Nvidia offers PhysX support to AMD / ATI

http://www.custompc.co.uk/news/602205/nvidia-offers-physx-support-to-amd--ati.html

Logical move: Intel->Havok is BIGGER (and more potentially dangerous) than Nvidia; maybe Nvidia+AMD in this case could be a solution for Nvidia to have a market fot PhysX/Ageia acquisition.
In my opinion of course...

The whole thing is totally barmy to me.

At the moment and in the current situation the companies need to be looking at REDUCING the cost of PC gaming so it can survive.

Making a physics card mandatory to play games so they can sell more cards?

Dumb, dumb, dumb it will just push more people away, "Oh man I can't play without ANOTHER £150 upgrade, oh forget then, (just like they did with Crysis)."

If you don't need the PhysX (or PhysX enabled) card to play the game, why buy the card?

Trying to get people to buy 1/2/3 additional graphics cards instead of just one.

Plain stupid, budgets are tight enough already and going to get tighter.

Pushing expensive "Beyond HD" screen resolutions:

Pointless, the customer isn't stupid and 1080P HD is more than adequate and the law of diminsihing returns kicks in hard. Also 40+" screens for your PC just don't work as your sitting close the screen rather than in a living room where you are further away.

PC Gaming needs to be reaching out to find new markets, not trying to extract even more from it's shrinking 13-40 target demographic who are already moving to consoles. Parents strapped for cash are going to be looking at spending less on entertaining their kids. The last thing they want is never ending expensive upgrades.

Darkheart

Good news for me though - when it is released, I get the physics for free with my GPU.

I've got this feeling they will probably charge extra for the PhysX enabled cards and even so any game that requires a physics enabled card is going to probably lose a lot of sales.

Darkheart

The nice thing is that you won't need a seperate physics processor. Thats the whole point. It uses your GPU. I wasn't interested in CUDA because it was a Nvidia only development. But if Nvidia are going to enable everyone to use it then it shows a lot more promise :)

Dark:
I don't think you understand. There used to be a PhysX card. Nvidia bought the company, and is porting the physics engine to work on their GPUs. There will be a driver update for all 8800 users which will enable this physics engine in games which previously supported the PhysX processor and new games which use the PhysX API. There is no additional cost to you.

Additionally they're offering to let ATI support the API on their chips as well, so game developers won't have to code for six different physics engines.

The reason they can do this is GPU's are now a lot more like CPU's since they implemented shaders. You can run all sorts of code on them. And once this physics API is out there they'll be able to add stuff to future CPU's which will speed up the physics even more if need be.

That is a silly attitude to take. Graphics technology moves forward. People aren't going to have the same cards forever. They have to introduce the engine AT SOME POINT in order to get the user base out there to be large enough to start using it regularly, just as they did with pixel shaders.

And I'm sure developers aren't going to make games which only run with a PhysX card for another 2-3 years at least.

The 8800 card doesn't actually have any Phys X hardware at all in it though, does it?

Phys X already comes in a software implementation, the main problem is that independant benchmarks show not much difference when using the dedicated Phys X hardware:

http://anandtech.com/video/showdoc.aspx?i=2751&p=1

I still think it's moving in the wrong direction, forcing PC gamers to upgrade by adding new hardware features at a time when many are struggling to pay their mortgages and buy food seems silly to me; time will tell.

Darkheart

they can just have different Physics schemes. Kind of like Crysis with its low, medium, and high physics.

that way you can use your CPU in software for low and medium, and your fancy GPU in high. I don't really see there being a problem.

Whoever wrote this article has no knowledge of the subject. This article says nothing about PhysX running off of AMD/ATI graphics hardware.

PhysX is already free, and it can run off the CPU. NVidia will enable PhysX to run on their GPUs. NVidia will allow users to run PhysX off the CPU, just like Ageia did. NVidia is not going to allow AMD/ATI to implement PhysX on their GPUs.

I guarantee you NVidia and Intel engineers are not too happy right now about their recent aquisitions. PhysX and Havok are sh*t. Fortunately, the CUDA platform is open so that Newton Game Dynamics can be implemented with it. I am trying to get Intel to look at Newton. Once I have some more demos for them I think they will consider it.

the one with no plan of it are you actually Josh, sorry.

NVIDIA is writting a CUDA implementation of PhysX.
CUDA thought is not restricted to NVIDIA. NVIDIA will just not write the drivers for the others. But NVIDIA uses to AMD available tools for its compilation etc (DGB and and an opensource compiler tool) so the only thing ATI needs to do is write a PTX driver for their SM4 cards and they can support the GPU accelerated PhysX as well and I would bet with the known speedups of GPU driven maths (factor 10-100 of a single tesla over a 16 core supercomp), ATI will write one more sooner than later, as they will lose the bussines costumers which want to use the cards for more than graphics otherwise very soon which they can not afford.

I know that for sure, as a CUDA researcher from NVIDIA held a multi hour presentation today at our university and that was part of it.

Do you have a link for that?

No, the presentation is not openly available I fear. Don't think NVIDIA will that allow again after recent "leaks" due to such stuff.

That ATI has not written PTX capable drivers yet, I think, you already know.

NVIDIA is not preventing them or anyone else with SM4 chips writting their own drivers thought. As mentioned they use a tool that seperates CPU and GPU code which anyone can buy (DGB was its name I think, not sure as it was of little interest towards CUDA, its a thing mainly driver devs need to know) and the PTX compilation is done with an open source compilation solution (Openxxx64, can't remember the xxx part anymore).
The whole API documentation etc as well is freely available not only to licensees.

Physics on the GPU is all well and good, but won’t that use resources that would otherwise be used for rendering? So can you throw around 50,000 particles at 60fps while rendering at 1440x900x32 with 4x anti-aliasing, and with smoke, shadow and water shaders active at the same time? I’m confused…

nope it won't.
The plan is to add in low end cards (8400GS and the like, can be bought very cheaply, worlds cheaper than a PhysX card actually) for all additional stuff or of you have SLI using one for graphics and the other for physics and other stuff.

You can naturally run it parallel on the same card as well but I don't think game designers will do that unless they do casual game designing with medium graphics and using the rest of the GPU for usefull stuff. But otherwise you will just offload it on core 2 if present etc.

The drivers for this have been leaked - not sure what the stance on these forums is for leaked software so I won't link - but search for NVIDIA_physX_8_06_12_driver - it works on all G92 (8800GT, etc.) cards upwards but not on the old 8800GTS (G80) and 8800GTX/Ultra. You will also need CUDA enabled drivers - ~175.16 onwards.

It runs on the CUDA pipeline using the streaming processors - as very few games make 100% use of these for the most part you can get significant hardware acceleration of physics without a penalty on the rest of your rendering performance...

If you have SLI - you can still use both cards for rendering at full speed and you'd have even more slack space that could be taken up by the physics processing.