NVIDIA shows signs ... [2008 - 2017]

[Jawed]

CUDA doesn't really offer anything that OpenCL does and OpenCL implementations will be available from all the hardware vendors.

I think the next version of CUDA will see some specifics that take performance way beyond what OpenCL provides for, e.g. if the hardware has built-in support for Scan then a wealth of key algorithms just got a major boost.

http://forum.beyond3d.com/showpost.php?p=1327227&postcount=1936

OpenCL, if it ever gets such a primitive, will be years behind and programmers using it will be eyeing the CUDA variants of their Scan-based algorithms with deep envy in the mean time.

Of course, it's still a guess as to whether Scan will be in CUDA 3.0, or how sophisticated it will be. But when NVidia owns the entire house it can party any which way it wants, unrestricted by committee.

So, how long before OpenCL is "mature"? Or, how long before OpenCL is dropped in favour of ...

Also, separately, I guess NVidia's strategy with PhysX, now that it is fully proprietary, is to get game developers to build gameplay physics, not just effects physics. Totally cutting off non-NVidia gamers. I wouldn't be surprised if Futuremark's game, whatever it's called, goes that way.

If Intel thinks it has a chance to get game developers writing purely Larrabee-specific code, then why doesn't NVidia have the same chance?

Jawed

 

[ChrisRay]

Missed it, sorry about that.

One other quick comment before I head off for the day. I noticed your pointed out that AMD has OpenCL drivers and demos. I know this. And they actually porting Havok to OpenCL. The question is. Why couldn't have AMD done this with PhysX?

Would Nvidia not let them? Did AMD Not want too work with Nvidia? DId ATI Not want to work with Nvidia? AMD chose to ally with Intel. Which has done everything in its power to run AMD into the ground.

In my mind this has less to do with technicalities such as "Propietary" ect. IE Havok is Propietary as well. ((Because when it comes Physics right now Intel/Nvidia control where they are headed. Not AMD which simply licensed it)).

I think it simply comes down to politics between the 3 companies. And these politics are coming from all 3 companies right now. I know Nvidia wants to be the driving force behind Physics and more specifically GPU Computing. So PhysX is highly optimised/designed with the GPU in mind in any recent title. While I dont think that is technically within the same interest of AMD/Intel right now which are still specifically CPU companies with GPU Subset ((building GPU subset)).

Once we see Intels GPU we will see if theres a political twist from intel regarding the nature/uses of GPU. I personally suspect there will be. Specially when you look at the design of it.

 

[neliz]

One other quick comment before I head off for the day. I noticed your pointed out that AMD has OpenCL drivers and demos. I know this. And they actually porting Havok to OpenCL. The question is. Why couldn't have AMD done this with PhysX?

Because as of now, PhysX runs on proprietary code? CUDA? nV tried this "oh, but everyone can run PhysX!" trick before. the AMD/Havok demonstration was a cooperation and not an AMD only affair.

 

[ChrisRay]

You missed the point Neliz. What prevents AMD from porting PhysX to OpenCL? Assuming AMD has licensed PhysX instead of Havok. Havok is Propietary technology that AMD licensed. Havok is not an open Standard. It is run and driven by Intel. Havok/PhysX are both closed standards Run by 2 competing companies. ((Intel/Nvidia)). You have to license either of them.

 

[Brad Grenz]

What prevents AMD from porting PhysX to OpenCL?

Are you really that naive? What prevents AMD from getting Physx accelerated on their GPUs is nVidia's interest in keeping the technology proprietary to their own hardware. They've already demonstrated the will to block AMD hardware from using something as basic as MSAA in TWIMTBP games using nothing more than a vendor ID check. What makes you think nVidia would just stand there and let AMD negate one of their last marketing advantages?

 

[ChrisRay]

Your operating under the assumption that Nvidia would not license PhysX to AMD. Licensing Royalties alone can be very profitable. But at leasts thats at least a better argument than "PhysX is propietary and Havok is Open" statements. I dont agree that this is the case. But at least I can see the argument. Do you guys truly believe that there was never even a discussion between AMD/Nvidia regarding the usage/licensing of PhysX?

 

[Kaotik]

Your operating under the assumption that Nvidia would not license PhysX to AMD. Licensing Royalties alone can be very profitable. But at leasts thats at least a better argument than "PhysX is propietary and Havok is Open" statements. I dont agree that this is the case. But at least I can see the argument. Do you guys truly believe that there was never even a discussion between AMD/Nvidia regarding the usage/licensing of PhysX?

Last time nVidia claimed that PhysX, including hardware implementation on video card, is "free for all", but you need to support CUDA

 

[ChrisRay]

Last time nVidia claimed that PhysX, including hardware implementation on video card, is "free for all", but you need to support CUDA

Thats very true. And a very good point. This is actually an issue I intend to raise when I talk to the PhysX/CUDA driver guys today. Because OpenCL may in itself change this. We'll see. I intend to try and pick Nvidia's brain today about PhysX/CUDA/OpenCL. Maybe i"ll learn something I can talk about. Since this was written before OpenCL finalized. Things might have changed.

 

[neliz]

You missed the point Neliz. What prevents AMD from porting PhysX to OpenCL? Assuming AMD has licensed PhysX instead of Havok. Havok is Propietary technology that AMD licensed. Havok is not an open Standard. It is run and driven by Intel. Havok/PhysX are both closed standards Run by 2 competing companies. ((Intel/Nvidia)). You have to license either of them.

My point was exactly what's described below. As soon as PhysX runs on OCL, why would AMD have to invest in it? At most some optimizations with developers that aren't payed for in twimtbp.

 

[Davros]

They've already demonstrated the will to block AMD hardware from using something as basic as MSAA in TWIMTBP games using nothing more than a vendor ID check.

why would they do this chris - and can you tell them to stop it ?

 

[dkanter]

I asked about PhysX running on AMD GPUs once and the GMs and PMs at NV said very clearly that the only way they would let that happen is if CUDA was supported on AMD devices.

AMD has no reason to see CUDA as a programming interface for their GPUs, since that would just enable NV's compiler to emit code that is bad for AMD GPUs, or require features missing in ATI GPUs.

Bottom line - ATI would have to be a bunch of idiots to promote CUDA, but that was exactly the price NV wanted for PhysX. It should be obvious why ATI chose to work with Havok.

Intel and ATI had always had a pretty good working relationship before the M&A, and the M&A didn't change that, and in many ways improved it.

David

 

[Davros]

found this :

Instinct Technology has collaborated with Dark Water Studios to create a movie showing how their middleware product Instinct Studio is able to use NVIDIA’s CUDA technology to offload AI from the CPU onto the GPU of the graphics card. The trailer shows GPU-controlled flocking behavior in DogFighter that allowed them to display a ridiculous (as in over 4000) number of aircraft on the screen at once, all controlled by a single CUDA-enabled graphics card.

The demo shows 4096 bot planes handled solely on a single GPU parallel to game rendering. The bot planes use steering behaviours for flocking, navigation and obstacle avoidance. The planes are fully lit and rendered (with shadows). The demo runs with interactive frame rate on main stream CUDA enabled graphics cards. In comparison the same simulation without utilizing CUDA achieved a similar frame rate on a decent machine with only 512 planes in our tests. The steering computation for 512 planes requires about 260.000 neighbour queries while for 4096 planes this grows to a whopping 16 Million queries. The algorithm can be easily parallelized, explaining the advantage of technologies like CUDA for this kind of problem. Even if there is a potential to optimize the algorithm for CPU the clear benefit for us is a heavily reduced development time.

 

[ChrisRay]

My point was exactly what's described below. As soon as PhysX runs on OCL, why would AMD have to invest in it? At most some optimizations with developers that aren't payed for in twimtbp.

In Order to use PhysX AMD has to license it. Just like Havok. And yes there is driver side work AMD would have to do to make PhysX work for them. Hell they are porting the entire Havok Library to make it work for them in OpenCL.

 

[digitalwanderer]

Again it begs the question though Chris, why would AMD support an API that their competitor controls? :|

 

[ChrisRay]

Again it begs the question though Chris, why would AMD support an API that their competitor controls? :|

Why Dont you ask Intel that? Whether ATI Division had a "Good" Relationship with Intel is really besides the point. ATI Division is just as much Intel's competitor As Nvidia is. And Intel is just as cut throat as you guys believe Nvidia is. Intel is guiding force behind Havok and they control its direction. AMD is Busy porting Havok to OpenCL.

Havok is not Open Source, Havok is not free to use. The question is redundant because AMD has already made this compromise. They just compromised with intel. Not Nvidia. If Nvidia Compromised. And let AMD port PhysX on their own to OpenCL with a license. How would this be any different? Or would this not be enough?

Ok. I get it. AMD doesn't want to use CUDA. And PhysX is on CUDA. But it doesn't neccasarily have to be and could possibly change in the future. I have seen nothing that suggests this is a completely closed door. And AMD has even spoken of the possibility of future talks with Nvidia an extremetech article posted back in March. I am in good faith here trying to listen to the arguments of why you guys dislike PhysX and what can be done to improve the situation. But some of the comments here have me scratching my head.

 

[digitalwanderer]

I believe ATi is the ones who initially dealt with Intel over Havoc, not AMD...they just got it along with ATi.

Not a big difference, but enough of a significant one I should think.

Ok. I get it. AMD doesn't want to use CUDA. And PhysX is on CUDA. But it doesn't neccasarily have to be and could possibly change in the future. I have seen nothing that suggests this is a completely closed door. And AMD has even spoken of the possibility of future talks with Nvidia an extremetech article posted back in March. I am in good faith here trying to listen to the arguments of why you guys dislike PhysX and what can be done to improve the situation. But some of the comments here have me scratching my head.

You really think nVidia would play nice with PhysX if AMD licensed it? What colour is the sky in your world Chris? :|

 

[ChrisRay]

I believe ATi is the ones who initially dealt with Intel over Havoc, not AMD...they just got it along with ATi.

Not a big difference, but enough of a significant one I should think.


You really think nVidia would play nice with PhysX if AMD licensed it? What colour is the sky in your world Chris? :|

That is not how it went down.

http://www.extremetech.com/article2/0,2845,2324555,00.asp

And Digi. Such is the price you pay of not having your own Physic library and their being no Open Standard for Physics. I dont believe Nvidia will ultimately play any more or less friendly than Intel. Right now Intel does not have a GPU so its less critical on a timing front for AMD. But I have every reason to believe Intel will be just as aggressive with Physics in the long run. Especially when they have a GPU to accelerate it. And Intel is clearly taking GPU Compute seriously.

 

[no-X]

Many successful games used Havok and many of these games ran very well on ATi hardware (HL2 series, Painkiller series...). ATi also cooperated with Havok on HavokFX. I'd expect much better relationship between ATi and Havok team, than between ATi and nVidia.

Havok is well accepted - it's SW based solution today, so it runs well on every PC. I think it's better to support well estabilished and widely accepted format.

From the point of hardware acceleration, nVidia is the real competitor, not Intel. How can ATi be sure, that there won't be PhysX related ID checks in nVidia sponsored titles to make the games running slower or uglier on ATi hardware? This is not a risk in the case of Havok. I wouldn't expect any complication from Intel, they don't care about gaming today and I think they are pretty happy, that AMD and ATi are promoting Havok in a much stronger way, then they do.

I have no idea how the final version of the OpenCL based Havok port will look like, but I can imagine it would be quite cool, if the solution would be backward compatible to all (or majority of) Havok titles.

 

[digitalwanderer]

That is not how it went down.

http://www.extremetech.com/article2/0,2845,2324555,00.asp

My bad, my memory sucks. Sorry.

But your link points out why AMD didn't go with PhysX Chris, they didn't want to trust nVidia on a GPU based physics solution. Intel's Havoc is more CPU based, ain't it?

And Digi. Such is the price you pay of not having your own Physic library and their being no Open Standard for Physics. I dont believe Nvidia will ultimately play any more or less friendly than Intel. Right now Intel does not have a GPU so its less critical on a timing front for AMD. But I have every reason to believe Intel will be just as aggressive with Physics in the long run. Especially when they have a GPU to accelerate it. And Intel is clearly taking GPU Compute seriously.

I see Intel taking CPU computing a bit more seriously than GPU computing, at least for the foreseeable future.

And I do disagree on your assessment, I believe nVidia will ultimately play less friendly than Intel.

 

[ChrisRay]

Many successful games used Havok and many of these games ran very well on ATi hardware (HL2 series, Painkiller series...). ATi also cooperated with Havok on HavokFX. I'd expect much better relationship between ATi and Havok team, than between ATi and nVidia.

To be fair. Nvidia was going to work with HavokFX too. And HL2 ect were all released before Intel bought Havok.

How can ATi be sure, that there won't be PhysX related ID checks in nVidia sponsored titles to make the games running slower or uglier on ATi hardware?

PhysX related ID checks. Thats another good question/feedback. I'll look into it. S. I would like to point out. That currently AMD users are not allowed to use PhysX at all in any tangible means. And there are more and more titles coming out now that don't support it. So they dont get any benefit at all from it.

So far the list of things I have gotten from you guys that I think are definately worth looking into are this. I cant gaurentee I'll get an answer on any of them. But I plan to press the issue tonight.

1) OpenCL. Would AMD be allowed to port it OpenCL? Or will Nvidia create OpenCL variant of PhysX.
2) ID Checks. Assuming AMD were to take on PhysX. Would Nvidia actively seek to hurt AMD's Physics Adoption Under OpenCL.
3)PhysX Performance under CPU/Software controlled enviroment is less than stellar. Will We see further optimisations to this. Or will the front stay strictly GPU and CPU as an after thought?

Other issues are purely opinion based to me. And my disconnect with you guys is on Havok. Havok is not the same company that was first born since Intel took it over. Anymore than Ageia is the same company since Nvidia bought it out. Havok = Intel. PhysX = Nvidia. And the way I see it Intel/Nvidia are out to make money. And I dont see any good reason for them to give away their technology for free to AMD just because they "Didnt have it". As both Intel and Nvidia put quite a heavy investment into both of these companies. I doubt I'll be easily convinced either way. However, Thank you for the time being giving me a chance to listen into your perspectives. It has been enlightening one way or another.

Chris