Nvidia's RTX 50 series recently dropped, bringing with it quite a lot of disappointment and reigniting hope for AMD's Radeon 9070 cards. While Nvidia claimed "4090 performance" out of the 5070, that was only with the company's controversial mutli-frame generation. Alongside that, all of the cards are basically impossible to obtain, with sky-high prices to boot. However, that's not all of the controversy, and one particular point of contention with the series has me seriously worried for the future of Nvidia's GPUs.
With the launch of the RTX 50 series, Nvidia silently killed support for 32-bit OpenCL and CUDA, without any documentation to suggest that it was happening. This may not seem like a problem on the surface; after all, most platforms are working towards dropping 32-bit support these days anyway. However, it came with some pretty dire consequences.
If you want to play games with PhysX, you'll need to get creative
Creative solutions are needed
The first and primary casuality lies in PhysX. It's a real-time physics engine open-sourced by Nvidia, and the idea was that it could help to enhance physics in games like Metro: Exodus and Borderlands 2. Now, running those older games that use PhysX will render them borderline unplayable on RTX 50 cards, though you can turn off PhysX to get performance back at the cost of physics performance.
This has led to gamers coming up with creative solutions in order to get PhysX working with these cards, including using a second GPU to handle those specific calculations. There's no other way to get them working, as Nvidia didn't release any kind of compatibility layer to translate those calls to run on 64-bit CUDA cores.
What makes this especially confusing is that Nvidia didn't announce this change, nor has it drawn attention to the documentation where it was changed. Nvidia killed support for compiling 32-bit CUDA applications with CUDA Toolkit 12.0, released in December 2022, and Nvidia updated that announcement to say that 32-bit applications would not be supported on the RTX 50 series.
This change didn't just affect gamers either; other applications like PassMark's benchmark which utilized 32-bit OpenCL code as part of its computation process also stopped working. While some developers have rewritten their code for 64-bit, there are plenty of older applications that will likely never receive an update and can't be used on the RTX 50 series if they rely on 32-bit compatibility.
Nvidia could kill any feature
And some might be just as important
The reason this worries me is not strictly relating to PhysX. To be honest, I don't particularly care for it, and you can run those games with the feature disabled. However, Nvidia pushed developers to support it, and some gamers have said that there is a pretty big difference between when it's enabled and when it's disabled.
When Nvidia pushes developers to support Nvidia-specific features, it creates a kind of lock-in that ultimately means those games are only playable as long as those features are supported. PhysX was seen as "revolutionary" by many at the time, particularly thanks to the fact you could execute all PhysX calculations on a separate GPU to the rest of your rendering.
However, if Nvidia can take a feature that was once "revolutionary" and drop support for it, what does that mean for current features that are supported by hardware in their more recent cards? I'm thinking of features like ray-tracing, where Nvidia is pushing developers to support it and make use of the hardware for it. Games are largely timeless, and imagine you couldn't run Cyberpunk 2077 in its full glory in ten years time because a future Nvidia card drops support for ray-tracing?
I'm not saying that will happen, but this highlights yet another problem when it comes to technology and preservation. You can play those games on an RTX 50 series card, but you'll be playing them in a worse fashion than someone could just a few years prior. That's obviously not good, and pretty worrying to think about in the long-term for all kinds of reasons.
I also don't think that some features will be at risk, but it's unfortunate that Nvidia released the 50 series without support to even execute 32-bit code. When we've seen other companies drop 32-bit support, there have been stopgaps put in place to still support executing those applications in the form of translation layers or virtualization.
Nvidia isn't the only company to have done this
And that's why I'm worried
To be clear, this is far from the most egregious example of features being killed off causing problems. There are plenty of older DirectX games that are no longer playable without DXVK, along with some games from the 90s and 00s that use DRM (like StarForce with Trackmania) no longer supported by Windows. This isn't a unique-to-Nvidia problem, but there aren't really solutions being developed either.
In the case of those DRM systems no longer being supported, you can look at it from the perspective of security, but at least there are alternatives via the high seas if you still want to play those games. As for 32-bit support being dropped, you can look at it from the perspective of security as well, but the diference is there is no alternative. You simply can't use 32-bit PhysX no matter what you do, sans buying a second card and using it in your PC.
Then there's the issue of communication. Nvidia didn't announce this change, they just simply didn't work and it was confirmed by Nvidia publicly after the fact. For example, Batman: Arkham Origins won't even run the highest level of PhysX on the CPU, meaning that you could potentially lose access to the best version of that game entirely in the future.
Backwards compatibility is one of the biggest strengths of PC gaming, with multiple workarounds typically being available for older games that don't work. You can play an 8-bit or even a 16-bit game on your PC, but anything with 32-bit PhysX is simply out of the question currently for the RTX 50 series. That's a shame, and goes against what PC as a gaming platform has always been about.