Chester, if you have any updated information or additional insights on Nvidia's decision to basically dump PhysX support in their Blackwell GPUs*, please let us know. As of now, they won't even provide a wrapper; thus, a 5090 is, with the right-wrong game that wants PhysX, slower that an RTX 950.
* That "*" is here because I don't know if they also just dropped PhysX in their professional Blackwell cards. There is still some professional software that actually uses PhysX, so those customers wouldn't be pleased if their software doesn't run so well on their new Blackwell GPU.
NVIDIA have not dropped support for PhysX on Blackwell, this information is incorrect. PhysX utilises the CUDA runtimes and NVIDIA previously announced after Ada was released that they had deprecated the 32-bit CUDA runtime and would remove it in Blackwell. As a result of this, older games and potentially also some professional tools that still utilise 32-bit PhysX now have to use an alternative fallback path to x87 CPU instructions, as the AVX based path was added much later.
Obviously this is still problematic as most of these games and tools are long since out of support, and will not receive updates to newer versions which support AVX. This, coupled with most modern CPUs slowly removing x87 instructions, is why the performance is problematic in some older titles and tools.
I don't understand what the difference is supposed to be between dropping support for 32 Bit PhysX and removing it? Either means the same, Blackwell won't run 32 Bit PhysX.
The difference is pretty simple. PhysX still works for 64 bit applications, which is the vast majority of use cases used today. It is only older 32 bit applications that haven’t received updated in the last 3-5 years that will be affected by this.
Do you work for Jensen Huang? Why else would you just blame the users who stupidly expected that they can continue to run software that has worked just fine for them, including on Ada? That is, until Nvidia just decided to render it obsolete in Blackwell - no reason given. What would have been the harm in continuing support for 32 Bit PhysX with a wrapper? I haven't heard or read a single good reason why that would have been a bad thing.
I never made a statement on whether this was the correct move or not, nor did I make the statements you are implying I did. I explained what had changed to cause the issue. I would appreciate you not making aggressive accusations and statements about me. Regarding solutions - I cannot explain why NVIDIA chose not to provide one. I imagine it is likely for games and tools that are now having issues that some sort of DLL replacement can be done by modding communities to enable the more modern and much faster AVX based fallback path to mitigate this issue.
I loved this article, a ridiculous nod to the past! When I read the first paragraph and saw ‘68xx’ , I thought to myself “would NVidia actually change its numbering scheme from 6x8x to 68xx just to mess with AMD, considering AMD did that thing this last generation to match NVidia’s?” And then I figured out the April Fool’s nature of the article and chuckled. But, yes, I believe NVidia would do that for their upcoming RTX 6000 series if they ever saw this article, just to mess with AMD.
Ah yes... I fondly remember my 6800 GT, beautifully hand-modded with an Arctic cooler using a Molex connectctor re-soldered to the back of its board for space reasons :)
I still remember the Nvidia tech demo for the 6800 GPUs.
Chester, if you have any updated information or additional insights on Nvidia's decision to basically dump PhysX support in their Blackwell GPUs*, please let us know. As of now, they won't even provide a wrapper; thus, a 5090 is, with the right-wrong game that wants PhysX, slower that an RTX 950.
* That "*" is here because I don't know if they also just dropped PhysX in their professional Blackwell cards. There is still some professional software that actually uses PhysX, so those customers wouldn't be pleased if their software doesn't run so well on their new Blackwell GPU.
NVIDIA have not dropped support for PhysX on Blackwell, this information is incorrect. PhysX utilises the CUDA runtimes and NVIDIA previously announced after Ada was released that they had deprecated the 32-bit CUDA runtime and would remove it in Blackwell. As a result of this, older games and potentially also some professional tools that still utilise 32-bit PhysX now have to use an alternative fallback path to x87 CPU instructions, as the AVX based path was added much later.
Obviously this is still problematic as most of these games and tools are long since out of support, and will not receive updates to newer versions which support AVX. This, coupled with most modern CPUs slowly removing x87 instructions, is why the performance is problematic in some older titles and tools.
I don't understand what the difference is supposed to be between dropping support for 32 Bit PhysX and removing it? Either means the same, Blackwell won't run 32 Bit PhysX.
The difference is pretty simple. PhysX still works for 64 bit applications, which is the vast majority of use cases used today. It is only older 32 bit applications that haven’t received updated in the last 3-5 years that will be affected by this.
Do you work for Jensen Huang? Why else would you just blame the users who stupidly expected that they can continue to run software that has worked just fine for them, including on Ada? That is, until Nvidia just decided to render it obsolete in Blackwell - no reason given. What would have been the harm in continuing support for 32 Bit PhysX with a wrapper? I haven't heard or read a single good reason why that would have been a bad thing.
I never made a statement on whether this was the correct move or not, nor did I make the statements you are implying I did. I explained what had changed to cause the issue. I would appreciate you not making aggressive accusations and statements about me. Regarding solutions - I cannot explain why NVIDIA chose not to provide one. I imagine it is likely for games and tools that are now having issues that some sort of DLL replacement can be done by modding communities to enable the more modern and much faster AVX based fallback path to mitigate this issue.
Why ask a question then get angry when you don't get the answer you're looking for?
They just wanted to save some money...
I loved this article, a ridiculous nod to the past! When I read the first paragraph and saw ‘68xx’ , I thought to myself “would NVidia actually change its numbering scheme from 6x8x to 68xx just to mess with AMD, considering AMD did that thing this last generation to match NVidia’s?” And then I figured out the April Fool’s nature of the article and chuckled. But, yes, I believe NVidia would do that for their upcoming RTX 6000 series if they ever saw this article, just to mess with AMD.
Ah yes... I fondly remember my 6800 GT, beautifully hand-modded with an Arctic cooler using a Molex connectctor re-soldered to the back of its board for space reasons :)
I had a lower end one of those in a laptop. Was actually impressed with it at the time 😄.