AMD: Devs only use PhysX for money

[Owenb]

I thought the SPE's in the PS3's cell could happily do it.

That's a console, not console's

The PS3 proves the point. Some first party games make good use of the vector units and they can do that by using it for physics. Fusion and Sandy Bridge are AMD/Intel's Cell- A complex CPU in charge of simple, fast number crunching units.

 

[Lightnix]

That's a console, not console's

The PS3 proves the point. Some first party games make good use of the vector units and they can do that by using it for physics. Fusion and Sandy Bridge are AMD/Intel's Cell- A complex CPU in charge of simple, fast number crunching units.

Hm, Sandy Bridge isn't really like that at all from what I can tell. It's just an evolutionary step in that it has more (although still homogeneous) cores, but what it does differently is double the width of its SIMD unit (from 128-bit to 256-bit, something that hasn't been done since the Pentium III with the introduction of SSE). The gains from both of these are massive (four times the maximum theoretical throughput clock for clock in SIMD heavy operations), but I'd argue not a massive revolution in architecture like you're suggesting.

 

[wolvers]

the only stipulation is they need to interface with it via CUDA.

Well it's not 'open' to me then is it!!

 

[starfighter]

I am more hopeful of an open platform than this stupid nVidia/PhysX way of doing things. however.

I am not sure about a separate card for Physics, it seems to me it should either be done on the CPU or GPU, you don't really need a 2nd card, especially one that is just a GPU used differently. why ot use the spare power of the 1st GPU. But then some people ike to spend money regardless, hence SLI/Xfire.

But it definitely will not be mainstream or make it to consoles. hell it is bad enough trying to get most computer manufacturers to put a decent (if any) graphics card into a "gaming machine". never mind a 2nd or 3rd.

And not forgetting computers are getting smaller/more portable and energy concious as well. It just isn't happening as far as more separate hardware goes. If you were "hardcore" and did want a 2nd card it would be better used in SLI/Xfire, but I suppose it makes old cards useful. That is it's only saving grace.

Games aren't CPU limited, surely if physics was hogging all the CPU they would be? People prefer things to actually look real rather than act real first. And besides games makers don't know what do with all these CPU cores, just use them for it.

 

[Strawberry]

Well it's not 'open' to me then is it!!

Yeah but you need to interface with DirectX on Windows, officially.

D3D is just as brand-limited as PhysX is. If D3D was as readily and powerfully available on eg. unix-based systems as it is on Windows, we would most likely see a lot more linux users (several of my friends and myself included).

The problem is just that the PC gaming industry has already migrated towards Windows, so D3D being restricted (officially) to Windows is not as much of a problem as PhysX being restricted to NVidia, because (currently) AMD/ATI is the more successful company, and forcing such a migration would be counter-productive.

 

[Phoenixdancer]

I purcahsed a 9800GT for £40 for the use in a Server PC and when i needed a Phsyx card...

To be honist i would rather a dedicated Physics card so my 5970 can use all its grunt on my games eyecandy

But thats me, Im an avid tinker man

BUT i would rather it not be Nvidias Phsyx i had to use... and have to run in loops to get it working with an ATI card.

We need a Code that will run with anything and still be able to use a dedicated card for it

 

[Strawberry]

We need modern dedicated physics cards. (At least until CPU-integrated GPGPU chips come along).

True, CUDA-only NVIDIA PhysX cards. That's something I'd pay for. No graphics driver, GPU, monitors etc bull**** just pure CUDA.

 

[kylew]

That's a console, not console's

The PS3 proves the point. Some first party games make good use of the vector units and they can do that by using it for physics. Fusion and Sandy Bridge are AMD/Intel's Cell- A complex CPU in charge of simple, fast number crunching units.

That would actually be "consoles".

 

[kylew]

PhysX is just as "open" as directx is in that regard, anyone can use the API if they wish, anyone can develop their own hardware implementation - the only stipulation is they need to interface with it via CUDA.

Come on, you know that's not true.

Look at nVidia trying to lock out ATi users from using a secondary nVidia card for physx.

You can't even say it's open providing you have a graphics card that can interface with CUDA.

 

[Lightnix]

Come on, you know that's not true.

Look at nVidia trying to lock out ATi users from using a secondary nVidia card for physx.

You can't even say it's open providing you have a graphics card that can interface with CUDA.

Edit: Actually I take this whole section that was here back. After a bit of reading it seems as though it might be possible for AMD to support PhysX legally, although the area looks a little grey.

I mean remember when that NGOHQ site said they were allegedly developing a CUDA layer for ATi hardware? And how they were allegedly receiving Nvidia help for it? Have you seen any updates on that since 2008? Didn't think so. That's because they were lying through their teeth about the whole ordeal. They were talking about developing it for the 3800 series of cards which don't even have support for shared memory, which is a requirement of CUDA (this is also why AMD aren't supporting the 3800 series with their OpenCL or DirectCompute drivers as they also require hardware shared memory support).

I brought that last point up because it's where the majority of 'Nvidia supports CUDA/PhysX on AMD hardware' statements started cropping up from. Unless someone wants to show me a press release from Nvidia, from 2009 onwards, that expresses openness to an AMD implementation of CUDA, then could we please stop making these ridiculous claims about Nvidia's open arms approach to this?

 

[Rroff]

Look at MS trying to lock out OpenGL (yes they have tried) my point wasn't how open or not PhysX was - that comparing it to DirectX as if that was "open" is inherently flawed.

 

[Rroff]

Anyone can approach nVidia in getting PhysX working on their hardware, the only stipulation is that you also have CUDA running on your hardware (PhysX sits ontop of this). The purpose of the terming in that EULA above is to prevent people writing unofficial wrappers to run code on another platform, if someone was to license PhysX for their platform then the EULA would be updated.

Aside from the fact AMD just can't be bothered with it, they also wouldn't be that happy about porting CUDA over as it would kill of their own stream entirely rather than the pathetic life it has now.

 

[kylew]

Look at MS trying to lock out OpenGL (yes they have tried) my point wasn't how open or not PhysX was - that comparing it to DirectX as if that was "open" is inherently flawed.

That in itself is completely irrelevant (irrelevant, not doesn't matter) to whether or not directx is open, it's an entirely separate situation.

Anyone can use DirectX, true, MS trying to lock out OpenGL doesn't suddenly mean that people can't use DirectX now does it?

 

[kylew]

Anyone can approach nVidia in getting PhysX working on their hardware, the only stipulation is that you also have CUDA running on your hardware (PhysX sits ontop of this). The purpose of the terming in that EULA above is to prevent people writing unofficial wrappers to run code on another platform, if someone was to license PhysX for their platform then the EULA would be updated.

Aside from the fact AMD just can't be bothered with it, they also wouldn't be that happy about porting CUDA over as it would kill of their own stream entirely rather than the pathetic life it has now.

There's no solid "truth" in that situation.

nVidia claim to have offered it without any restrictions or conditions, which is very unlike nVidia.

AMD claim that nVidia did this for PR, but behind closed doors didn't want to work with them.

Additionally, do you blame AMD for not wanting nVidia to have possible leverage over them? They could threaten to gimp PhysX if AMD didn't do what they wanted.

I'm not in to any closed standards, and that's exactly what PhysX is.

 

[sturmtruppe]

Anyone can use DirectX, true, MS trying to lock out OpenGL doesn't suddenly mean that people can't use DirectX now does it?

a mac user can't use directx.

 

[Lightnix]

Anyone can approach nVidia in getting PhysX working on their hardware, the only stipulation is that you also have CUDA running on your hardware (PhysX sits ontop of this). The purpose of the terming in that EULA above is to prevent people writing unofficial wrappers to run code on another platform, if someone was to license PhysX for their platform then the EULA would be updated.

Aside from the fact AMD just can't be bothered with it, they also wouldn't be that happy about porting CUDA over as it would kill of their own stream entirely rather than the pathetic life it has now.

Actually I did find one section in their CUDA license that would (at least safely) prevent AMD from developing a CUDA implementation for their own hardware, and I admit it's a stretch, but they have a defensive suspension section which basically says if AMD were to get into a legal battle with Nvidia for any reason, Nvidia has the right to terminate the license. That's not ideal, since large chip making companies are typically in legal battles with each other perpetually.

Edit: Actually the phrase is 'legal proceeding', that could mean anything. :\

 

[Rroff]

There's no solid "truth" in that situation.

nVidia claim to have offered it without any restrictions or conditions, which is very unlike nVidia.

AMD claim that nVidia did this for PR, but behind closed doors didn't want to work with them.

Additionally, do you blame AMD for not wanting nVidia to have possible leverage over them? They could threaten to gimp PhysX if AMD didn't do what they wanted.

I'm not in to any closed standards, and that's exactly what PhysX is.

Gimping it at a later date would be fairly hard due to the nature of the API - it hands off to the compute language indiscriminately - you'd have to go to massive length and bribe everyone developing with it to create configurations that wouldn't run optimally on 3rd party hardware.

 

[wolvers]

Look at nVidia trying to lock out ATi users from using a secondary nVidia card for physx.

You can't even say it's open providing you have a graphics card that can interface with CUDA.

Exactly, that was the nail in their coffin for me.

Of course there's the fact that it's pants, brings nothing to the gaming experience and seems to be constantly outdone by other solutions that don't force me to buy certain hardware!

 

[RavenXXX2]

Physx is dead, nothing of any substance coming out this year supports it, die...die POS.

 

[Mr Paul]

Anybody want a Ageia PPU?