Details on Nvidia PhysX - please read

[kid41212003]

ATI refused PhysX in first place, one of the guy from ngohq modified the driver that made PhysX cards work with ATI card, and he even got support from NVIDIA. I believe now he got a job from NVIDIA.

imagine how ATI get sued for putting a physX logo on a 5870 box, if nvidia didnt get royalties. free to use the software, is not the same as 'free to sell hardware advertising it to work with our stuff'

Before they can do that, they probably contact NVIDIA first, and I mean if they really want to put PhysX logo on their boxes, which I think they never want to.

 

[Mussels]

ATI refused PhysX in first place, one of the guy from ngohq modified the driver that made PhysX cards work with ATI card, and he even got support from NVIDIA. I believe now he got a job from NVIDIA.

please read my posts on the previous page. 'free to run' is not 'free to advertise' or 'without limitations'


look at it in another light: mac vs windows.

its like microsoft saying: sure thing man, you can run ALL windows apps for free on mac OSX.

However, our logo must be on all macs that support this - and you gotta pay royalties for it. Oh and YOU have to guarantee it works perfectly on all windows programs, not us. Oh and we make no assurances that it wont be optimized for our hardware over yours.

 

[newtekie1]

let me put it this way: do you really think nvidia would allow the following for free, given a choice:

Their logos on ATI hardware boxes

Absolutely, I'd love my logo on my competitors product

ATI support/branding on games

What does this have to do with PhysX?

driver/tech support to help them code cuda to work on STREAM

And why does CUDA have to run on STREAM? I mean, I know it doesn't, but I'd like to know why you seem to think it does.

test games before release to make them work on ATI cards

The beauty of CUDA. If CUDA works on ATi hardware, and PhysX works on CUDA, and PhysX works in the game, then there is no additional testing needed. CUDA apps are rather hardware idependent.

please read my posts on the previous page. 'free to run' is not 'free to advertise' or 'without limitations'


look at it in another light: mac vs windows.

its like microsoft saying: sure thing man, you can run ALL windows apps for free on mac OSX.

However, our logo must be on all macs that support this - and you gotta pay royalties for it. Oh and YOU have to guarantee it works perfectly on all windows programs, not us. Oh and we make no assurances that it wont be optimized for our hardware over yours.

Ignoring the obvious wrong that assumes Microsoft controls all Windows apps...

Yeah, it is kind of like Apple advertising bootcamp can run Windows and all Windows apps for free...oh you probably didn't mean it like that...

 

[Mussels]

Absolutely, I'd love my logo on my competitors product

Nvidia would love their logos on ATI hardware. ATI would not love it, nor would it be free.

What does this have to do with PhysX?

PhsyX titles are TWIMTBP titles. if ATI supported physX and it became popular, they'd be killing the GITG program and basically giving NVidia advertising rights to every single game.

And why does CUDA have to run on STREAM? I mean, I know it doesn't, but I'd like to know why you seem to think it does.

I may have made a typo - i should have said PHYSX needs to run on stream

The beauty of CUDA. If CUDA works on ATi hardware, and PhysX works on CUDA, and PhysX works in the game, then there is no additional testing needed. CUDA apps are rather hardware idependent.

Alright, fair point. what about performance issues? what about limitations to how much can run at any one time on various GPU's? like, lets say nvidia make it so they can handle 500 phsyX items at once and ATI can only handle 400 - would nvidia tell game devs to hold back for ATI's sake, or would they say "buy nvidia cause we do it better!"

 

[kid41212003]

It's for NVIDIA cards, they bought the company, they hired people to made it work with their GPUs.

It's like saying Eye something is not work on NVIDIA cards.

I'm always think of it this way:

I get this when I choose this brand, and I get another "different" thing when I choose a different brand.

But still, it's not like when you have an ATI card you can't play games that have PhysX.

Adobe wrote a translator to translate programs code (not originally for Istuffs) to work with Apple, and Apple updated their Ipad so it will not work with programs using that translator. That's what I call BS.

 

[bobzilla2009]

Alright, fair point. what about performance issues? what about limitations to how much can run at any one time on various GPU's? like, lets say nvidia make it so they can handle 500 phsyX items at once and ATI can only handle 400 - would nvidia tell game devs to hold back for ATI's sake, or would they say "buy nvidia cause we do it better!"

the latter by a long way, although ati gpu's have always been good with regards to computing. So the only way they could force a hd5 series card to run terrible is to make it so only the gtx470 and gtx480 could run the game at all or make sure that the physX code just doesn't run too well on ATi cards. But it is nvidia after all, so they would do that.

Directcompute is the way to go, maybe openCL if they get their act together a bit quicker.

 

[newtekie1]

Nvidia would love their logos on ATI hardware. ATI would not love it, nor would it be free.

You say it wouldn't be free, I say it would. We'll just have to agree to disagree, because there isn't anything really supporting either side, since the deal never got that far along.

PhsyX titles are TWIMTBP titles. if ATI supported physX and it became popular, they'd be killing the GITG program and basically giving NVidia advertising rights to every single game.

They've already killed the GITG program, I can't even remember the last game that was part of the GITG program... ATi has taken a strong stance on not helping developers to optimize games, and instead optimizing the drivers for the games themselves.

You are correct, nVidia would have been able to advertise PhysX for every game that uses it, but ATi could have advertised they supported it...well you say they wouldn't have...

I may have made a typo - i should have said PHYSX needs to run on stream

Still don't see what Streams has to do with any of this...

Alright, fair point. what about performance issues? what about limitations to how much can run at any one time on various GPU's? like, lets say nvidia make it so they can handle 500 phsyX items at once and ATI can only handle 400 - would nvidia tell game devs to hold back for ATI's sake, or would they say "buy nvidia cause we do it better!"

I've already talked about the performance issues that many bring up, so I'm not going to rehash that. And now you are just trying to make up illogical reason to make nVidia look bad, there is nothing even suggesting that nVidia would hinder performance on ATi cards. Plus, as already mentioned, a 3rd party was going to be doing the developing, so nVidia couldn't do any funny business even if they wanted to.

 

[human_error]

You say it wouldn't be free, I say it would. We'll just have to agree to disagree, because there isn't anything really supporting either side, since the deal never got that far along.


They've already killed the GITG program, I can't even remember the last game that was part of the GITG program... ATi has taken a strong stance on not helping developers to optimize games, and instead optimizing the drivers for the games themselves.

Before people start saying "nvidia would let ati use physx" or "ati don't help developers at all" I suggest everyone reads http://www.bit-tech.net/bits/interviews/2010/01/06/interview-amd-on-game-development-and-dx11/1 where it is claimed nvidia refused to let ati run physx on ati hardware at all, and goes into detail on how ati helps developers to ensure compatability with ati hardware, but some developers do not talk to ati at all (TWIMTBP games developers usually) and so ati have to release driver patches/update to ensure compatability after the game has launched.

 

[Mussels]

newtekie: ATI stream is ATI's version of cuda.

They cant run cuda, so anything running, would have to run on stream...

are you seriously this in-depth in the discussion, with no idea what ATI stream is, or how its different to CUDA?

 

[bobzilla2009]

The GITG program = AMD: The future is fusion

 

[newtekie1]

newtekie: ATI stream is ATI's version of cuda.

They cant run cuda, so anything running, would have to run on stream...

are you seriously this in-depth in the discussion, with no idea what ATI stream is, or how its different to CUDA?

I know exactly what Stream is, yet I still fail to see what it has to do with PhysX. Why exactly does PhysX have to run on Stream is my question?

 

[bobzilla2009]

I know exactly what Streams is, yet I still fail to see what it has to do with PhysX. Why exactly does PhysX have to run on Streams is my question?

because stream is the ati version of cuda, it's how such programs interact with the gpu.

 

[Mussels]

I know exactly what Stream is, yet I still fail to see what it has to do with PhysX. Why exactly does PhysX have to run on Stream is my question?

because otherwise ATI cant run it?

 

[newtekie1]

because otherwise ATI cant run it?

So you're just ignoring the 3rd party adapting ATi's drivers to natively support CUDA then?

 

[bobzilla2009]

So you're just ignoring the 3rd party adapting ATi's drivers to natively support CUDA then?

when i actually see that proven i have to just dismiss it. All the evidence for the physX adaptation was one picture. One that could and certainly would have been photoshopped. If it was so effective it WOULD have been 'leaked' somehow, no one can throw around a lawsuit if they don't know who did it.

Also the excuse was that they couldn't get a hd4870, and thats why it fell apart. O RLY? if they wanted to, they could have just bought one. It reeks of general nonsense and veiled propaganda imo. Unless of course it's a new team doing it now? and not the one from 2-3 years ago?

I read a lot about that so called team trying it before and it all seemed like bull to me, especially the 'nvidia were very eager to help us port it to ati cards for free'...

 

[Mussels]

So you're just ignoring the 3rd party adapting ATi's drivers to natively support CUDA then?

there is simply no way CUDA ran natively on ATI.

Emulated, ported to STREAM, ran on some early beta of directcompute or openCL, whatever it was, it was NOT natively ran.

 

[human_error]

So you're just ignoring the 3rd party adapting ATi's drivers to natively support CUDA then?

Using an abstraction layer to translate calls in one language/syntax to another one is emulation, not native support (plus it was physx which was translated, not cuda). They were essentially emulating cuda support by making stream calls which were the equivalent to the cuda calls, however this was never released and so it is likely that there is not an exact stream equivalent to every cuda command, and visa-versa. This means that the abstraction layer would have to add more calls to emulate cuda sufficiently enough for physx to run and due to the extra work needed it would have not been very efficient or fast compared to running on a cuda platform.

Plus if it is unreleased with no comprehensive proof then it means nothing (didn't nvidia hire the guy who apparently made the abstraction layer on the condition that he never release it so to prevent any possibility of ati physx support).

 

[Mussels]

i do get sick of people quoting one third party website which released nothing than a few possibly faked screenshots... i mean shite, the screenshots showing an ATI video card with boosted scores from physX was possible with an ATI and nvidia card in the same system, could have been as lame a hoax as that.

 

[newtekie1]

when i actually see that proven i have to just dismiss it. All the evidence for the physX adaptation was one picture. One that could and certainly would have been photoshopped. If it was so effective it WOULD have been 'leaked' somehow, no one can throw around a lawsuit if they don't know who did it.

Also the excuse was that they couldn't get a hd4870, and thats why it fell apart. O RLY? if they wanted to, they could have just bought one. It reeks of general nonsense and veiled propaganda imo. Unless of course it's a new team doing it now? and not the one from 2-3 years ago?

I read a lot about that so called team trying it before and it all seemed like bull to me, especially the 'nvidia were very eager to help us port it to ati cards for free'...

there is simply no way CUDA ran natively on ATI.

Emulated, ported to STREAM, ran on some early beta of directcompute or openCL, whatever it was, it was NOT natively ran.

there are some issues that need to be addressed, since adding Radeon support in CUDA isn’t a big deal - but it’s not enough! We also need to add CUDA support on AMD’s driver level and its being addressed as we speak.

Note he says CUDA, not PhysX.

And it wasn't just an issue of getting an HD4000 series card, I'm pretty sure nVidia bought him one pretty quickly, the issue came down entirely to adding CUDA support to ATi's drivers.

 

[bobzilla2009]

i do get sick of people quoting one third party website which released nothing than a few possibly faked screenshots... i mean shite, the screenshots showing an ATI video card with boosted scores from physX was possible with an ATI and nvidia card in the same system, could have been as lame a hoax as that.

yes, and especially since a guy working at ATi later in year (i'll try and find the interview) said specifically that nvidia told them to 'go whistle' (exact words) when they asked about the physX on ATi cards. Nvidia are the gits, and the fact that ATi have outsold them for 2 generations is making them worse.

Note he says CUDA, not PhysX.

why did he stop? because they couldn't get a hd4870! lame excuse so they could end the hoax. I'm sure nvidia would have bought them one too, so why didn't they? hmm?

 

[human_error]

yes, and especially since a guy working at ATi later in year (i'll try and find the interview) said specifically that nvidia told them to 'go whistle' (exact words) when they asked about the physX on ATi cards. Nvidia are the gits, and the fact that ATi have outsold them for 2 generations is making them worse.


why did he stop? because they couldn't get a hd4870! lame excuse so they could end the hoax. I'm sure nvidia would have bought them one too, so why didn't they? hmm?

I posted the interview link already :-

Before people start saying "nvidia would let ati use physx" or "ati don't help developers at all" I suggest everyone reads http://www.bit-tech.net/bits/interviews/2010/01/06/interview-amd-on-game-development-and-dx11/1 where it is claimed nvidia refused to let ati run physx on ati hardware at all, and goes into detail on how ati helps developers to ensure compatability with ati hardware, but some developers do not talk to ati at all (TWIMTBP games developers usually) and so ati have to release driver patches/update to ensure compatability after the game has launched.

The guy stopped development of the app because he was hired by nvidia - the best way to stop development was to hire the guy as a bribe to stop him developing further or releasing any code.

As for the cuda vs physx emulation the guy claimed to have he was referring to cuda call support on ati cards which meant translating cuda calls to stream calls for the purpose of running physx. It could be expanded further (any cuda apps, not just physx) but would require more work as every possible cuda call would have to be translated into the correct stream or series of stream calls (of which there could be multiple solutions with different levels of accuracy or efficiency, which would need to be selected differently depending on the task - which is a lot of work).

 

[bobzilla2009]

I posted the interview link already :-

sorry must have missed it

anyway, the post of the cuda on ati drivers was one picture with lots of promises of videos and proof, and even a driver release about 3 days after! None of it happened at all. So it obviously never worked.

 

[newtekie1]

yes, and especially since a guy working at ATi later in year (i'll try and find the interview) said specifically that nvidia told them to 'go whistle' (exact words) when they asked about the physX on ATi cards. Nvidia are the gits, and the fact that ATi have outsold them for 2 generations is making them worse.



why did he stop? because they couldn't get a hd4870! lame excuse so they could end the hoax. I'm sure nvidia would have bought them one too, so why didn't they? hmm?

That was a very early reason, a hurdle which was quickly cleared. Again, the reason for stopping developement was ATi would not add, or allow him to add, support for CUDA to their drivers.

I posted the interview link already :-



The guy stopped development of the app because he was hired by nvidia - the best way to stop development was to hire the guy as a bribe to stop him developing further or releasing any code.

As for the cuda vs physx emulation the guy claimed to have he was referring to cuda call support on ati cards which meant translating cuda calls to stream calls for the purpose of running physx. It could be expanded further (any cuda apps, not just physx) but would require more work as every possible cuda call would have to be translated into the correct stream or series of stream calls (of which there could be multiple solutions with different levels of accuracy or efficiency, which would need to be selected differently depending on the task - which is a lot of work).

He continued developement under nVidia, even gave several interviews, and nVidia publicly said they supported his efforts. He wasn't exactly hired by nVidia either, like he wasn't directly working for them, he was added to their developement team, kind of like s sub-contractor.

And the interview, and the comments in it, were from waaaay after all of this. After ATi had already shut the door on CUDA running on ATi hardware natively, and nVidia in turn entirely cut PhysX support if ATi hardware was detected. On dick move leads to another....and another...and another... I hate politics.

 

[Mussels]

what you're talking about here, is the same as running x86 software on itanium, or an xbox360 game on a mac.

there is simply NO WAY cuda ran natively on ATI hardware.

I'm seriously sorry if you fell for the crap, but it just screams hoax.

 

[bobzilla2009]

That was a very early reason, a hurdle which was quickly cleared. Again, the reason for stopping developement was ATi would not add, or allow him to add, support for CUDA to their drivers.

He continued developement under nVidia, even gave several interviews, and nVidia publicly said they supported his efforts. He wasn't exactly hired by nVidia either, like he wasn't directly working for them, he was added to their developement team, kind of like s sub-contractor.

And the interview, and the comments in it, were from waaaay after all of this. After ATi had already shut the door on CUDA running on ATi hardware natively, and nVidia in turn entirely cut PhysX support if ATi hardware was detected. On dick move leads to another....and another...and another... I hate politics.

Maybe ATi would lose the ability to use Havoc on their gpu's or something similar, but any real proof that it did work would have been nice. Anyway, the future of physics in games is either in open source or just putting it back onto the ol' cpu. It's not like they have to do much atm. The most demanding games like crysis rarely hit 40-50% cpu load on most decent processors, even with a standard (in the case of crysis, pretty decent) physics engine.

Also, why would ATi say that nvidia wouldn't help them port physX to ati cards AT ALL, but nvidia were actively supporting a random guy on the internet porting CUDA? Is it because CUDA would run horribly if it was ported, whereas physX would run very very well on stream? i think so, and it makes sense.