NVIDIA GeForce RTX 50 Series Faces Compute Performance Issues Due to Dropped 32-bit Support

[eidairaman1]

Thank you for your contribution to the discussion. It’s truly insightful and informative.

You're welcome

 

[sbacc]

Nvidia is just following Apple, Apple always provide past architecture support to their customer but only to a degree and deprecate relatively abruptly (reading this cuda 32bit story, Mac OSX Lion drop of support for pre Intel code came to mind).
That was the big difference with Microsoft that tend to keep support for much longer, not caring much about bloating their soft, but even them seem to have change and start taking a similar approach.
I would argue that, developing and releasing a translation layer to keep those old codes running would be a more cavalier approach. But this cost money for no real return and considering the un-charity level of big US corp today, this is a max-level pipe dream at this point...

 

[LittleBro]

Even though Nvidia had said it planned to end support for 32-bit CUDA apps, they should have clearly stated which version will be pilot in this case. AMD does not support OpenCL software-wise for years either, but hardware-wise still does. So Nvidia dropped even hardware support now.

Those who think that OpenCL is not being used anymore, ask all cryptominers.

As for Passmark software, it's piece of shit software and this has been known for years. I'd compare Passmark to GeekBench in terms of amplitude of "shitness".

Anyway, those who need 32-bit CUDA support can go for RTX 4000 series. They are no worse than RTX 5000 series, you really don't need MFG support for older games that rely on PhysX.

I agree with @Assimilator on that Nvidia should have made some workaround (transition layer) available to retain backward compatibility.

So, Nvidia invested all time, money and effort to AI stuff, dropped support for anything no longer related, thus minimizing costs for gamers hardware and maximizing for one for AI. Which tells no other story but that one in which Nvidia does not give a f*ck about gamers anymore.

And where does this all lead to? For those who can't clearly see, Nvidia aims to provide small ARM-based boxes, more like terminals, for users to connect to their GeForce NOW cloud computing service for playing games. Monthly fee included. Something like Xbox Pass with streamed games instead of rendered locally. People are more and more staggering to afford new GPUs, prices are insane. This will be Nvidia's response: get this small cheap box, you've got 6 months of membership fee included, then you pay monthly fee, so we can keep it raising and you won't be able to do anything about it since we no longer make gaming GPUs.