run it in an emulator. Perhaps some clever person will write a wrapper for the 64-bit CUDA driver that allows old 32-bit applications to run, similar to the Glide wrappers for DirectX that you can use with PC games from the late 90s.
EDIT: Inside baseball type info - when Microsoft drops support for an operating system, that means, among other things, no more security updates and, importantly, no more development tool updates. So if I've been supporting 32-bit Windows systems with an application, even if I want to keep supporting it, I really can't, because the latest versions of Microsoft tools won't support it, and I can't keep my development environment stuck in the past. This is not really a big deal, since Windows 11 has robust backward compatibility with 32-bit applications. 16-bit support was killed with Win 11, and some companies did cry about it, but Microsoft justifiably told them to learn how to use DOSBox if it really matters that much.
Most of your standard hardware is fine. If you have a 32-bit application, don't worry, it can talk to the network card in 64-bit Windows just fine. There's a layer to translate one to another. Same with your mouse. I have a Microsoft Intellimouse from 2001 that still works in Windows 11. Even old 32-bit DirectX games still work in 64-bit Windows (which is why Mirror's Edge runs
at all). The problem is with anything that uses a fully proprietary protocol...like CUDA. That one is out of Microsoft's hands. NVIDIA's CUDA technology is 100% theirs, end to end, built by them, and provided by them. As of late last year, they can no longer build it for 32-bit systems.
The entity to blame for this is NVIDIA. When they were developing CUDA, many, many end users, including me, complained to them that a proprietary programming language is intrinsically risky, and they should really form a consortium with other companies, like Microsoft and the Linux Foundation, to build around OpenACC and develop the OpenACC standard. If they had done that, it would be on the operating system vendor to support it, and guarantee all-around better compatibility and support. Instead, NVIDIA bought the only cross-platform OpenACC compiler, PGI, and killed it to force people to CUDA. (The FTC never seems to consider this monopolistic behavior.)
So, here we are, exactly what everyone said would happen has happened, and old CUDA programs can't run on new hardware because NVIDIA has done everything to keep the API for its GPUs out of the operating system and in its walled garden. It's annoying, but unfortunately, its competitors in the accelerated compute space were all run by retards, so we're stuck with CUDA.
