Issues with CUDA

[Nenu]

The PPU wouldn't gain any preformance if you went PCI-E with it.
It's not a question about bandwith, we are not talking graphics here...it's a matter of massive parallel interactions...not triangles.

The PPU in current form is far from saturating the PCI bus...and why would I waste GPU power on anything but graphics?

But hey, perhaps you super-dooper-magic-bus suddenly will make GPU-physcis a reality...

You can bet, its on its way!

 

[Atech]

You can bet, its on its way!

Oh really?
Why don't you wake me up when you got more than PR-fluf, please?

 

[AnnoyedDragon]

Excuse me, but don't confuse Havok FX with Cuda. Cuda is a GPGPU framework and can do any kind of physics you want. It's aimed at a newer generation of hardware than Havok FX was.

Don’t presume that just because a technology can do something that it is going to. The PPU wasn’t used to its potential for game play physics because of install base and Internet connection limitations, based on the Steam hardware survey current CUDA install base is around 18%. That isn’t enough for game play physics, so right now CUDA physics is no better than Havok FX.

Well, I believe that Crysis won't allow you to change the physics level in an online game.
There's also the difference between DX9 and DX10-mode, where in DX10-mode there's more physics, like trees that can be destroyed and such.

DirectX 10 has nothing to do with the very high physics mode, nor is it a requirement for object destruction. The no breaking trees under DX9 online limitation in Crysis is pretty much artificial.

I wouldn't be too sure of that. Firstly, nVidia has a very large marketshare (especially in the game-buying market), and over time this entire marketshare will have GF8+ GPUs that are capable of Cuda-accelerated physics. It could be over 50% of all gamers (62% of all Steam-users have nVidia hardware, only 31% uses ATi...

You might want to check the Steam hardware survey again, also keep in mind it is the 8 series that can run CUDA not any Nvidia card. Frankly I don’t think people should be warming up to the idea of an Nvidia monopoly in the GPU market.

Secondly, it wouldn't be the first time that games included special features that were only available on one brand of hardware.

How many impacted game play? Which in turn made only owners of that hardware able to play the game?

But as I said before, currently nobody even knows what the ATi GPUs are capable of in terms of physics or GPGPU in the first place, because ATi never disclosed much of their architecture, and never released an SDK like Cuda.

ATI has their own GPU programming method, it is called CTM. It is pretty much ATI's version of CUDA and like CUDA only works on their brand of hardware. This of course poses the problem of ATI not wanting to abandon their own GPGPU method to pay royalties for Nvdia’s, not necessarily a fault on their part but of both companies for not working on a standard.

You don't *have* to buy two GPUs, it will work with a single GPU aswell...

From the Nvidia forum on CUDA.

Ask the people using CUDA just how well that works.

If a GPU is used for emulation, it becomes a physics engine. It cannot be used for both at the same time. At best, you must switch in and out of "modes" while operating.

Take the GPU into physics, and you drop the SLI capability as well.

That's why the more successful CUDA programmers go for either a stand-alone engine like Tesla or a second graphics card. GeForce owners testing the CUDA waters are finding this out the hard way.

In other words it is possible that graphics and CUDA cannot be share on the same GPU, I have yet to get a solid confirmation on this. Nvidia says it can be done, but then again they also said they were bringing CUDA to consoles…

Even if it does work on a single GPU; you have seen a few of the tech demos lying around, you are not going to get much after the graphics have take their share of the GPU.

 

[Nenu]

Oh really?
Why don't you wake me up when you got more than PR-fluf, please?

lol accept reality

 

[Scali2]

Don’t presume that just because a technology can do something that it is going to. The PPU wasn’t used to its potential for game play physics because of install base and Internet connection limitations, based on the Steam hardware survey current CUDA install base is around 18%. That isn’t enough for game play physics, so right now CUDA physics is no better than Havok FX.

You are confusing the possibilities of a technology with its applications. Havok FX was based on DX9-class hardware, and couldn't do anything other than 'special effects physics' by definition. Cuda has no such hardware limitations, neither does the PhysX API in general, or the PhysX PPU.
Havok FX was meant as an addon for the regular Havok API which would handle gameplay physics.

DirectX 10 has nothing to do with the very high physics mode, nor is it a requirement for object destruction. The no breaking trees under DX9 online limitation in Crysis is pretty much artificial.

That's not the point, the point was that the developers made sure that nobody had an advantage because their system ran at a different physics level.

You might want to check the Steam hardware survey again, also keep in mind it is the 8 series that can run CUDA not any Nvidia card. Frankly I don’t think people should be warming up to the idea of an Nvidia monopoly in the GPU market.

You might want to read my post again, and note the words 'over time'. It's only normal that everyone will upgrade their GPU from time to time. And as we can see from the Steam survey, by far the largest percentage of users buys nVidia. So over time the largest percentage of users will have Cuda-capable GPUs.

How many impacted game play? Which in turn made only owners of that hardware able to play the game?

There have been occasions where certain hardware changed, or at least improved gameplay. But never has it completely excluded other people, because there was always some form of software emulation. I don't see how that would be different. As pointed out, games today already have multiple levels of physics. Hardware acceleration won't change that concept, it will just make some levels hardware-accelerated.

ATI has their own GPU programming method, it is called CTM. It is pretty much ATI's version of CUDA and like CUDA only works on their brand of hardware. This of course poses the problem of ATI not wanting to abandon their own GPGPU method to pay royalties for Nvdia’s, not necessarily a fault on their part but of both companies for not working on a standard.

I know about CTM, hence my point that ATi has nothing that is anything like Cuda. Firstly, CTM is aimed at X1000-series hardware, and therefore far less powerful than Cuda.
Secondly, CTM is very lowlevel programming, pretty much assembly, where Cuda is more or less regular C++ programming. So Cuda is easier to develop, more powerful, and has a larger potential userbase.

In other words it is possible that graphics and CUDA cannot be share on the same GPU, I have yet to get a solid confirmation on this. Nvidia says it can be done, but then again they also said they were bringing CUDA to consoles…

I'm not sure who that quote is from, but the Cuda SDK actually includes some examples of Cuda combined with D3D or OpenGL, and it works pretty well. And that is with the standard programming interface. nVidia could easily implement some extra shortcuts and optimizations for physics in the drivers, making it work even better.
And where exactly did nVidia say they were bringing Cuda to consoles? Never heard anything like that, although I can imagine that they would say that if there's a next-gen console with GF8+ hardware planned.

 

[Atech]

In other words it is possible that graphics and CUDA cannot be share on the same GPU, I have yet to get a solid confirmation on this. Nvidia says it can be done, but then again they also said they were bringing CUDA to consoles…

Or GPU physics to 6 and 7 series GPU's...that didn't happen either

 

[Scali2]

Or GPU physics to 6 and 7 series GPU's...that didn't happen either

Yes, but they were banking on Havok FX, which never surfaced, as far as I know.
Now that Intel owns Havok, I doubt it ever will. You could at most blame nVidia for trusting Havok to deliver the software, rather than taking matters into their own hands... but they've corrected that mistake now, by buying Ageia.

 

[Atech]

lol accept reality

Okay..my PPU vs your 8800...what shall we play first?

Oh wait...that right...you can't do anything, sorry

 

[Nenu]

Okay..my PPU vs your 8800...what shall we play first?

Oh wait...that right...you can't do anything, sorry

lol, now i know why you are pissed, you bought a PhysX card

 

[AnnoyedDragon]

You are confusing the possibilities of a technology with its applications. Havok FX was based on DX9-class hardware, and couldn't do anything other than 'special effects physics' by definition. Cuda has no such hardware limitations, neither does the PhysX API in general, or the PhysX PPU.
Havok FX was meant as an addon for the regular Havok API which would handle gameplay physics.

Havok FX was not based on DirectX 9; it utilised GPU math computing via the shader pipelines, it worked on *any* GPU with shader model 3, even if that GPU was OpenGL only like with the PS3. Using this method however meant it couldn’t communicate with the CPU, so the physics could not have events attached to them to influence game play.

It is true CUDA does not have this same problem, but as long as it has the tiny install base and incompatibility with both ATI and consoles it will not be utilised in a way that makes it a mandatory requirement. Which means that like Havok FX it will only be used for effects; it can do affects, but they are not practical in its current state.

You cannot presume time will resolve this issue automatically, CUDA install base will rise but the incompatibility issues will still be present.

That's not the point, the point was that the developers made sure that nobody had an advantage because their system ran at a different physics level.

No, they made sure that you had to buy Vista before they ticked the box needed to let you play with destructible game play online. It has nothing to do with protecting gamers from those with better hardware, they could have just marked servers with destructible game play and let player decide if they can handle it. Making it DX10 only was simply a favour for Microsoft and is no indication if you CPU and Internet connection is up to task.

You might want to read my post again, and note the words 'over time'. It's only normal that everyone will upgrade their GPU from time to time. And as we can see from the Steam survey, by far the largest percentage of users buys nVidia. So over time the largest percentage of users will have Cuda-capable GPUs.

Again you are assuming time is a cure all and everyone will naturally adopt CUDA over time, not that simple.

There have been occasions where certain hardware changed, or at least improved gameplay. But never has it completely excluded other people, because there was always some form of software emulation. I don't see how that would be different. As pointed out, games today already have multiple levels of physics. Hardware acceleration won't change that concept, it will just make some levels hardware-accelerated.

The multiple levels are effects; game play physics will be the lowest setting. See this is were the concept of lowest common denominators come in, you target the weakest common spec in order to maximise potential audience. Everything above that is perks for those with better hardware, but they cannot impact game play in a way that makes it play different from the lowest common denominator.

It is how the industry works, target the weakest aspect of each platform you plan to launch on.

I know about CTM, hence my point that ATi has nothing that is anything like Cuda. Firstly, CTM is aimed at X1000-series hardware, and therefore far less powerful than Cuda.
Secondly, CTM is very lowlevel programming, pretty much assembly, where Cuda is more or less regular C++ programming. So Cuda is easier to develop, more powerful, and has a larger potential userbase.

You know about it, but didn't think it was worth mentioning in your post? The point isn't which is the most powerful, if Nvidia is even willing to license CUDA to ATI they will have to convince them to drop their own programming language first. This is going to delay the time it will take before we begin to see GPUs impact game play, maybe indefinitely if ATI pulls a 360.

I'm not sure who that quote is from, but the Cuda SDK actually includes some examples of Cuda combined with D3D or OpenGL, and it works pretty well. And that is with the standard programming interface. nVidia could easily implement some extra shortcuts and optimizations for physics in the drivers, making it work even better.
And where exactly did nVidia say they were bringing Cuda to consoles? Never heard anything like that, although I can imagine that they would say that if there's a next-gen console with GF8+ hardware planned.

They said it in the announcements they were buying Ageia, they keep claiming CUDA is an open platform when it will only run on Geforce 8 hardware.

Honestly I have stuff to do today so I cannot keep posting responses, but Scali2 and a few other people in here are unrealistically optimistic about CUDA. Whenever there is a very significant issue they down play it as not being a problem or something that ‘will’ somehow be resolved in the future, this isn’t going to be an easy or short transition and it is not in the interest of consumers for Nvidia to dominate the market.

 

[Atech]

lol, now i know why you are pissed, you bought a PhysX card

Actually I am happy with it...still no hardware to beat it
But you are you typing to me...did you get more than PR-fluf?
Like working software?

 

[imzjustplayin]

Please, study Cuda and the GF8-architecture before making such uninformed rants. A GF8 is nowhere near as dedicated as you think it is. Also, you don't seem to understand that we have unified shaders, which can do anything you like at any time... So you don't dedicate an entire GPU to physics, you just use some idle shaders to do physics. Hence my reference to HyperThreading.
You don't *have* to buy two GPUs, it will work with a single GPU aswell... It's just that ATi's latest product is a very popular 2-GPU solution, and nVidia has a product like that planned for the GF9-series. It seems that dual-GPU cards are the future.

If you're in an SLI setup, then I'd hope that one would be dedicated. Furthermore, this proves my point that GPU physics is a bad idea because all it means is that progression in the physics department is going to be hampered since there is barely any physics acceleration. Using "idle" shaders sounds like to me that the help would be so miniscule (since we're usually GPU limited anyways) that the benefit of another processing core would be far outweighed which completely defeats the WHOLE PURPOSE OF PHYSICS ACCELERATION!

 

[Scali2]

Havok FX was not based on DirectX 9; it utilised GPU math computing via the shader pipelines, it worked on *any* GPU with shader model 3, even if that GPU was OpenGL only like with the PS3. Using this method however meant it couldn’t communicate with the CPU, so the physics could not have events attached to them to influence game play.

I said DX9-level hardware, I didn't say it required DX9 itself. You actually say it too, because SM3.0 is actually a class of DX9 shaders, and has nothing to do with OpenGL or PS3. It's just how we classify hardware these days.

It is true CUDA does not have this same problem, but as long as it has the tiny install base and incompatibility with both ATI and consoles it will not be utilised in a way that makes it a mandatory requirement. Which means that like Havok FX it will only be used for effects; it can do affects, but they are not practical in its current state.

But you don't seem to understand that PhysX already exists and already runs on CPUs and consoles. nVidia is just *adding* Cuda support. This allows the PhysX API to transparently offload physics operations to Cuda-capable hardware, just as it now is able to do the same for PPUs.
Cuda and PhysX are not the same thing, so stop confusing Cuda requirements with PhysX requirements. PhysX is already widely in use in both PC and console software, it was even before the PPU existed (back when it was still called NovodeX, before it was acquired by Ageia).

No, they made sure that you had to buy Vista before they ticked the box needed to let you play with destructible game play online. It has nothing to do with protecting gamers from those with better hardware, they could have just marked servers with destructible game play and let player decide if they can handle it. Making it DX10 only was simply a favour for Microsoft and is no indication if you CPU and Internet connection is up to task.

That is completely beside the point here. Don't bring your pet peeves with DX10 or Vista into a thread that is about physics.

Again you are assuming time is a cure all and everyone will naturally adopt CUDA over time, not that simple.

Actually, it is. Tell me where exactly my logic of the largest percentage of the people buying nVidia and nVidia supporting Cuda on most current and future products does not result in most people naturally adopting Cuda over time?

The multiple levels are effects; game play physics will be the lowest setting. See this is were the concept of lowest common denominators come in, you target the weakest common spec in order to maximise potential audience. Everything above that is perks for those with better hardware, but they cannot impact game play in a way that makes it play different from the lowest common denominator.

It is how the industry works, target the weakest aspect of each platform you plan to launch on.

Not entirely true actually. In the case of Crysis, higher levels of physics mean more deformable and destructible objects, changing ways to kill enemies or be killed, and new ways of taking cover. The game isn't exactly the same on all physics levels.

You know about it, but didn't think it was worth mentioning in your post?

I assumed everyone here knew the difference between CTM and Cuda, guess I overestimated the audience.
Bottom line is, Cuda requires a specific GPU architecture, with hardware features that solely aid GPGPU, and have nothing to do with DX10 or OGL support. The way it looks, ATi doesn't have this architecture, so arguing that Cuda only supports GF8 is moot at this point. Cuda couldn't run on ATi GPUs even if they wanted to.
Then again, since ATi's marketshare is about half of nVidia's, it's ATi's problem, not nVidia's. ATi doesn't have the critical mass to make a difference.

They said it in the announcements they were buying Ageia, they keep claiming CUDA is an open platform when it will only run on Geforce 8 hardware.

Link to actual nVidia announcement saying such?
Are you sure they said Cuda, not PhysX? You seem to be confusing the two constantly.

Honestly I have stuff to do today so I cannot keep posting responses, but Scali2 and a few other people in here are unrealistically optimistic about CUDA. Whenever there is a very significant issue they down play it as not being a problem or something that ‘will’ somehow be resolved in the future, this isn’t going to be an easy or short transition and it is not in the interest of consumers for Nvidia to dominate the market.

Well I actually worked with Cuda, what do you know about Cuda and GF8 hardware?
No offense, but judging by your posts, you're not very informed on the subject.
It was the same when the PPU arrived. People downplayed it because "CPU can do physics aswell, and look, we get 4 cores now". Totally disregarding the fact that a PPU was far more powerful than any CPU on the market. Sure, there is a problem with actually getting software to support it... But that doesn't mean the technology isn't superior to CPUs running physics.
Also, I don't think Cuda will actually outperform the PPU, but I think it will fall somewhere between CPU and PPU performance, making it an interesting way to offload some physics to the GPU, especially in a multi-GPU system where there's idle time.
But I've already said all this, and nobody actually argued these points directly. So I hold on to them until someone proves me wrong.
Other than that, I really don't care about all this talk about politics and economics. This is the [H]ardforum and [H]ardware is what I'm interested in. I don't care who makes it, or who doesn't support it, and what company may go bankrupt because they can't keep up with technology, or what company gets a monopoly because they had a technology powerful enough to take over the world... That's just how it is.

 

[Scali2]

If you're in an SLI setup, then I'd hope that one would be dedicated.

Why? The physics workload may not be anywhere near enough to tax the GPU, it may be able to still significantly accelerate the graphics while it's at it.
We'll just have to see when the first working versions of Cuda-accelerated PhysX become available. Current physics workloads in games are aimed at CPUs that are busy with tons of other stuff aswell and are basically an insult to the raw power of a PPU, and I bet a high-end GF8 will laugh at these puny workloads aswell.

Furthermore, this proves my point that GPU physics is a bad idea because all it means is that progression in the physics department is going to be hampered since there is barely any physics acceleration. Using "idle" shaders sounds like to me that the help would be so miniscule (since we're usually GPU limited anyways) that the benefit of another processing core would be far outweighed which completely defeats the WHOLE PURPOSE OF PHYSICS ACCELERATION!

We'll just have to see. Just because you're GPU-limited doesn't mean that all the GPU's resources are in use. It just means that a certain resource in the chain is having trouble keeping up. This doesn't have to be the shader array.
In fact, in many cases the GPU is actually being limited by the driver, something that DX10 hopes to alleviate. In this scenario, offloading physics from the CPU to the GPU would be interesting even if it doesn't make the physics faster... because it would make the CPU able to spend more time on the driver.

 

[imzjustplayin]

Why? The physics workload may not be anywhere near enough to tax the GPU, it may be able to still significantly accelerate the graphics while it's at it.
We'll just have to see when the first working versions of Cuda-accelerated PhysX become available. Current physics workloads in games are aimed at CPUs that are busy with tons of other stuff aswell and are basically an insult to the raw power of a PPU, and I bet a high-end GF8 will laugh at these puny workloads aswell.

I'd only be "puny" if it was being underutilized. If you were to take the peak workload the physX PPU can do and put that on a GPU, it'd be far from "puny" by any means and I have a strong belief that even under the most ideal conditions, the GF8 would not be able to match its performance. The physX is a dedicated part, we have Core2duos with intel integrated graphics still running Quake3 like ass. Imagine if we stuck with CPUs today, no GPUs whatsoever, think about what video games would look like today. You know what they'd look like? It'd be looking at Quake 3, thats what.

Also don't be hypnotized by the concept that intel's latest integrated GPUs can run 3D Mark 2003, they're using more dedicated GPU hardware than they previously did which is why 3D mark 2003 is EVEN a possibility. If you were to run 3D mark 2003 on a core2duo with 4 cores, it'd still be a slideshow.

I know you didn't bring up CPUs but a lot of other people still consider multiple CPU cores as a viable extension to physics processing when the real answer is a dedicated PPU.

 

[imzjustplayin]

Why? The physics workload may not be anywhere near enough to tax the GPU, it may be able to still significantly accelerate the graphics while it's at it.
We'll just have to see when the first working versions of Cuda-accelerated PhysX become available. Current physics workloads in games are aimed at CPUs that are busy with tons of other stuff aswell and are basically an insult to the raw power of a PPU, and I bet a high-end GF8 will laugh at these puny workloads aswell.



We'll just have to see. Just because you're GPU-limited doesn't mean that all the GPU's resources are in use. It just means that a certain resource in the chain is having trouble keeping up. This doesn't have to be the shader array.
In fact, in many cases the GPU is actually being limited by the driver, something that DX10 hopes to alleviate. In this scenario, offloading physics from the CPU to the GPU would be interesting even if it doesn't make the physics faster... because it would make the CPU able to spend more time on the driver.

You should change that to "hoped", it was SUPPOSE to speed up things yet vista seemingly did the opposite to games, especially "DX10 titles".

 

[BillParrish]

.

Considering the news Nvidia is going to redesign the PhysX API to use CUDA for physics acceleration on the GPU, why isn’t anyone concerned?

Because it does not cost me anything for a new capability while there was no way I was going to shell out the cash for an additional speciality card that was not all that greatly supported, er make that "not at all" in the games I play ?

Y'all funny, only people annoyed should be Nvida stockholders that just saw a good chunk of Nvida cash pay for something Nvida is going to give away, apparently as a "feature" stick to further beat AMD severly around the ears and head with. Things grow or evolve or die and sometimes they get hit a speeding train and become FUBAR , eh, whaddayougonna do? If a PPU is such a great idea someone else will make a new and better one eventually, or not.

 

[Scali2]

I'd only be "puny" if it was being underutilized. If you were to take the peak workload the physX PPU can do and put that on a GPU, it'd be far from "puny" by any means and I have a strong belief that even under the most ideal conditions, the GF8 would not be able to match its performance. The physX is a dedicated part, we have Core2duos with intel integrated graphics still running Quake3 like ass.

Isn't that exactly what I just said?
"Also, I don't think Cuda will actually outperform the PPU, but I think it will fall somewhere between CPU and PPU performance"

Also don't be hypnotized by the concept that intel's latest integrated GPUs can run 3D Mark 2003, they're using more dedicated GPU hardware than they previously did which is why 3D mark 2003 is EVEN a possibility. If you were to run 3D mark 2003 on a core2duo with 4 cores, it'd still be a slideshow.

Why are you telling me this?

I know you didn't bring up CPUs but a lot of other people still consider multiple CPU cores as a viable extension to physics processing when the real answer is a dedicated PPU.

Yes, exactly what I just said. Mind you, this Cuda thing may still be a good step up from CPUs, even if it doesn't outperform PPUs.
At least it may end the catch-22 situation we have today... Why buy a PPU when there are no games... and why develop PPU support for games when nobody buys a PPU?
Cuda-capable hardware is already out there. It may give PhysX a nice foothold in the market, and then it'd be easier to expand GPU hardware for better physics support, or invest in further developing the dedicated PPU.

 

[Scali2]

You should change that to "hoped", it was SUPPOSE to speed up things yet vista seemingly did the opposite to games, especially "DX10 titles".

Yea well, we haven't seen a proper DX10-engine yet.
The DX10 driver system does work, it's just that nobody used it properly yet, including Crysis.
I've dabbled a bit with converting my DX9 engine to DX10, and while I don't have a complete, functional DX10 engine yet, what I've done so far works very well in DX10.

Don't confuse DX9 performance with DX10 either. DX9 will be slower on Vista than on XP by design, because Vista uses a wrapper to translate DX9 calls to the new DX10 driver model, which adds overhead. DX10 doesn't have this overhead, so DX9 performance on Vista says nothing about DX10.

 

[AnnoyedDragon]

That’s right; everyone who criticises CUDA is just confused and ignorant, unable to see with the same clarity as it’s supporters

I’ll address your responses and accusations of ignorance when I have more time at some point in the future. But the perception of Nvidia cannot do wrong and their solution is perfect in every way, and if you think otherwise you are just ignorant or mistaken, is becoming tiresome.

CUDA has issues, real undeniable issues that are going to impact how successful it can potentially be. If you don’t see any of that and think everything is just ‘rosy’ then I have to wonder what is fuelling this blind faith. Nvidia fanboyism? Hype? Plain over optimism? There are things about CUDA you cannot simply rationalise away, the fact some people in here seem to think they can makes me wonder what their motives are to spread such fud.

 

[Scali2]

That’s right; everyone who criticises CUDA is just confused and ignorant, unable to see with the same clarity as it’s supporters

I’ll address your responses and accusations of ignorance when I have more time at some point in the future. But the perception of Nvidia cannot do wrong and their solution is perfect in every way, and if you think otherwise you are just ignorant or mistaken, is becoming tiresome.

CUDA has issues, real undeniable issues that are going to impact how successful it can potentially be. If you don’t see any of that and think everything is just ‘rosy’ then I have to wonder what is fuelling this blind faith. Nvidia fanboyism? Hype? Plain over optimism? There are things about CUDA you cannot simply rationalise away, the fact some people in here seem to think they can makes me wonder what their motives are to spread such fud.

I think you're the one spreading fud here. Nobody claimed Cuda is the be-all-end-all of computing... We're just discussing the adaptation of Cuda to physics.

Instead, you come in here with unfunded claims and unrelated statements and try to put Cuda down at all costs... How about discussing things at a technical level instead?
A good start would be to accurately describe these 'things about Cuda' instead of posting some vague quote from some vague forum without even offering a link, so we have no idea who said it, in what context, etc... or if it even is a actual quote in the first place.

 

[AnnoyedDragon]

My problem, Scalie2, is every bit of valid criticism made by myself and others regarding CUDA has simply been shrugged off and ignored as unimportant. Only 18% of the market being able to use CUDA today ‘unimportant’, no CUDA support on ATI and console platforms ‘unimportant’. The way people are talking it is like they want an Nvidia monopoly; because that would solve allot of these problems, of course consumers get fucked in the process.

CUDA is going to bring next generation hardware physics to PC gaming and if you say anything to the contrary you are ignorant or confused. Maybe I would be able to look at the good side if concerns were actually addressed; but they are not being addressed, they are being put aside as something that will eventually solve themselves in the future.

And that’s the problem, myself and others are being expected to be happy with that answer and being repeatedly told we are in the wrong for daring to bring the subjects up.

You want that quote? Fine here, post 43. I am tired of being told I don’t know what I am talking about when you make mistakes in posts yourself, then deny them in the next one, yet I am expected to be happy with “it will just work out” with CUDAs issues. Your posts are taking a more offensive approach, talking about ‘overestimating’ people and other insults like that, if you keep it up the mood isn't going to stay civil.

 

[Scali2]

My problem, Scalie2, is every bit of valid criticism made by myself and others regarding CUDA has simply been shrugged off and ignored as unimportant. Only 18% of the market being able to use CUDA today ‘unimportant’, no CUDA support on ATI and console platforms ‘unimportant’. The way people are talking it is like they want an Nvidia monopoly; because that would solve allot of these problems, of course consumers get fucked in the process.

As I said, they aren't technical issues... It doesn't make the technology any less effective.
In fact, if this technology turns out to work well, it may just become another reason to buy new nVidia cards. So it's not 'valid criticism', it's just you having some kind of hangup about nVidia possibly getting a monopoly.

CUDA is going to bring next generation hardware physics to PC gaming and if you say anything to the contrary you are ignorant or confused.

That's not what I said... But your posts indicated a lack of understanding about Cuda and PhysX, as I already addressed.
I never claimed that Cuda will bring next generation hardware physics to PC gaming. In fact, I even said that I think the PPU solution will actually give superior performance.
Nevertheless, Cuda physics has some potential advantages over CPU-based physics and PPUs. It will offload your CPU, and you don't need to buy an extra card to use it.
You are actually the one who is ignoring all valid technical advantages of Cuda physics, and focus only on the "Ohnoes nVidia may get a monopoly"-issues.
What you're saying sounds more like you can't accept that Cuda physics might succeed, because you cannot accept the consequences that it may have for ATi.

You want that quote? Fine here, post 43. I am tired of being told I don’t know what I am talking about when you make mistakes in posts yourself, then deny them in the next one, yet I am expected to be happy with “it will just work out” with CUDAs issues.

Where exactly did I make mistakes in my posts?
And, no, you posting a link does not make you look like you know what your talking about. It's not exactly a developer forum anyway... I'm a developer and I don't agree with what he posted there. So what use is his post?
Especially when your own input makes no sense at times.

 

[InorganicMatter]

You want that quote? Fine here, post 43. I am tired of being told I don’t know what I am talking about when you make mistakes in posts yourself, then deny them in the next one, yet I am expected to be happy with “it will just work out” with CUDAs issues. Your posts are taking a more offensive approach, talking about ‘overestimating’ people and other insults like that, if you keep it up the mood isn't going to stay civil.

I have no idea where that person is getting their information, but it's wrong. Our very own [H] has an excellent article all about the GeForce 8's internals you really should read before bashing CUDA or stream processing:
http://www.hardocp.com/article.html?art=MTIxOCw1LCxoZW50aHVzaWFzdA==

 

[Scali2]

I have no idea where that person is getting their information, but it's wrong. Our very own [H] has an excellent article all about the GeForce 8's internals you really should read before bashing CUDA or stream processing:
http://www.hardocp.com/article.html?art=MTIxOCw1LCxoZW50aHVzaWFzdA==

Thanks for the link. That is exactly what I have been talking about... The GeForce 8800 series is 'designed for Cuda', and as far as is known, ATi's GPUs have no such design, but stick more or less to the old DirectX 9 style architecture, with improved shader capabilities.
The diagram on page 8 nicely demonstrates how different it is with the parallel data cache. A feature that exists solely for Cuda, and has no meaning for graphics.

It also says this:

The hardware architecture itself uses the same graphics architecture that is there in the GPU and simply treats it in a different way for doing thread computing. It uses the stream processors, texture and a special parallel data cache to do all of its work. The beauty of this system is that it can run at the same time that you are using the GPU for accelerating 3D graphics. Gamers will pick up on this real quick and realize that a single GeForce 8800 GTX and GTS is capable of doing both 3D acceleration and gameplay physics at the same time in the same game. The entire architecture is based on the standard C language, so adoption shouldn’t be that complicated.

Which is completely the opposite of the quote from that nVidia forum.

 

[AnnoyedDragon]

How can an incompatibility with the vast majority of hardware in the market ‘not’ be a technical issue? Again it is like you are expecting a monopoly; the only way Nvidia wouldn’t have to worry about ATI compatibility issues is if Nvidia dominated them, either through killing them off or putting them in a position where they are forced to license CUDA.

A tiny fraction of the market supporting this thing looks like a ‘big’ issue to me, with the limited platform support being the next ‘big’ problem, being told I should stop thinking about it and simply shouldn’t worry doesn’t sit well.

You forget I backed hardware accelerated physics for over a year, I want a hardware solution. But I don’t want one that punishes gamers for the ‘brand name’ of their hardware, if there was a standardised solution we wouldn’t be having this conversation right now. I don’t care if something similar may or may not have happened in the past, it is a problem and it has to be taken into consideration when looking at CUDA now.

A parallel processor is a parallel processor right? Well not when it is only accessible via one company in a market of several; it is actually making me lean more towards software physics than hardware, CPUs have no where near the same computational power but at least physics on them will work on every hardware platform available.

So a guy talking about CUDA ‘on the official Nvidia forum’ right next to the ‘official CUDA section’ where CUDA programmers chat on a daily basis may not be as reliable as I thought. I stated in the post I had yet to confirm the information, if you think you have valid evidence to suggest they are wrong then fine case closed. But I am not going to put the install base and platform support issues to rest until someone gives me a decent reason to, not simply ‘because’ it will ‘probably’ be sorted at some point.

Where exactly did I make mistakes in my posts?

When you didn’t mention ATI’s solution when saying they didn’t have a GPU programming SDK like CUDA; then said you knew all about it when it was brought up, claiming you didn’t mention it because you ‘overestimated’ people which is a really offensive comment may I add. Then when you said Havok FX was DX9 based, after being corrected it was only shader model 3 reliant and not DirectX you twisted it to make it sound like that is what you was saying to begin with.

No way to prove/disprove foreknowledge on the Internet after it has already been posted, just remember everyone makes mistakes.

 

[Scali2]

How can an incompatibility with the vast majority of hardware in the market ‘not’ be a technical issue?

Because it isn't. Whether or not people have the hardware to use this technology has no effect on the existence and functionality of the technology itself.
What you're saying is something like "Rocket technology doesn't work because most people don't own a rocket".

Again it is like you are expecting a monopoly; the only way Nvidia wouldn’t have to worry about ATI compatibility issues is if Nvidia dominated them, either through killing them off or putting them in a position where they are forced to license CUDA.

That is a political issue, not a technical issue.
I don't care about politics.

A parallel processor is a parallel processor right? Well not when it is only accessible via one company in a market of several;

You're not making sense. You're trying to say that if only one company makes a parallel processor, it isn't parallel. That's just nonsense.

So a guy talking about CUDA ‘on the official Nvidia forum’ right next to the ‘official CUDA section’ where CUDA programmers chat on a daily basis may not be as reliable as I thought.

Indeed it is not. Most people are full of shit. The internet gives them a medium to post that shit, day in day out. Welcome to the internet.

When you didn’t mention ATI’s solution when saying they didn’t have a GPU programming SDK like CUDA; then said you knew all about it when it was brought up, claiming you didn’t mention it because you ‘overestimated’ people which is a really offensive comment may I add.

Yes, I said ATi didn't have a GPU programming SDK like Cuda. Which it doesn't, because as I explained, CTM is nowhere near Cuda, hence it isn't 'like Cuda'. That was my point. I specifically used the words 'like Cuda' to signify that what they did have was no match for Cuda. I later clarified myself by explaining how Cuda has special hardware support like the Parallel Data Cache, and how it uses a C++-like programming language, where CTM is still aimed at X1000-hardware and programming at assembly level.

Then when you said Havok FX was DX9 based, after being corrected it was only shader model 3 reliant and not DirectX you twisted it to make it sound like that is what you was saying to begin with.

I didn't say it was DX9-based. Re-read my post. It says it was "based on DX9-class hardware". Never anything about being based on any DX9 software.
And yes, Shader Model 3 is part of the DX9-spec. The specific term and version number is specific to DirectX, and have nothing to do with OpenGL or PS3. As I already said, the DX9 spec is just the most comman way to classify hardware generations and functionality. You do it yourself, you just don't even realize you're doing it, because apparently you aren't familiar with how OpenGL and PS3 name their things.

So basically you're the one making a mistake, and failing to understand what I said, even after I tried to clarify myself.
And you think it's weird that I don't have the impression that you know what you're talking about?

 

[R3MF]

CUDA has issues, real undeniable issues that are going to impact how successful it can potentially be. If you don’t see any of that and think everything is just ‘rosy’ then I have to wonder what is fuelling this blind faith. Nvidia fanboyism? Hype? Plain over optimism? There are things about CUDA you cannot simply rationalise away, the fact some people in here seem to think they can makes me wonder what their motives are to spread such fud.

what are these issues?

 

[Scali2]

what are these issues?

I think we've narrowed it down to "Non-nVidia cards and older nVidia cards don't support Cuda"...
But really, we all were fully aware of that already... it's painfully obvious. It's not exactly anything new either. Surprise surprise, back when Intel invented the microprocessor, it wasn't compatible with anything else either... It wasn't until much later when clones like AMD appeared.
I also recall the era of the first soundcards, first Adlib, then Soundblaster... and for a long time every card had to be Adlib/SB-compatible to be worth anything, until they standardized audio in DirectX at last.
And we can all remember the first 3d cards, Voodoo with Glide API... Nobody else supported that.
Hey, it all worked out.
So really, I don't care about nVidia being the only one. The important thing is that this kind of technology gets out there. We can worry about licensing and APIs later. DirectX will probably get some kind of addon when there are multiple vendors offering hardware solutions. Currently nVidia is the only one. ATi and Intel will join nVidia at some point, and only then can a standard be derived.

 

[AnnoyedDragon]

Because it isn't. Whether or not people have the hardware to use this technology has no effect on the existence and functionality of the technology itself.
What you're saying is something like "Rocket technology doesn't work because most people don't own a rocket".

That’s complete nonsense, I am not questioning the technology I am questioning the feasibility of its integration into gaming when only a small number of people can use it. Install base saturates, you still can’t use it for anything other than effects because then the games won’t work for ATI and console users.

That is a political issue, not a technical issue.
I don't care about politics.

Well you should; it dictates whether or not Nvidia’s CUDA solution will improve game play or just be a pretty effect, regardless of how good it is it won’t survive if it cannot be integrated into gaming effectively. Besides monopolies are something every consumer should be concerned about, do you want to pay bloated prices?

You're not making sense. You're trying to say that if only one company makes a parallel processor, it isn't parallel. That's just nonsense.

People looked at the PPU and GPU and thought what was the difference? They are both parallel processors so what does it matter which one it is done on?

Well Nvidia’s solution only works on Nvidia cards while the PPU worked with any hardware brand name you wanted. They are both parallel processors capable of physics; but the GPU has a limited potential reach, making it not as simple as moving physics from one parallel processor to another.

Yes, I said ATi didn't have a GPU programming SDK like Cuda. Which it doesn't, because as I explained, CTM is nowhere near Cuda, hence it isn't 'like Cuda'. That was my point. I specifically used the words 'like Cuda' to signifify that what they did have was no match for Cuda.

Doing well or not, ATI’s solution is still a competitor to CUDA, saying they don’t have one is misleading even if their one is crap.

I didn't say it was DX9-based. Re-read my post. It says it was "based on DX9-class hardware". Never anything about being based on any DX9 software.
And yes, Shader Model 3 is part of the DX9-spec. The specific term and version number is specific to DirectX, and have nothing to do with OpenGL or PS3. As I already said, the DX9 spec is just the most comman way to classify hardware generations and functionality. You do it yourself, you just don't even realize you're doing it, because apparently you aren't familiar with how OpenGL and PS3 name their things.

You should have just said pixel shader 3, honestly it is easier and less potentially misleading than "based on DX9-class hardware".

So basically you're the one making a mistake, and failing to understand what I said, even after I tried to clarify myself.
And you think it's weird that I don't have the impression that you know what you're talking about?

And judging the responses I am having to type you are not understanding some of what I am saying; of course in your case there is going to be some image saving excuse isn’t there? For me your just going to stamp idiot or something equally insulting.

Speaking of clarifying yourself, I am still waiting for you to clarify why exactly you think the platform support issues are not a problem? I’m supposed to not worry about this while most of what I have been given is closer to crystal ball gazing than hard facts.

what are these issues?

-Only the 8 series is compatible with CUDA, it has not reached a majority install base so targeting it for game play in the short term cannot be justified

-Incompatibility with current generation consoles when the games market is console orientated; with the majority of game titles today having to support both console and PC platforms, only PC platform exclusive games can potentially support CUDA game play physics.

-Incompatibility with ATI based graphics cards, creating a segment of the market that will be unable to access CUDA physics simply because they are on a different brand of GPU.

-CUDA game support bumping up the GPU system requirements for games, requiring a SLI solution to truly take advantage of it. This is unpractical for the majority of the market who are on mainstream priced hardware; if all they could afford was a 8600GT to begin with the needed performance jump for CUDA is out of their reach. People on high end systems may wonder why they should care if someone in the lower end cannot afford it? Well that lower end dictates game play, if they cannot afford it you won’t get it.

Pretty much it is limited to PC, limited to Nvidia and limited to the 8 series. This is too small a segment to justify designing a game that will only work for them. Even when the install base increases and CUDA is licensed to ATI, incompatibility on consoles and the dramatically increase GPU requirements will deter developers using it for game play physics.

Scali2 is down playing these problems, making them sound like they are not that big an issue. I honestly think it is going to significantly hold back CUDA from bringing anything new to game play for a couple of years, at least until all compatibility and performance issues are dealt with. People arguing for CUDA seem to be relying on all the competing hardware developers putting aside their differences and uniting under Nvidia’s CUDA flag. Something that is not only unrealistic, but if it does take place it will be too far in the future to care about CUDA at this point.

 

[Scali2]

That’s complete nonsense, I am not questioning the technology I am questioning the feasibility of its integration into gaming when only a small number of people can use it.

You claimed it was a 'technical issue'. Looks like you changed your mind on that now.

Well Nvidia’s solution only works on Nvidia cards while the PPU worked with any hardware brand name you wanted. They are both parallel processors capable of physics; but the GPU has a limited potential reach, making it not as simple as moving physics from one parallel processor to another.

There is only one PPU, made by only one manufacturer, and the PhysX API only supports that one PPU. In that sense it's no different.
You could combine that PPU with any GPU... but you can still do that now. You just don't necessarily need to if you have Cuda support, a nice bonus for nVidia customers.

Doing well or not, ATI’s solution is still a competitor to CUDA, saying they don’t have one is misleading even if their one is crap.

They don't have a solution. You can hardly call an assembly interface on last-gen hardware a competitor to Cuda and PhysX.

You should have just said pixel shader 3, honestly it is easier and less potentially misleading than "based on DX9-class hardware".

I think they're pretty much equivalent in daily use, and I think most people here would agree with me, so you're making a big deal out of nothing.

Speaking of clarifying yourself, I am still waiting for you to clarify why exactly you think the platform support issues are not a problem? I’m supposed to not worry about this while most of what I have been given is closer to crystal ball gazing than hard facts.

As I said, they're not technical issues. You can argue about them all day, but in the end you have no idea how well this technology will be adopted until it has actually been on the market for a while.
I don't care about that sort of 'crystal ball logic'. I go to tech forums to discuss technology. Things that have solid answers to questions, because they're based on logic and physics, not on what some guy pulls out of his arse. I hate people who think they can predict the future. They're always wrong anyway. Look at how people predicted the success of AMD's Barcelona, lol. Bunch of idiots. There was nothing in the technology indicating success for AMD, they just got caught up in the hype. This time there seems to be a 'negative hype' about Cuda... same as there was a 'negative hype' about the PhysX PPU. The technology works, but people don't want to adopt it.

I only focused on the fact that you claimed there were technical issues with Cuda that would stop it from being a working solution, or whatever you wanted to make it sound like.
I wanted to point out that the technology is pretty solid.
And again, in your list, you sum up ONLY Cuda issues. You completely ignore the fact that the API already has support for multithreaded CPUs, the PhysX PPU and various consoles, and nVidia has already testified that is not dropping that support.
So really, who cares if it's not going to use Cuda on every platform? It will just use other means of acceleration if present, be it a multithreaded CPU, a PPU, or a custom console CPU. I'm not downplaying anything, this is the reality. PhysX is already out there, supporting these platforms... the only platform it doesn't support yet is... Cuda.
Nobody ever expected Cuda to work on ATi hardware anyway, that should be pretty obvious. I really don't see what you're trying to argue. As of yet, ATi has no physics solution at all, so arguing about whether or not it wil be compatible with PhysX is useless at this point. We can't say anything about compatibility of something that doesn't exist in the first place.
Also, you are pulling that SLI-requirement out of your arse, because it is way too soon to know anything about performance levels on single vs dual-GPU setups.
The 8600-series doesn't support Cuda anyway, so your argument is flawed. Nobody who bought an 8600 was promised Cuda or physics acceleration, unlike the 8800 series.
Edit: Looks like they changed it now... 8600 actually is listed with Cuda support. I'm 100% sure that it wasn't back when I bought my 8800GTS... It was actually one of the reasons to go for 8800 series instead of 8600. I also wonder if it is actually supported yet. Driver support for Cuda has been rather flaky so far... Only supporting XP 32-bit anyway.

 

[InorganicMatter]

How can an incompatibility with the vast majority of hardware in the market ‘not’ be a technical issue?

Funny. That sounds like a certain proprietary and expensive PCI card that was just killed off...

Again it is like you are expecting a monopoly; the only way Nvidia wouldn’t have to worry about ATI compatibility issues is if Nvidia dominated them, either through killing them off or putting them in a position where they are forced to license CUDA.

A tiny fraction of the market supporting this thing looks like a ‘big’ issue to me, with the limited platform support being the next ‘big’ problem, being told I should stop thinking about it and simply shouldn’t worry doesn’t sit well.

A tiny fraction of the market only supports DirectX10, yet I don't see developers shying away from DirectX10. New APIs come along, and require new hardware, the old stuff doesn't support it, too bad. As far as ATI, that's their own fault. I seem to remember them being in denial about GPU physics a few months ago and now it's suddenly blown up in their face. It's too bad that they didn't get in on the action, but it's really no one's fault but their own. CUDA is an excellent idea and architecture that ATI should have seriously considered licensing from NVIDIA, yet chose not to (probably another royal screwup, courtesy of Wrector Ruinz).

 

[AnnoyedDragon]

In an effort to avoid these page long posts I’ll try to sum it up quickly without quotes.

Because it is a technical issue that is making me question the feasibility of its integration in the first place. The exclusive compatibility with only Nvidia hardware, haven’t I gone on about it long enough that it should be obvious that is what I am referring to? Yes, we knew from the start ATI is not compatible, but that’s what I have a problem with. As long as that remains the case I don’t see CUDA ‘game play’ taking off on PC, developers won’t want to limit their potential audience.

Regarding there only being one PPU, it could be used with either ATI or Nvidia, it has no hardware bias. You can have a PPU with any other hardware you want, CUDA? Nvidia only. Stuff like that deters developers, they are willing to give a hardware sponsor extra perks but not exclusivity of a game. That would be like if Intel made Alan Wake work on only Intel processors, regardless of the justification you shouldn’t alienate your audience based on the brand name of their hardware. If they have the performance, hell they deserve to play it.

And… I’m done. Not done done because I did have more to say in response, but over in my side of the world it is late. I just hope I’m getting across that CUDA being Nvidia exclusive is what bothers me, if they licence it to ATI I would have less reasons to not like it. But as it stand it will be a while before that happens, if it ever does, so until we know for sure CUDA looks unpractical to me.

On a side note InorganicMatter, How is DX10 comparable to CUDA? It has no impact on game play, it is supported on both ATI/Nvidia cards and it can be turned off to allow none DX10 hardware to play the game. It doesn’t matter if you support DX10 for the minority because it isn’t going to limit your audience.

 

[TerranUp16]

Well, here's something else to chew-on:

What if nVidia develops the capability to SLI two different card models? That means I can still get my 9800GX2 (or whatever) primarily for graphics and then go and get an 8800GT or 9600GT or something (or perhaps something cheaper) primarily for physics. But both have the capability to help each other out w/either the graphics or physics (thanks to the magic of stream processors). Add-in linking integrated graphics with those cards, and you have a system that now has a lot of flexibility as far as physics and graphics processing goes. And this isn't even evaluating the fact that I'm sure the cpu can help-out as well. Granted, that's a lot of things to coordinate, but they're all multi-purpose and are useful for more than just physics. I don't think nVidia has had much of a reason to develop SLI to work with different gpu's, but now they definitely do. Sure, you could claim that buying a cheaper gpu primarily for physics is like buying a PPU, but does your PPU do graphics as well?

 

[Scali2]

Because it is a technical issue that is making me question the feasibility of its integration in the first place. The exclusive compatibility with only Nvidia hardware, haven’t I gone on about it long enough that it should be obvious that is what I am referring to? Yes, we knew from the start ATI is not compatible, but that’s what I have a problem with. As long as that remains the case I don’t see CUDA ‘game play’ taking off on PC, developers won’t want to limit their potential audience.

Regarding there only being one PPU, it could be used with either ATI or Nvidia, it has no hardware bias. You can have a PPU with any other hardware you want, CUDA? Nvidia only. Stuff like that deters developers, they are willing to give a hardware sponsor extra perks but not exclusivity of a game. That would be like if Intel made Alan Wake work on only Intel processors, regardless of the justification you shouldn’t alienate your audience based on the brand name of their hardware. If they have the performance, hell they deserve to play it.

Firstly, it is NOT a technical issue, as you've already admitted in your previous post.
Secondly, it's up to the developers to decide. Just because you don't like it doesn't mean that no developer will support it.
We'll just have to wait and see. I wouldn't be surprised if various games implemented a special level of physics in games that is only playable with a PPU or Cuda.
Alternatively they could use Cuda for people with slow CPUs and an onboard GPU to accelerate physics, or something like that, which is a possibility you've ignored all along.

 

[silent-circuit]

Why is anyone paying attention to this guy in the first place? The whole thread borders on trolling, start to finish.

It can be summed up as "progress is bad an I don't have $160 to throw down on an 8800GS, or at very least can't justify the price, or I just don't like Nvidia's face."

Oh -- and I ran 3DfX (Voodoo3 3500 AGP), and ATI (Rage3D AGP, Radeon AGP, Radeon 9700 AGP, Radeon X800XT AGP) GPUs before my 8800GTXs, and AMD CPUs (1 Ghz Athlon socket A, Athlon 64 3500+ socket 939) before my Intel E6420 and X3220, so don't pull that "he's a fanboy!" crap on me. I go where the power is.

The thing I see repeated here (and the main reason this debate seems completely idiotic to me) is this whole debate over the "use" of CUDA. Nvidia wrote a wrapper -- PhysX API to CUDA -- and that's all. The API is the same. It will still run on the PPUs already on the market. It will run on lower end 8 series GPUs you can pick up for $60 or so if you have an empty 16x PCIe slot. They intend to make it run on more hardware. It is to their advantage to make it as readily accessible as possible, so more developers use it and pay licensing fees. Ageia's obsession with their overpriced, under-performing PCI based PPU was a very bad thing for the huge market segment of gamers who didn't have $300+ to throw away once the rest of their system was already perfect. Nvidia buying Ageia will open the use of PhysX to far more people and maybe give developers real incentive to implement higher end physics effects. This is a potentially good thing, as PhysX may mature and actually end up worthwhile... and at worst? It remains the joke it currently is, and there's zero damage cause by the acquisition over all.

 

[AnnoyedDragon]

Everyone is entitled to an opinion on this Silent-Circuit, not just the people already for the technology.

I said elsewhere in the forum that I really doubt an 8400GS is going to be much use in physics; we are talking about 16 unified shaders here, based on performance demos available that won’t even beat a CPU rendering it pointless. 8600GT has 32 unified shaders and is within the same price range as the PPU, 32 while a significant amount isn’t going to kill a single 8800 in a game either, making it the most likely candidate for a secondary physics GPU. Targeting anything in the 8800 performance is a bad idea; you have to make sure more than just the high end can access this.

Despite everyone claiming they are taking the ATI and console incompatibility into account; no one seems to think it matters, I wouldn’t call it ‘zero’ damage for a game designed to use CUDA for game play to not work on either of these two. Using licensing as a cure all when we have no idea if Nvidia is even going to make this technology available to competitors is a bit of a leap of faith; even if they did, there is no guarantee they will take it.

 

[Scali2]

I said elsewhere in the forum that I really doubt an 8400GS is going to be much use in physics; we are talking about 16 unified shaders here, based on performance demos available that won’t even beat a CPU rendering it pointless. 8600GT has 32 unified shaders and is within the same price range as the PPU, 32 while a significant amount isn’t going to kill a single 8800 in a game either, making it the most likely candidate for a secondary physics GPU. Targeting anything in the 8800 performance is a bad idea; you have to make sure more than just the high end can access this.

I don't understand what you're trying to say here...
Firstly, what performance demos are you talking about? I have yet to see any working PhysX stuff on Cuda, nVidia only just started developing it. So I think you are jumping to conclusions about performance. Nobody knows how it performs yet, not even nVidia.
Secondly, what's your point... Low-end hardware cannot be used for high-end gaming? Well, we already knew that. Low-end CPUs and GPUs are just as useless at gaming as a low-end Cuda system is for physics. I really don't see the point.

Despite everyone claiming they are taking the ATI and console incompatibility into account; no one seems to think it matters, I wouldn’t call it ‘zero’ damage for a game designed to use CUDA for game play to not work on either of these two. Using licensing as a cure all when we have no idea if Nvidia is even going to make this technology available to competitors is a bit of a leap of faith; even if they did, there is no guarantee they will take it.

As you said yourself, everyone is entitled to their opinion. Did it ever occur to you that some people don't care about that sort of political issues?
Do you realize that some people just buy their hardware and software based on what *they* want to do, completely disregarding compatibility issues, because they are insignificant to them?
I mean, I might want to have some fast and accurate realtime physics for some scientific simulations that I'm doing for research purposes. All I would care about is that I get results quickly and accurately. Whether other people can run it on their PS3 or not... it doesn't matter one bit to me, because I'm not using a PS3. And it probably doesn't matter to these people either, because they're not interested in my simulations anyway.
You seem to live in a VERY small world.

 

[silent-circuit]

-snip-

Despite everyone claiming they are taking the ATI and console incompatibility into account; no one seems to think it matters, I wouldn’t call it ‘zero’ damage for a game designed to use CUDA for game play to not work on either of these two. Using licensing as a cure all when we have no idea if Nvidia is even going to make this technology available to competitors is a bit of a leap of faith; even if they did, there is no guarantee they will take it.

"...I wouldn’t call it ‘zero’ damage for a game designed to use CUDA for game play to not work on either of these two."

This statement is the crux of your current (you've been through several by now) argument. It is flat out wrong. You seem to be under the impression that developers would at some point have to actually write CUDA code. They will not -- ever. They'll just have to make use of the PhysX API.

Then, if a game includes PhysX support (because they used the PhysX API, just as they would've to support Ageia's PPU before the CUDA wrapper, writing no Nvidia-specific code of any kind) the latest revision of the PhysX software will look for an 8 series Nvidia card (along with the PPU, as it always would've, as well as software fallback, which Nvidia claims to be expanding -- good news for those with quads -- and other yet to be announced hardware) and talk to the Nvidia card via CUDA. CUDA is only the "messenger" between PhysX, an existing API, and the 8 series cards. No one but Nvidia and their PhysX team will be writing anything in CUDA. In short, don't shoot the messenger.



"Using licensing as a cure all when we have no idea if Nvidia is even going to make this technology available to competitors is a bit of a leap of faith; even if they did, there is no guarantee they will take it."

They've already stated, in the preliminary interview (I saw a quote earlier -- I believe in a thread you started) that they would be opening this up to a wider market and more types of hardware. PhysX can be used like Havok -- you don't actually need a PPU or whatever, it just makes the effects faster (in theory). There've been hacks around to force software mode in PhysX demos for quite some time, and the performance hit is minimal. With this in mind, it's already compatible with all x86 based platforms including current Macs and the Xbox 360. I'm not sure about the PS3, but writing a version for it should not by any means be impossible, and Cell's parallel nature would actually be advantageous. In the end there is no console incompatibility.

As for the competitors "not taking" compatibility, I'm not entirely sure what you're getting at here, but... if the API supports a given piece of hardware for use as a PPU or physics aid, then the competitor won't have much choice. If the hardware is available, present in the machine, it will be used.



To everyone in this thread:
They aren't "using" CUDA in a potentially detrimental way -- it's just a wrapper. The API hasn't changed substantially, PPUs will still work, there's still x86 CPU support and supposedly it's being expanded, along with support for more hardware. The argument is pointless, and it's far too early to pass a difinitive judgement on any of this stuff. Give it six months to a year -- development cycles still take time, believe it or not.

 

[InorganicMatter]

I said elsewhere in the forum that I really doubt an 8400GS is going to be much use in physics; we are talking about 16 unified shaders here, based on performance demos available that won’t even beat a CPU rendering it pointless. 8600GT has 32 unified shaders and is within the same price range as the PPU, 32 while a significant amount isn’t going to kill a single 8800 in a game either, making it the most likely candidate for a secondary physics GPU. Targeting anything in the 8800 performance is a bad idea; you have to make sure more than just the high end can access this.

You don't know ANYTHING about how it will perform, no one does, so stop acting like an 8400GS isn't going to cut it. The amount of vector calculus that goes into processing something like specular lighting alone is a monstrous amount of mathematics; there is no way even the most complicated of dynamic physical systems could have as much math as graphical processing. Think about how little information needs to be processed: particle shape, mass, x/y/z acceleration, x/y/z velocity. Now take that versus something like applying specular lighting, one of MANY special effects, to the graphics:
http://en.wikipedia.org/wiki/Specular_highlight
Yeah, draw your own conclusions.