Did nVidia cap software PhysX?

[FearMeAll]

This is starting to rekindle the anger I had towards nvidia with the 5800ultra..just lies and cheats with drivers..grrrrrrrrrrr. I thought this sort of crap was over.

[Ancient76]

ROFLMAO you don't even know what PhysX is and how it works... it can run in both GPU and CPU. When you choose GPU it runs on nVidia cards, when you choose CPU it runs on your CPU. Months ago it used multiple threads in CPU mode to gain advantage of multiple cores/processors. Now nVidia has removed this capability and forced it to run in a single thread, so CPU PhysX will artificially look much worse than GPU PhysX.

It was working before, and now nVidia capped it.

I think my grandma knows that!

Why would Nvidia optimize this game for AMD and Intel CPUs! Are they even qualified to do that? Is CPU designed for physics! No it's not, it's bottleneck.

The point of Nvidia Physx is to off load calculations from the CPU, because GPU is much better for accelerating physics.

I want that games run on GPU 100% if it's possible. I don't want to pay 300-400€ or more for some CPU, just to have 3-5 fps more.

[walterman]

Why would Nvidia optimize this game for AMD and Intel CPUs! Are they even qualified to do that?

Come on, any computer engineer can do this. And companies like nVidia have the best of the best.

Is CPU designed for physics! No it's not, it's bottleneck.

Neither the GPU is designed for physics, and the physics are a bottleneck for your framerate.

The point of Nvidia Physx is to off load calculations from the CPU, because GPU is much better for accelerating physics.

I want that games run on GPU 100% if it's possible. I don't want to pay 300-400€ or more for some CPU, just to have 3-5 fps more.

Yes, you can offload the PhysX calculations to the GPU, and lose a percentage of your framerate, while all your CPU cores are idling. Or you can use all those unused CPU cores to run some physics calculations, while your GPU keeps your framerate as intended. The key is to use all the computing power inside your machine.

And if you want to see an example:
http://www.coopdb.com/modules.php?name=Forums&file=viewtopic&p=13461#13461

My application runs faster with CPU+GPU, than with GPU only.

And about the main theme of the thread, yes, the scores of a core i7 running PhysX by software with 8 threads, could be interesting, and not so bad when compared to a GPU.

[uibo]

Why would Nvidia optimize this game for AMD and Intel CPUs! Are they even qualified to do that? Is CPU designed for physics! No it's not, it's bottleneck.


I think it was even mentioned in this thread that PhysX used to be multi-threaded but now is not. I'm not sure if it is just this game or all? Since GPUs are "massively parallel" processors using more than one thread should not be a problem.

[LordJuanlo]

Yeah, that's the main concern of this thread. We are not asking nVidia to optimize PhysX for multiple threads, because this WAS ALREADY DONE by Ageia. We are wondering why nVidia has disabled this capability, and our guess is that they want to make their GPU acceleration look much better than standard CPU PhysX, by making the last one much slower than it was originally.

[Heinz68]

NVIDIA: The Way It's Meant To Be Gimped. (http://www.urbandictionary.com/define.php?term=gimped)

4. Gimped

The practice of making software run poorly on a compatitor's hardware.
nVidia's involvement in the development of Batman Arkham Asylum has gimped antialiasing performance on ATi hardware and PhysX performance on multicore CPUs. They should rebrand their developer relations program as nVidia: The Wat it's Meant to be Gimped.

[Viral]

Seriously this has properly rekindled my disdain for nvidia that I first discovered some time back in late 2003 (took me a while to wake up and realise ATI were a viable alternative).

[Johnny C]

Yeah, that's the main concern of this thread. We are not asking nVidia to optimize PhysX for multiple threads, because this WAS ALREADY DONE by Ageia. We are wondering why nVidia has disabled this capability, and our guess is that they want to make their GPU acceleration look much better than standard CPU PhysX, by making the last one much slower than it was originally.

The reasoning is the same for all of nVidia's recent actions: Capping Physx (this thread), Disabling Physx w/ATi, Batman and AA fiasco, Intel cartoons which are less than tasteful, woodscrew engineering of the "Future"

They're just Spoiled Sports....so very sad....

[killahsin]

They did, they disabled the ability to use multiple threads when using CPU PhysX. This was enabled before nVidia bought AGEIA.

Um what? do you have quotes of the code that was removed?

to clarify physix is still optimized to use many 'threads' but consoles run 'physics' generally through one 'thread'. so my question again is was this code actually removed as you are all here claiming. Or was this just some corner cutting.

[LordJuanlo]

Did you actually look at the numbers posted on the first page of this post?.

I don't have the quotes but I have eyes. I remember running AGEIA demos in software, as well as Cell Factor (in software too) and I watched the CPU usage. And I saw all CPU cores working.

In Batman, when you enable full PhysX and you don't have GPU acceleration, you should get at least the same CPU usage than you get with PhysX disabled. AT LEAST. Furthermore, you should get MORE CPU usage because you are taxing your CPU when using full PhysX.

[Atomizer]

Did you actually look at the numbers posted on the first page of this post?.

I don't have the quotes but I have eyes. I remember running AGEIA demos in software, as well as Cell Factor (in software too) and I watched the CPU usage. And I saw all CPU cores working.

In Batman, when you enable full PhysX and you don't have GPU acceleration, you should get at least the same CPU usage than you get with PhysX disabled. AT LEAST. Furthermore, you should get MORE CPU usage because you are taxing your CPU when using full PhysX.

I think killashins point was cell factor might have been better optimized for multiple cores, where as batman might not, which makes physx irrelevant(who knows if thats true or not though)

[LordJuanlo]

All right guys, you can believe whatever you want to believe, I remember one of the features of Ageia PhysX, and it was its scalability on multiple CPU cores when using software mode. It was not written on the apps, but built in the PhysX engine.

No one cares anymore, it's owned by nVidia now and they can do whatever they want with it, and most people seems fine with what they are doing.

On first page it's very clear that PhysX off is using more CPU power than PhysX high, if you are fine with it, good to you

[Atomizer]

All right guys, you can believe whatever you want to believe, I remember one of the features of Ageia PhysX, and it was its scalability on multiple CPU cores when using software mode. It was not written on the apps, but built in the PhysX engine.

No one cares anymore, it's owned by nVidia now and they can do whatever they want with it, and most people seems fine with what they are doing.

On first page it's very clear that PhysX off is using more CPU power than PhysX high, if you are fine with it, good to you

Well obviously PhysX will be using less CPU with GPU physx enabled, think about that for a moment

[LordJuanlo]

Again... I'm not talking about enabling GPU PhysX, I'm talking about CPU PhysX. If you set PhysX to high AND you don't have GPU, all calculations that normally would be done by the GPU are done by the CPU. So... CPU usage should be higher than no PhysX!!!!!

Think about that for a moment

[Johnny C]

Again... I'm not talking about enabling GPU PhysX, I'm talking about CPU PhysX. If you set PhysX to high AND you don't have GPU, all calculations that normally would be done by the GPU are done by the CPU. So... CPU usage should be higher than no PhysX!!!!!

Think about that for a moment

I knew this thread was heading this way.....