nVidia, ATi, PhysX & Upcoming Games

[Tamlin_WSGF]

I'm assuming you don't mean open source but based on a standard API. Yes one day someone will write an OpenCL physics library. But that day isn't here yet and Nvidia is making hay while the sun shines. A hypothetical future has no bearing on whether PhysX makes business sense today.

That day is here, isn't it?:

AMD Open Physics Initiative Expands Ecosystem with Free DMM for Game Production and Updated version of Bullet Physics
Updated Pixelux and Bullet Physics offerings enable more realistic games that run on any OpenCL™- and DirectCompute-capable platforms

http://www.amd.com/us/press-releases/Pages/amd-ecosystem-2010mar8.aspx

and its free.

When it comes to GPU accelerated physics so far, its pretty worthless for any mainstream user unless they have a dedicated physx card in addition to their mainstream card. Unless they want to turn down all eyecandy to get physx and low framerates. The performance hit is too big with physx in comparison to the gain. Batman with CPU PhysX hack was better to use then the GPU accelerated PhysX for mainstream user (people often have a decent CPU coupled with their GPU).

 

[noquarter]

I think the biggest problem with GPU accelerated PhysX is that Nvidia bought it. If Ageia (or Novodex) still owned it, both those companies were huge on licensing - I have no doubt they would have developed an OpenCL/DirectCompute based PhysX platform and licensed it to both Nvidia and ATI (or licensed directly to devs). With a neutral middleman like that there would have been no risk for both ATI and nVidia to adopt the tech similar to Microsoft providing DirectX as a platform.

Because Nvidia purchased it however, it makes no sense whatsoever for ATI to license the tech from Nvidia (btw Nvidia only offered CUDA for free - PhysX still required a license fee, which would screw ATI once PhysX became established). Another way to make money off PhysX is to open it up to other hardware to increase market penetration then charge devs a licensing fee.


Also, it's ridiculous to compare ATI pushing hardware agnostic DX11 with Nvidia pushing vendor locked PhysX. They don't even mean the same thing.

 

[trinibwoy]

That day is here, isn't it?:

Sure if you want to equate powerpoint slides with an actual product. AMD talks a lot but talk is cheap. They are piggybacking on stuff like Bullet (which ironically was and is still primarily CUDA based, OpenCL support is still in the works) cause right now they have no concrete answer to PhysX. Same thing with OpenCL - they tried to convince everyone that they're championing open compute standards yet at the end of the day Nvidia often ends up with better OpenCL driver support and performance (see Anand's review or visit B3D). Can't blame them for doing so as it's their only angle right now until they have something solid to work with.

In any case it's not a pissing match between CUDA and OpenCL. Developers will use the package with the best combination of cost, compatibility, support and toolchain to suit their needs. Sometimes it'll be PhysX, sometimes it won't. When they choose PhysX Nvidia will of course try to cajole them into tacking on GPU based effects too to help push their hardware which is the goal at the end of the day.

I agree on the performance hit but it's not something unique to PhysX. Consider DoF and Tessellation in Metro for example. Besides with Fermi IIRC stuff like Batman is now quite playable on a single card with high PhysX enabled.

 

[Tamlin_WSGF]

Sure if you want to equate powerpoint slides with an actual product. AMD talks a lot but talk is cheap. They are piggybacking on stuff like Bullet (which ironically was and is still primarily CUDA based, OpenCL support is still in the works) cause right now they have no concrete answer to PhysX. Same thing with OpenCL - they tried to convince everyone that they're championing open compute standards yet at the end of the day Nvidia often ends up with better OpenCL driver support and performance (see Anand's review or visit B3D). Can't blame them for doing so as it's their only angle right now until they have something solid to work with.

How can you piggyback on middleware? The GPU physics tools are given away to selected developers, so its more then powerpoint slides

AMD announced that it’s giving away a free version of Pixelux’s DMM2 material physics engine to selected game developers.

http://bulletphysics.org/wordpress/?p=175

In any case it's not a pissing match between CUDA and OpenCL. Developers will use the package with the best combination of cost, compatibility, support and toolchain to suit their needs. Sometimes it'll be PhysX, sometimes it won't. When they choose PhysX Nvidia will of course try to cajole them into tacking on GPU based effects too to help push their hardware which is the goal at the end of the day.

(game)Developers will use what works on most machines, so they don't need to worry about many different implementations and API's. Developers develop on standards, not closed proprietary crap that makes PC into consoles in the end. Thats why we see GPU accelerated PhysX only in TWIMTBP titles where Nvidia had a heavy participation in development of the titles. TWIMTBP titles with physx where Nvidia didn't participate that much don't have GPU accelerated physx.

I agree on the performance hit but it's not something unique to PhysX. Consider DoF and Tessellation in Metro for example. Besides with Fermi IIRC stuff like Batman is now quite playable on a single card with high PhysX enabled.

[H] turned off DoF, due to a massive performance hit vs. gain. With tessellation, the performance hit wasn't that big, so they left it on. Same goes with PhysX. It drains performance extremely and gives very little back imo. I'd prefer that its offloaded to CPU and given more optimizing for multi-gpu instead.

PhysX smoke takes more hits then dynamic volumetric smoke it seems. Check out on 2.10 here:
http://www.youtube.com/watch?v=Qxg9rK086SU

Some got impressed by the physx effects given in Batman, since they were displayed as:
PhysX on: Smoke with physx controlled behavior
PhysX off: No smoke

I was not impressed.

 

[trinibwoy]

How can you piggyback on middleware? The GPU physics tools are given away to selected developers, so its more then powerpoint slides

Again, no product. You piggyback on middleware by sending out press releases about other people's work (like the one you just linked to).

Developers will use what works on most machines

PhysX works on every machine as well. If you're talking solely about GPU physics there is no competition. Bullet is a paper tiger - like I said above, its OpenCL support is still in the works. You also keep ignoring the things that are far more important than the API - i.e. the tools that help developers integrate and use the physics functions.

[H] turned off DoF, due to a massive performance hit vs. gain. With tessellation, the performance hit wasn't that big, so they left it on. Same goes with PhysX.

http://www.computerbase.de/artikel/...x_470/6/#abschnitt_batman_arkham_asylum_physx
http://www.fudzilla.com/content/view/18242/1/1/7/

Not quite the same at all.

 

[Tamlin_WSGF]

Again, no product. You piggyback on middleware by sending out press releases about other people's work (like the one you just linked to).

The whole point of middleware is that it works on several platforms. If sending out press-releases about work that you have had part in "piggybacking", then the whole TWIMTBP a big piggyback ride. I can't see how piggybacking has any value to the discussion then.

PhysX works on every machine as well. If you're talking solely about GPU physics there is no competition. Bullet is a paper tiger - like I said above, its OpenCL support is still in the works. You also keep ignoring the things that are far more important than the API - i.e. the tools that help developers integrate and use the physics functions.

Bullet physics is already used in titles, so I would hardly call it a paper tiger. Public opencl support is coming in v3, but that doesn't mean the version given to developers as mentioned above press release doesn't support opencl. Press release says opencl and directcompute support.

You say I ignore the tools, but I don't see the point in discussing the toolkits when there is no review of the toolkits provided to developers as in the above presskit. What I find important, is that developers have tended to use API's with the broadest userbase. Thats why many thinks there wasn't so many DX10 games. Consoles and DX9 machines were the largest userbase.

http://www.computerbase.de/artikel/...x_470/6/#abschnitt_batman_arkham_asylum_physx
http://www.fudzilla.com/content/view/18242/1/1/7/

Not quite the same at all.

Lets look at those:
http://www.fudzilla.com/content/view/18242/1/1/7/
GTX285:
1920x1200 4XAA
without physx = 101FPS
with physx = 47FPS

Major performance hit

http://www.computerbase.de/artikel/...x_470/6/#abschnitt_batman_arkham_asylum_physx

ATI cards with PhysX on CPU = 21fps regardless of resolution = Crappy support for CPU in PhysX or capped. As I remember from Tomshardware's review of Batman, CPU utilization actually went DOWN when using GPU accelated physx on CPU instead of UP which is natural.

 

[noquarter]

The reason middleware isn't working out for ATI is because Nvidia keeps buying out the middleware to vendor lock it. I really think nVidia 3d Vision is just a licensed pair of Samsung shutter glasses with a $50 markup too. I believe you can do the same with ATI + Samsung glasses + iZ3d or DDD middleware.

I still think it's better to let the middleware fight it out for the best solution. If Ageia/Novodex was trying to sell PhysX as a development platform they'd make sure it ran fast on CPU's and GPU's (provided their PPU market crashed), but since Nvidia is trying to sell PhysX as a bullet point it's only going to work well on nVidia cards.


Also I'm sure we all know the difference between PhysX and GPU accelerated PhysX.. just easier to shorthand it since we are almost always talking about GPGPU PhysX processing..

 

[Tamlin_WSGF]

The reason middleware isn't working out for ATI is because Nvidia keeps buying out the middleware to vendor lock it. I really think nVidia 3d Vision is just a licensed pair of Samsung shutter glasses with a $50 markup too. I believe you can do the same with ATI + Samsung glasses + iZ3d or DDD middleware.

I still think it's better to let the middleware fight it out for the best solution. If Ageia/Novodex was trying to sell PhysX as a development platform they'd make sure it ran fast on CPU's and GPU's (provided their PPU market crashed), but since Nvidia is trying to sell PhysX as a bullet point it's only going to work well on nVidia cards.


Also I'm sure we all know the difference between PhysX and GPU accelerated PhysX.. just easier to shorthand it since we are almost always talking about GPGPU PhysX processing..

Exactly.

 

[coldstorm]

The reason middleware isn't working out for ATI is because Nvidia keeps buying out the middleware to vendor lock it. I really think nVidia 3d Vision is just a licensed pair of Samsung shutter glasses with a $50 markup too. I believe you can do the same with ATI + Samsung glasses + iZ3d or DDD middleware.

I still think it's better to let the middleware fight it out for the best solution. If Ageia/Novodex was trying to sell PhysX as a development platform they'd make sure it ran fast on CPU's and GPU's (provided their PPU market crashed), but since Nvidia is trying to sell PhysX as a bullet point it's only going to work well on nVidia cards.


Also I'm sure we all know the difference between PhysX and GPU accelerated PhysX.. just easier to shorthand it since we are almost always talking about GPGPU PhysX processing..

why yes you can with the 3d glasses. but until they move physx to open standard ati will not support. we have to wait for bullet or havok in opencl to level to playing field as far a physics in games. physx is most games don't really improve the experience that much.

 

[trinibwoy]

The whole point of middleware is that it works on several platforms.

Middleware is simply a package that provides standard functionality so that a developer doesn't have to re-invent the wheel. The definition of middleware isn't tied to whether you're working on a proprietary or open solution. There is lots of middleware out there that's tied to a specific platform. Cryengine 2.0 for example.

Public opencl support is coming in v3, but that doesn't mean the version given to developers as mentioned above press release doesn't support opencl. Press release says opencl and directcompute support.

Just more promises.

Major performance hit

Why are you conveniently excluding GF100 based cards? Is it because they achieve over 60fps with AA and high PhysX enabled which would completely invalidate your claims?

ATI cards with PhysX on CPU = 21fps regardless of resolution = Crappy support for CPU in PhysX or capped.

Yes, and how is crappy CPU support relevant to the consumer making a purchasing decision between two graphics cards? That's what the OP is about after all.

 

[Tamlin_WSGF]

Middleware is simply a package that provides standard functionality so that a developer doesn't have to re-invent the wheel. The definition of middleware isn't tied to whether you're working on a proprietary or open solution. There is lots of middleware out there that's tied to a specific platform. Cryengine 2.0 for example.

Middleware ties software together. GPU accelerated function of PhysX is part of driver functionality and can hardly be called middleware.

Just more promises.

Whats next is always a promise. However, they have delivered this solution to selected developers as per press release, so promises have been delivered.

Why are you conveniently excluding GF100 based cards? Is it because they achieve over 60fps with AA and high PhysX enabled which would completely invalidate your claims?

No, my claims are the massive performance hit vs. gain. I didn't think it was necessary to use the GTX480 numbers, but if you wish:

Batman 1920X1200 4XAA, no physx = 143FPS
Batman 1920X1200 4XAA, physx = 69FPS

Major performance hit!

Yes, and how is crappy CPU support relevant to the consumer making a purchasing decision between two graphics cards? That's what the OP is about after all.

Crappy CPU support is relevant to the OPs question about the future of physx.

As for purchasing decision, the performance hit makes GPU accelerated PhysX useless for most users. Especially seeing how it performs in a console port like Batman.

 

[Gorankar]

PhysX is no better off now than it was under Ageia. It's free so a lot of indie or low rent devs use the CPU version of PhysX, but almost none of them add GPU acceleration to their games using PhysX. A couple of gpu accelerated AAA titles per year with varying levels of eye candy or performance enhancement ranging from pretty cool to meh, per year. How many years have gone by now since it came out? If it has not been accepted as "the standard" yet, why do people still think it will be? Stuff not controlled by Nv is on the way, and Intel and AMD are embracing nearly everything that comes down the pike that even remotely competes against PhysX or CUDA.

It is the same now as then. If you really like the added eye candy in that tiny little minority of games that use GPU accelerated PhysX, you use an Nv card. That's all there is to it.
Though to be honest, I could never see actually buying a Nv card solely for PhysX. Use my 9800gtx+ that I already had, sure, but buy a new card with just PhysX in mind, I just don't see it. Not for one or two games a year. GPU accelerated PhysX has been all but shunned by AAA developers. And the few times it does show up in AAA games, it seems like Nv actually had to code it in themselves.

 

[trinibwoy]

Middleware ties software together. GPU accelerated function of PhysX is part of driver functionality and can hardly be called middleware.

Guess we'll have to stick with our own definitions based on our own experience with the topic.

Whats next is always a promise.

Nothing wrong with that but don't try to palm off promises as viable alternatives to a shipping product.

No, my claims are the massive performance hit vs. gain.

Yes, I know but it's completely irrelevant when the resulting performance is still eminently playable. "Performance hit" is only relevant when it takes you into unplayable territory. Lots of features have large performance hits. Try lowering resolution or IQ settings and you'll see how fast you get into the hundreds of fps.

As for purchasing decision, the performance hit makes GPU accelerated PhysX useless for most users. Especially seeing how it performs in a console port like Batman.

Considering the OP was about a Fermi purchasing decision I'm not sure how "most users" matter. The question was whether someone considering a Fermi based card would include PhysX support in their list of criteria.

 

[Tamlin_WSGF]

Guess we'll have to stick with our own definitions based on our own experience with the topic.

Let do that.

Nothing wrong with that but don't try to palm off promises as viable alternatives to a shipping product.

If a promise has been shipped to developers in form of physics tools, I consider it a viable alternative for them. If Bullet announces the next version as opencl and directcompute, I consider it worthy to include. I mean, you have been discussing fermi from the first time it was advertised as upcoming. Why do you wish to exclude next version of Bullet in a discussion about the future of physics? Even OP discusses about what to come.

Yes, I know but it's completely irrelevant when the resulting performance is still eminently playable. "Performance hit" is only relevant when it takes you into unplayable territory. Lots of features have large performance hits. Try lowering resolution or IQ settings and you'll see how fast you get into the hundreds of fps.

Its very relevant. PhysX within "playable territory" with Batman was possible from the beginning by using a dedicated physX card. Its still possible regardless of which card you have as primary to use a secondary Nvidia card for PhysX. The performance hit causes most people to turn it off, not to add a second card.

Considering the OP was about a Fermi purchasing decision I'm not sure how "most users" matter. The question was whether someone considering a Fermi based card would include PhysX support in their list of criteria.

OP wasn't asking if people want a fermi card or not, but if they would consider Nvidia for physx.

OP was questioning following:

1. So, how relevant will PhysX continue to be in the future? How many other developers are utilizing PhysX for their upcoming titles?

A: For GPU accelerated physx, I think only in a few titles where Nvidia has been heavy into development of. Other titles will have only CPU PhysX (or Havok/Bullet) as we've seen so far IMO.

2. If you are with ATi at the moment, would you consider moving to nVidia to get a better gaming experience with games you're interested in that utilize PhysX?

A I don't consider PhysX more worthy to use then before when it was under Aegia. If I should ever consider PhysX, I would use a dedicated card for it regardless of my primary card.

3 Just curious as to the opinions of those around here, regarding the situation and how, if at all, it affects them and their decisions regarding what hardware to go with.

A. Most people are mainstream users, so I believe that the 50%+ performance hit makes it useless for them.

 

[Deviationer]

Microsoft should just add their own physics+GP-GPU processing tech to DX11/12. Also the OpenGL group should do the same thing. Their both standards so why not? Why let nvidia control it all?

 

[Gorankar]

Microsoft should just add their own physics+GP-GPU processing tech to DX11/12. Also the OpenGL group should do the same thing. Their both standards so why not? Why let nvidia control it all?

That is the thing. GPU physics has not taken off at all. A ton of games use the PhysX api for CPU physics, but only around 20 or so released games use GPU or PPU acceleration for PhysX. Of those, maybe six or seven are AAA titles, nine at most if you are really generous. That is it since PhysX came out in 2004 and Nv took it over in early 2008. With such low adoption, it is prolly a non issue to MS at this point. It might as well be Trueform.

 

[Balthazor]

Proprietary APIs almost always end up dying out if they are tightly restricted to a particular vendor.

Remember GLide?

EAX is still around... but hardly a major factor now. Creative had a good run with EAX in the late 90s / early 2000s, but it has been declining in use steadily, even though it is able to run on top of OpenAL, which is supported under Vista/Windows 7. And this despite the fact that Creative actually licensed the X-Fi hardware to another vendor; too late, in my opinion. Granted, Creative hasn't been throwing the kind of money at game developers that Nvidia has been, but I think that particular bit of gaming hardware history is relevant.

GPU PhysX, if Nvidia continues to artificially vendor-restrict its use through driver blocks, will eventually go away also. How quickly? I'm not sure.

The question is, as I'm rephrasing the OP, "Is GPU PhysX going to be relevant enough in the near future to warrant making it a major purchasing factor?" The current release rate of good GPU PhysX games seems to point to no - it shouldn't be a major factor. It is a factor, clearly, but not a major one. Certainly not enough for me swap out all my ATI hardware.

Nvidia's current business objective for GPU PhysX appears to be total domination of the GPU market, by only allowing GPU PhysX if your system is all-Nvidia. This 'total domination' scenario is almost certainly not going to happen in the foreseeable future.

I think a more realistic business objective would be to maximize GPU PhysX-influenced hardware sales of GPU cards to consumers, regardles of any other hardware in their system. I think this approach would make Nvidia more money.

Its not like Nvidia has to produce a PPU - any of their GPUs could be purchased by an ATI user for use with PhysX. But there are apparently more than a few who prefer the ATI graphics rendering experience, or the ATI value, or the ATI power and thermal efficiency, rather than go all-Nvidia just for PhysX.

 

[sbuckler]

Proprietary APIs almost always end up dying out if they are tightly restricted to a particular vendor.

So what about proprietary DX vs open OpenGL? How come the world mostly uses DX despite it only working on some versions of windows and the xbox, and despite OpenGL being around longer then DX and heavily used long before glide and DX were even being dreamed up on a bunch of power point slides.

For that matter why are we still using proprietary windows at all when there are perfectly good open (and free) linux os's?

Saying open always wins is rubbish. The system with the best support and strongest marketing wins. For gpgpu right now that is cuda, and chances are in the future it will probably be direct x compute. OpenCL is likely to end up like OpenGL - a second choice to the MS backed standards mainly used where MS is not allowed (e.g. PS4).

As for gpu physx vs bullet - neither will take off on the gpu until the xbox 720 and PS4 which support hardware physics arrive. Until then all you'll see is software physics (cpu phsyx/havok & bullet), the only gpu physics you'll see is if someone promotes it. That basically means physx as nvidia is the only one that really puts effort into this sort of thing.

 

[Tamlin_WSGF]

So what about proprietary DX vs open OpenGL? How come the world mostly uses DX despite it only working on some versions of windows and the xbox, and despite OpenGL being around longer then DX and heavily used long before glide and DX were even being dreamed up on a bunch of power point slides.

OpenGL's features are determined between hardware vendors in Kronos, while DirectX is determined by hardware vendors in cooperation with Microsoft. The whole purpose of DirectX and OpenGL is to have a common platform. There are no royalties involved in either OpenGL or DirectX. They are both open standards, though only OpenGL is open source.

Those two API's are widely supported by vendors and therefore also widely supported by developers.

GPU accelerated PhysX is only used in a few titles, despite heavily marketed by Nvidia and even in those few titles Nvidia had to go heavy into development of these. Doesn't seem that developers themselves choose PhysX even though its available and they already use CPU PhysX in their games. With the 50%+ framerate loss from PhysX, I doubt it will be a vital part of any game in the future. Even in console ports like Batman, PhysX is useless for mainstream users.

 

[Balthazor]

So what about proprietary DX vs open OpenGL? How come the world mostly uses DX despite it only working on some versions of windows and the xbox, and despite OpenGL being around longer then DX and heavily used long before glide and DX were even being dreamed up on a bunch of power point slides.

For that matter why are we still using proprietary windows at all when there are perfectly good open (and free) linux os's?

Saying open always wins is rubbish. The system with the best support and strongest marketing wins. For gpgpu right now that is cuda, and chances are in the future it will probably be direct x compute. OpenCL is likely to end up like OpenGL - a second choice to the MS backed standards mainly used where MS is not allowed (e.g. PS4).

I never said that the open standard always wins, just that the gaming market tends to end up rejecting vendor-restricted APIs. None of the examples you gave are restricted to a particular vendor, whereas GPU PhysX is restricted to not just running on an Nvidia GPU, but further restricted to only those systems that use exclusively Nvidia GPUs.

 

[TehQuick]

Lols, sbuckler is sooooo convinced that PhysX is here to stay, when everything points toward it's eventual demise.

 

[trinibwoy]

OP was questioning following

Yes, and you're projecting into the future based on some press releases. There's no actual substance there to make a comparison.

They are both open standards, though only OpenGL is open source.

None of the examples you gave are restricted to a particular vendor

DirectX is not an open standard. It is tied to a single vendor - Microsoft. The only reason people don't complain is because that vendor has virtually 100% of the PC gaming market. One company having a monopoly doesn't make something a standard.

Its not like Nvidia has to produce a PPU - any of their GPUs could be purchased by an ATI user for use with PhysX. But there are apparently more than a few who prefer the ATI graphics rendering experience, or the ATI value, or the ATI power and thermal efficiency, rather than go all-Nvidia just for PhysX.

Nvidia could very well peddle their low end cards as PhysX accelerators if they wished. I don't see the business sense in that though since they completely eliminate any competitive edge their higher margin products have over the competition with respect to PhysX. Not to mention it would make it much easier for Nvidia owners to choose an ATi card when upgrading. Honestly, you can't take a consumerist view when trying to understand business strategy. It'll never work.

 

[Balthazor]

DirectX is not an open standard. It is tied to a single vendor - Microsoft.

I'm sorry I wasn't clear. I wasn't speaking of the OS that can run the software, but the hardware that runs the software. There are no restrictions on what hardware can run DirectX, other than the obvious restriction that the hardware must also run on Windows.

I thought the contrast with GPU PhysX and Nvidia was evident, but I hope that explanation makes it more so.

I completely agree that if Nvidia had a proportional market share of the gaming GPU market that Microsoft has of the OS market, there would be substantially fewer people complaining about Nvidia-exclusive GPU PhysX.

Nvidia could very well peddle their low end cards as PhysX accelerators if they wished. I don't see the business sense in that though since they completely eliminate any competitive edge their higher margin products have over the competition with respect to PhysX. Not to mention it would make it much easier for Nvidia owners to choose an ATi card when upgrading. Honestly, you can't take a consumerist view when trying to understand business strategy. It'll never work.

I don't have a problem understanding Nvidia's business strategy, I just don't think it will work with regards to GPU PhysX for the reasons I presented. As a consumer I am biased, and of course I want healthy competition to help motivate the manufacture of robust product offerings, but also as an Nvidia shareholder I'm concerned about their recent performance and the direction they took GPU PhysX.

I disagree that allowing GPU PhysX to work in a mult-vendor GPU system would "eliminate any competitive edge... with respect to PhysX." Nvidia cards would still be the only cards that could run PhysX. It would still force gamers who want PhysX to purchase Nvidia. And I think it a lot more likely for an ATI user to purchase an Nvidia card for PhysX than for them to entirely swap out their GPU hardware for PhysX alone.

 

[astrallite]

PhysX is an open standard. DirectX is a closed standard (with 90% of computers running it). Interesting. ATI is evil with its proprietary actions, apparently. You learn something everyday.

 

[trinibwoy]

I thought the contrast with GPU PhysX and Nvidia was evident, but I hope that explanation makes it more so.

Yep, your point was clear. Was just highlighting that the pushback on PhysX has more to do with the Nvidia vs ATi divide than it has to do with it being proprietary. If Nvidia was the only purveyor of discrete graphics cards it wouldn't be an issue.

And I think it a lot more likely for an ATI user to purchase an Nvidia card for PhysX than for them to entirely swap out their GPU hardware for PhysX alone.

Well that's the crux of our disagreement there. I'm not convinced at all that many people would opt for a second card for PhysX. The number of SLI and Crossfire systems are miniscule and those setups are arguably a lot more mainstream and applicable to a lot more titles than PhysX ever will be.

 

[Gorankar]

PhysX is an open standard. DirectX is a closed standard (with 90% of computers running it). Interesting. ATI is evil with its proprietary actions, apparently. You learn something everyday.

They are both proprietary standards, and both are free for software developers to use.

However, while DX is free for all hardware vendors to use, GPU accelerated PhysX and CUDA are locked to Nv hardware.

 

[Slachtbeest]

Well, seems I've started something that became quite a "hot topic" here, since last I checked, heh. I wasn't expecting such excellent, in-depth feedback, so glad to see it, as many responses have given some great input, info and just good general thoughts on the subject.

I agree with most of what's been said here, including bringing CPUs more "into light" regarding the subject, for while it's mainly about the GPU-end of things, it does make sense to take into consideration CPU performance, for games that utilized the CPU for PhysX, in a purchasing decision, if things ever came down to that for anyone.

If developers offloaded the work to the CPU in order to get the same exact level of effects and performance, that would obviously be preferable, considering that, as someone else had said, it does a lot to keep GPU manufacturers from alienating consumers.

If developers stick with the GPUs for utilizing PhysX for the most part, then that's where I, and many others, obviously see the problem.

I suppose that, if more developers began relying heavily on PhysX as the "backbone" for effects for any given game that would not be rendered properly by ATi's GPUs, then that would make it more relevant.

So, I suppose that, for the moment, it's more of a "case-by-case" basis kind of thing, though that can be a touch decision when it's a game/games that someone might particularly be very much into playing.

If PhysX was used say, strictly for something "small" such as the smoke effects in Batman, I personally would not be bothered by the loss of that, unless the game was one that utilized a lot of smoke, fog etc. with which PhysX had a large part to play in how it was rendered and if it added to the immersion of any given game.

But to me, that's the issue: we're all heavy gamers, and the experience is what it's all about. The more immersive, the better the experience. So, if a developer utilizes PhysX very heavily to create the effects they have that make up a very large part of the enjoyment/experience of a game, such as what seems to be the case with Mafia 2, as example, then that could indeed be a bit "turn-off" for those not able to see the game the way it was meant to be seen.

With Mafia 2, again just as an example, they've utilized PhysX very heavily to create the extremely multi-layered destruction they've created in the game. As someone else has said, it remains to be seen until the game is released, whether or not they shot themselves in the foot by relying on it "too heavily" to the point that, anyone running AMD's GPUs, would not be able to experience the game as in-depth due to the devs having done this.

Again, as someone else has said, any smart developer would find a way to make it work on another brand of GPU, as to not alienate any gamers. But, if they haven't found/taken a time to create the same experience for the users of another brand of GPU, depending on just how heavily PhysX has been used to create such an immersive experience, anyone not running nVidia could suffer the loss of having a lesser experience with any particular game, which defeats the whole purpose of gaming.

With something such as Mass Effect 2, there's nothing even noticeable (that I can see) created by the use of PhysX, so it doesn't much matter there, except for the fact that enabling AA via the CCC causes some serious lag, and if not forcing AA, I've found that it looks better with nVidia than ATi for some odd reason, in terms of aliasing.

I think the issue becomes more relevant on an individual basis, meaning that many gamers are only into one or two specific genres/types of games. If those particular games happen to use PhysX very heavily to create effects that would otherwise be "muted" or "dumbed-down" due to running an AMD GPU, then it becomes a little more relevant for that particular person in regards to a decision as to which GPU manufacturer to go with, as they will want the best possible experience with the game(s) they enjoy.

So, I feel that there are different "levels" of relevance. There's the "overall" relevance in the general gaming industry, which doesn't seem all that large, if not even shrinking. But then there's the "personal" relevance, for those who might be heavily into some game(s) that utilize PhysX via GPU, which would force them to go with nVidia if they wanted the "full experience" and immersion the game has to offer.

I don't personally feel PhysX generally offers a more in-depth experience, as someone suggested I might have meant. I meant what others have understood, that when it's used well and very heavily in a way that offers a more in-depth experience for any particular game, that it then could become an issue for certain people, or even just gamers in general, who are not running nVidia.

I run both nVdia and ATi, and I've no "preference" for any particular technology used to create an in-depth experience, as long as any given developer manages to create an in-depth experience with what ever they've chosen to use.

Visuals can indeed make-or-break the immersion in gaming, and in general, heavy gamers want to see more and more impressive, interesting and just general fun things on-screen.

So, using the example given above, I could use myself and Mafia 2 as an example of everything I've said here...

I'm greatly looking forward to the game, and it's something I want to be able to experience and enjoy to it's fullest. So, if the devs relied so heavily on PhysX to help create an extremely in-depth experience that would vary greatly using an AMD GPU, that then gets me thinking about how much I'd be "missing", and how much less of an immersive experience it might be, which then leads to the "GPU vs GPU" issue.

Of course, as seems to be the main point with most in this thread, none of this would even be an issue, were certain technologies not so strictly proprietary and/or developers didn't utilize something so heavily in only one particular way, and "leave out" the rest. Such as by not utilizing the CPU to render the same effects as a GPU-reliant PhysX game.

While the issue of PhysX is not one that effects nearly a large percentage of games out there, it can become an "issue" for those who want to get the most out of any given game that has utilized it heavily, but perhaps are not running nVidia at that time, which could severely lessen the experience for them... and that's just garbage.

I could easily just switch systems, since I have multiple systems and GPUs of both camps, but it's just not right that any given technology would be "brand-limited" in any way, to any degree, IMO.

Also, as someone else said, developers don't make games for any specific brand of GPU, certainly. But, if there's a certain technology the prefer to go with that hinders the experience of the gamer who is running another brand of GPU, and they're not smart enough to know better than to alienate that crowd, then, well... to put it simply: that just sucks.

We'll have to see what happens with Mafia 2, or any other upcoming title that might utilize PhysX to what ever degree, but the bottom line is, for me: "equal rights for all brands", so-to-speak.

Whether that means technologies being openly "shared" without the issue of "competition" getting in the way, or developers just being smart in how they can create something that gives the experience to the users of any hardware, I just see no reason that anyone should have to have a "lesser experience" just because of the GPU they happen to be running.

Certainly, any company has the right (and would be smart to) create a proprietary technology that offers something excellent. However, if it ever comes at the expense of the the gamer and the whole point of what a GPU is meant to do in the first place... play games in a manner that makes them fully enjoyable for all... then I just see no point to all the "proprietary" nonsense in this particular industry, as all it does it create frustration for the very individuals they're catering to in a very specific medium of entertainment.

 

[kllrnohj]

PhysX is an open standard.

Where the fuck did you get that ridiculous idea?

First, PhysX isn't a standard. Second, PhysX isn't open in any way, shape, or form. It is completely closed and proprietary, with Nvidia controlling what the library can do and what it can run on.

But let me say this again: PhysX is NOT a standard.

 

[noquarter]

^ I think he was being sarcastic in that post btw

That was a pretty long but interesting reply OP, and a clearly reasonable assessment. I did want to point out games like Mass Effect 2 use software PhysX which requires PhysX to be installed but runs entirely on the CPU so you don't miss any effects by using ATI.

I also wanted to add one more thought I have about GPGPU PhysX. To me the effects we've seen added have been way overcalculated. It shouldn't take an entire GPU worth of transistors to simulate some swirling leaves.

Nvidia is ignoring the key of game engines is to emulate real world effects by using shortcuts just like rasterizing textures is a huge shortcut over ray-tracing and fast inverse square root is a shortcut in 3d calcs. As a gamer you would never notice the difference between a 'best guess' swirl of leaves and a 100% accurate physics modeled swirl of leaves.

The reason they are doing this is obvious though, if you can overcalculate a physics effect just for the sake of calculating, you can try to justify using 750 million transistors to calculate this effect and claim there's no other way to do a swirl of leaves. A perfectly passable effect can be done on CPU by just reducing the complexity of the calculations to a reasonable level and most would never notice the difference, or use the GPU to raise the complexity on all calcs to a moderate degree (sounds like much more work) instead of overcalculating the effects on a single element just to exercise the GPU.

GPGPU physics is a great technology but not the way it's being used now and I believe it will greatly lose potential as high core count CPUs overtake the market.