Nvidia dropped 32-bit CUDA support with the 5000 series, not just 32-bit PhysX

[evernessince]

It appears as though all 32-bit CUDA apps may be impacted instead of just 32-bit PhysX.

Source: https://nvidia.custhelp.com/app/answers/detail/a_id/5615/~/support-plan-for-32-bit-cuda
Source:

https://www.reddit.com/r/hardware/comments/1iz8v8p

Source: https://www.videocardbenchmark.net/directCompute.html

A portion of PassMark's test is in 32-bit, which results in the 4090 getting a higher score than the 5090:

Nvidia should have included a translation layer or something. This is just silly.

 

[lexluthermiester]

Nvidia should have included a translation layer or something. This is just silly.

I agree. This really is just silly. Not all math and compute tasks need to be in 64bit.

 

[x4it3n]

It's sad for backward compatibility etc. but all modern machines are using 64-bit now. It was supposed to happen sooner or later. Who knows when Microsoft will drop 32-bit support too..

 

[evernessince]

It's sad for backward compatibility etc. but all modern machines are using 64-bit now. It was supposed to happen sooner or later. Who knows when Microsoft will drop 32-bit support too..

Windows already dropped 32-bit support. The reason you can still run 32-bit apps is because windows emulates a 32-bit environment: https://learn.microsoft.com/en-us/windows/win32/winprog64/running-32-bit-applications

The problem at hand isn't that Nvidia ended support for 32-bit CUDA, it's that it did so without providing any fallback. As you pointed out everything is depricated sooner or later. That said, it's common good practice to provide some sort of fallback to ensure that you aren't pulling the rug out from any of your users. Aside from gaming, certain industries such as medical imaging, finance, and manufacturing use specialized 32-bit CUDA software that has been stable for years. They may not have the budget, time, or ability to switch to new software. This could put them in a crunch where they are unable to obtain hardware to replace potential failures (as limiting to older hardware only means they can only acquire from a constantly shrinking pool) or cost them a lot of capital in replacing the system unnecessarily.

In addition, it breaks dependencies chains where you have one or more Library or Plugin that's 32-bit CUDA. Loosing PhysX support in some games is a lucky example where the only thing dropping 32-bit CUDA support did was forgo PhysX hardware acceleration but in other applications it may result in the entire application breaking or impact every feature down the dependency chain that relies on 32-bit CUDA.

 

[Dr. Dro]

I've mentioned this before, but 32-bit Windows driver support was dropped back some 7 years ago in 2018, the final driver being the 391.35 release which is also the last release to support the Fermi generation. This is not all that different from AMD removing the Mantle runtime after the 19.4.3 driver, sometimes things have gotta go.

I honestly had to resort to the dum-dum move and ask Grok what applications haven't been updated, and it points me to either old video games or very old software from the 2007 to 2015 timeframe, stating that CUDA 10 (dated Sep 2018, so around the time the 391.35 driver released) was the last to support 32-bit, with CUDA 12 (the current release) dropping it as baggage altogether after it spent some time deprecated in the last release. Makes sense to me. The AI assistant basically considers all software that hasn't been updated since as either deprecated or abandoned, which also makes sense to me. It's been so bloody long, after all.

Either way, I guess the RTX 4090 found its ultimate niche as the last super powerful GPU compatible with it, and that should definitely help keep its price up.