ATI/Nvidia GPU Physics not as good as Ageia's

[sladesurfer]

http://www.theinquirer.net/default.aspx?article=32810

According to The INQ, an anonymous game developer reported that ATI/NVIDIA physics is nowhere near as good as Ageia's physics card. The developer reported that both ATI and NVIDIA's physics solution only "cosmetically" affects the physics. Even high end cards like the X1900 XT and 7900 GT can not accomplish the same quality level of physics that Ageia's cards can.

[Stuperman]

WAIT JUST A GOD DAMN MINUTE! you mean that a chip designed with the intent of accelerating graphics accelerates physics more slowly than a chip designed to accelerate physics? Next you'll tell me that my RAM is better random access memory than my page files.

[BlackX]

http://www.theinquirer.net/default.aspx?article=32810

So why the hell it was said that X1600 is 3 times better then Ageia? :slapass:

I believe this is true and I'll buy Ageia PhysX, just hope price will drop.
300$ for PPU :censored:

[Cobalt]

We've known this for quite some time. There were several articles on it when the idea of a GPU PPU started. Unfortunatly for Ageia (and us) if they continue to charge such an extortionate price for the card then the ATi/nV solutions will be prefered from a consumer point of view.

So why the hell it was said that X1600 is 3 times better then Ageia? :slapass:

I believe this is true and I'll buy Ageia PhysX, just hope price will drop.
300$ for PPU :censored:
Its called marketing. And the X1600 probably can compute 3 times faster but not the right type of calulations.

[ahmad]

Seems like Ageia is making deals with these "anonymous" devs.

[Cybercat]

Seems like Ageia is making deals with these "anonymous" devs.Is it hard for you to concieve that this might be true?

[Shpoon]

Oh no! Now I'm not going to waste my money on another videocard because I want the physics one, which performs better on the one supported game!

Like I'd buy either atm.

[BlackX]

We've known this for quite some time. There were several articles on it when the idea of a GPU PPU started. Unfortunatly for Ageia (and us) if they continue to charge such an extortionate price for the card then the ATi/nV solutions will be prefered from a consumer point of view.


Its called marketing. And the X1600 probably can compute 3 times faster but not the right type of calulations.

Yeah and this sux. I think comparing PPUs we care about the quality of effects not about the quantity.

[Starscream]

so to be short:
a card that costs 300 euro is better at physix then a 100 euro grafix card.

i always accepted the fact that Ageias card would be faster at physix then a 7600 or x1600 the thing that matters is by how much.

seeing Ageias solution costs 3x more it better also perform 3x better.

that and Ageia has ot get their ass up and stop putting all their attention at the PR department and actualy make shure somethign gets delivered to us.

As their PR department is working asif there are already tons of physix games oput etc.
As Ageias soluion stil hasnt proved itself.

[Ubermann]

And how did this "source" bench this ?
There are no drivers nor are there any benchmarks..

I dont care if GFX card is slower or faster but this "news" totaly sux!
I has no info at all..

[nn_step]

until they put a PCIe card on the market. they can take their advantage and shove it up their......

[situman]

WHo cares! Someone should just buyout Agea and integrate their chip into a gpu. If I am going to spend 300 bucks on a vidcard, and another 300 on the damn phsyics card, I might as well go SLI.

[comment]

so to be short:
a card that costs 300 euro is better at physix then a 100 euro grafix card.

i always accepted the fact that Ageias card would be faster at physix then a 7600 or x1600 the thing that matters is by how much.

seeing Ageias solution costs 3x more it better also perform 3x better.

that and Ageia has ot get their ass up and stop putting all their attention at the PR department and actualy make shure somethign gets delivered to us.

As their PR department is working asif there are already tons of physix games oput etc.
As Ageias soluion stil hasnt proved itself.You (and others in this thread) are missing the point. It's not just a question about PhysX being "faster at physics" then GPU/VPU based physics.

Ageia's PhysX PPU is in a league of its own, being the only technology that is able to offload "gameplay physics" from the CPU. Neither ATI's nor nVidia's physics solutions can manage that. Enhancing physics acceleration is not only about looks. It’s also – and IMO first and foremost – about the way we play games, and enhanced “gameplay physics” radically changes that.

[Cobalt]

@comment: I don't thnk anyone is missing the point. You have to stop thinking about it from a purely technical perspective. At the moment (and when GPU physics is released) the ageia costs 3x more with little benifits. Some games actually perform worse with the physx because the GFX has to work harder. I'm not going to pay £200 to get worse performance, no matter how pretty it looks. On the other hand if I can use my old GFX to enhance gameplay (even if it is just for looks) then I'll jump on the opportunity.

Numbers mean nothing. The Cell processor is a good example of that. Despite its massive theoretical power it can't be used. The same problem is in effect here. Physx is very specific and good at what it does but nothing supports it. GPUs aren't very good but are much cheaper.

[LOE]

ahahahahaha LOL an "anonymous" DEV reports AGEIA has a more powerfull solution than ati and nvidia...

that's sad, and the fact the INQ knows it is BS it is still reporting it. I lost my "faith" into inquier the moment I sent them an article concerning a true problem with the hope they publish it so ppl may realise the TRUTH. But they bended and twisted the whole story in a way that is hard to imagine, managet to LIE about each fact... I hope at least those a$$holes got some good cash from the person who "ORDERED" them to LIE about stuff.

[Rovtar]

who the fu*k needs Ageia if it doesn't have game support (only ghost recon if i know right)

[Starscream]

You (and others in this thread) are missing the point. It's not just a question about PhysX being "faster at physics" then GPU/VPU based physics.

Ageia's PhysX PPU is in a league of its own, being the only technology that is able to offload "gameplay physics" from the CPU. Neither ATI's nor nVidia's physics solutions can manage that. Enhancing physics acceleration is not only about looks. It’s also – and IMO first and foremost – about the way we play games, and enhanced “gameplay physics” radically changes that.



ofc Ageias Physix card si way better and completely made for physix etc but that doesnt matter.
atm Ageias physix card hasnt proven itself one bit and already Ageia is throwing a marketing campaign stating they kick ass.
At the end of the day it doesnt matter whos better it matters who sells the most.

[Cobalt]

Thats what I was trying to get across

[LeoXV]

Best Bang for your buck

two grafix cards
supported games- run Ati/Nvidia physics
unsupported games- run Sli/CrossFire

grafix card + PhysX
PhysX-supported games- faster than Ati/Nvidia physics
unsupported games- will only drain power

[LOE]

don't you understand, physics is driven by a CHIP... ati and nvidia are FAR SUPERIOR in chip design than AGEIA, the PPU units are not very different than a shading unit. ATI and Nvidia have years of experience, and there is no doubt then can make hardware that is a lot faster and chepaer than AGEIA

AGEIA's claims are stupid, the chip can process whatever data you feed it, physics won't come for at least 6 months - enough time for ati and nvidia to include each feature required bu devs.

This article is total BS... "anonymous" DEV my a$$ :)
So why anonymous? Afraid nvidia might send a hitman to waste him, LOL, that's as pathetic as ATI slides :)

[[XC] Lead Head]

Best Bang for your buck

two grafix cards
supported games- run Ati/Nvidia physics
unsupported games- run Sli/CrossFire

grafix card + PhysX
PhysX-supported games- faster than Ati/Nvidia physics
unsupported games- will only drain power

No. 90% of the poepel going GPU physics, are going to get a lower end card..

[Cybercat]

so to be short:
a card that costs 300 euro is better at physix then a 100 euro grafix card.

i always accepted the fact that Ageias card would be faster at physix then a 7600 or x1600 the thing that matters is by how much.

seeing Ageias solution costs 3x more it better also perform 3x better.

that and Ageia has ot get their ass up and stop putting all their attention at the PR department and actualy make shure somethign gets delivered to us.

As their PR department is working asif there are already tons of physix games oput etc.
As Ageias soluion stil hasnt proved itself.It's not about performance, it's about the fact that Ageia's physics can be interactive, a GPU can't do that.

The article never mentioned performance.

don't you understand, physics is driven by a CHIP... ati and nvidia are FAR SUPERIOR in chip design than AGEIA, the PPU units are not very different than a shading unit. ATI and Nvidia have years of experience, and there is no doubt then can make hardware that is a lot faster and chepaer than AGEIA

AGEIA's claims are stupid, the chip can process whatever data you feed it, physics won't come for at least 6 months - enough time for ati and nvidia to include each feature required bu devs.

This article is total BS... "anonymous" DEV my a$$ :)
So why anonymous? Afraid nvidia might send a hitman to waste him, LOL, that's as pathetic as ATI slides :)Yes, because any old chip can run physics! Hell, why not take an MP3 decoder out of an iPod and program it to run physics?

It doesn't matter if ATI and NVIDIA's chip is more complex (and by complex, we can only judge my transistor count and make assumptions, we can't make direct comparisons), it's made to run graphics, it's not suited to run physics, and any attempt to do so is a hack with extreme limitations.

[[XC] Angstrom]

It's not about performance, it's about the fact that Ageia's physics can be interactive, a GPU can't do that.

The article never mentioned performance.

Exactly, and i'll go for Ageia any day, I have an x4 slot just waiting for fully interactive physics.

[Cobalt]

don't you understand, physics is driven by a CHIP... ati and nvidia are FAR SUPERIOR in chip design than AGEIA, the PPU units are not very different than a shading unit. ATI and Nvidia have years of experience, and there is no doubt then can make hardware that is a lot faster and chepaer than AGEIA

AGEIA's claims are stupid, the chip can process whatever data you feed it, physics won't come for at least 6 months - enough time for ati and nvidia to include each feature required bu devs.
Shows how much you know about chip design. Sure most chips can process most data given the right instructions but that doesn't mean that the chip will be any good at doing it. Why do you think that up untill now we've had seperate vertex and pixel shaders? Or a different type of chip to process graphics and sound? They do it much better than a general purpose processor like a CPU.

ATi and nV have years of experience making graphics chips and now they have turned their considerable resources to physics it wasn't that hard for them to get a good team together; they can just hire whoever they want. Ageia are a small company ATM so its no wonder that they can't produce chips as cheaply as the graphics companies.

[LOE]

shows how much you know about shading units...

CPUs have a single pipeline, but very complex, and can do whatever you can think of... GPUs have more piplines but they are not that complex. A gpu pipe is responsible for simple texturing and thats all, and since there are many of them GPUs do their job far better than a CPU. The same thing goes for physics, it is not that complex to calculate but it requires a lot of calculation due to the fact there are many many objects that interact with physics. But physics has nothing to do with the gpu pipeline... it's all about the shading units... as you know in a GPU there are many shading units, each one capable of executing one instruction per clock. 16 shading units at 600 Mhz can execute 9 600 000 000 instructions per second... thats a LOT

some history

before DX8 the GPU was only texturing and geometry, with the introduction of DX8 and SM1 the GPU was capable of executing some instructions to pixel level. With the introduction of SM3 the lenght of theese instructions was significantly increased, giving the shader unit the ability to process more complex calculations.

SM3 shaders due to their very long instruction can be used for many things outside graphics, a few weeks ago ATI demonstrated sound processing with a GPU, something I was talking about probably 1 year ago. And it's not just sound... feed the shader unit data and instruction what to do, and it will do it.

shading units are simple programable processors, they are the only programable part of the GPU, and it is just a matter of driver support so that object properties go to the shader units, forces applied to them go to the instruction queue and the shading unit calculates the result

it is not as simple as it sounds, but in general thats the way things work

the AGEIA solution is not that different in structure than a gpu, it is more simple and probably more dedicated to physics, but remember microsoft is going to integrate physics in DX, and it will be GPU driven, and ofcoruse GPU optimised, and having in mind SM4 is comming with a lot more headroom you can probably imagine that executing even very complex code will be possible.

Ati and Nvidia are mature chip designers - ageia is a hot shot. The price AGEIA asks for their solution is WAY OVERPRICED, thats cause they tought there will be no competition.

Some MORE history - in DX7 the GPU took the T&L load off from the CPU. Noone introduced a T&L standalone chip that costs 300$. Like T&L, physics should and probably will become a GPU feature. It is stupid to make a different accelerator for every little thing. Since DX8 the GPU is much more than only graphics, and will become even more compex and powerfull in time.

AGEIA's chippery is no match for ati and nvidia, and I am pretty sure the same thing goes with drivers.

What is wrong with you, last week everyone was screaming excited that gpus process physics much faster and much chepaer. Now after an anonymous comment all of a sudden everyone starts "ati and nvidia have nothing to do with processing physics, they simply cannot" I think for claims like this to be true it takes a lot more than anonymous complains...

ALSO how did this developer "tried to implement the physics" then there is NO SDK for GPU physics? Don't developers need a development kit in order to develop? I mean what ATI demonstrated is private to them and havok. You have to know how to do something in order to do it. This was an early demo, how come a dev is using it before even beeing finalized, not to mention the lack of any resources on using it... And how much interactive physics is depends only on the way it is programmed, it has nothing to do with hardware limitations. You can make perfectly good physics even ot a CPU but it won't hadle it. But a fast GPU with many shader units will have no problems with it...

Think about it... :slapass:

PS: sorry for the bad grammar
PS2: do you trust the INQ for everything they say? and if yes since when? the inq should change their header color to YELLOW, I told mike this a long time ago, mostly cause of the BS Demerjian enjoys posting, but it's not just him you know ...

[Cobalt]

Microsofts intergration of physics is a new API not a part of DX (even if they share the "direct" name). The point is that the physx chip has been optimised heavily for physics and a shader unit is optimised heavily for, guess what, shading!

To say they are similar is correct. They are both simple, programable processing units, but the devil is in the details and its the little differences that count. The calcuations a shading units is designed to carry out are different to that a physics processor needs to make. Sure you have basic vector calculations but you need a lot more than that if you want decent physics which is where the problem lies. The GPUs can do the vector bits with ease but the moment you need interaction it slows right down because it need more instructions in order to carry them out. The Physx isn't just a chip: it also has its own API and instruction set to go with it that makes it a lot easier. Maybe it'll all change when DXphysics comes out but untill then we know very little of what performance or quality is going to be like.

And we never said we trusted anything that the INQ said in fact I never read the actual article just the snippet at the top but it doesn't stop what I said being true. Besides you can't take the view that everything the INQ says is automatically a lie.

[LOE]

even if there are some limitations on current hardware, they will be gone in 4-5 months with the SM4 capable GPUS

anyway I don't really see much games that are going to use AGEIA

the ghost recon is just a marketing trick - everyone remembers FARCRY 64bit - better textures than the 32 bit versin - and everyone was like "WOW DUDE" but the very same textures could be used in the 32 bit version as well

the same stori with ghost recon - some stones pop out of the ground when you shoot it.. WOW... and with AGEIA things actually slow down, cause of the extra details the GPU has to render

there is point of arguing about this, sooner or later we will find out, I personaly don't like the idea of giving away 300$ so I can play a game. I don't like the idea that extra hardware should be required for better experience when GPUs have the horse power.

AGEIA is only trying to make money, noone can blame them for this, but they should have been more reasonable with their pricing - the chips costs no more than 50$ - plus PCB and other components hardly will top 70-80$

[Cobalt]

I don't know where you got the idea that I said that it was worth it. I think I said in my first post that it wasn't worth shelling out for but that doesn't stop it from being more suited to the task. Which was my whole point.

Do you think that ATi's costs are any higher? It probably costs them even less to produce because they can get better econmies of scale into action. The cost of a component is based on demand more than actualy manufacturing costs. Ageia predicted there would be high demand and priced their card too high. Then because of the high cost and no one believeing their marketing hype, demand was low and very few people brought one comapered to their predictions. Now in order to recover costs they have to keep the price high so they get a better return per unit.

[ahmad]

There is no need to argue this. Some anon dev says Ageia is more powerful, without actually testing anything. So we can't say he is right, but we can't say he is wrong either.

But if you look at it, the x1900 (for example) has massive calculation power. Obviously its more expensive than Ageia's solution (and I would expect it to be more powerful than the PPU), but if you look at it in the long run - you buy 2 high end video cards: if the game supports physics, one card does physics, the other graphics, and you are a happy camper. Game doesn't support physics? Thats fine, use them both to improve your gaming performance.

From a practical point of view, a standalone Ageia PPU is a waste. Not to mention the enormous price tag with very few benefits atm.

But from what I can see (knock on wood), the multithreaded design of GPUs should be superior to a dedicated processor like ageia's ppu (and most GPUs come with 256MB now).

[Cybercat]

How do you know Ageia's PPU isn't multithreaded? It's a floating point processor like any other, and from what I gather it's similar to the Cell, but obviously very simplified in comparison.

I think there are improvements that are necessary, but personally I'd prefer not to buy a graphics card for physics.

[ahmad]

I think there are improvements that are necessary, but personally I'd prefer not to buy a graphics card for physics.

Ageia is a new company, therefore anything they put out will be much more expensive than something lets say ATi and Nvidia have already done ages ago. You pay $400+ (CDN) for something that is merely a dedicated processor. There is nothing impressive about the hardware it holds right now, its the idea thats bringing them cash: a dedicated PPU.

And I didn't say it wasn't multithreaded, I am saying whatever it is, the hardware is guaranteed to be inferior to an x1800+ or a 7800GTX.

[Cybercat]

Ageia is a new company, therefore anything they put out will be much more expensive than something lets say ATi and Nvidia have already done ages ago. You pay $400+ (CDN) for something that is merely a dedicated processor. There is nothing impressive about the hardware it holds right now, its the idea thats bringing them cash: a dedicated PPU.

And I didn't say it wasn't multithreaded, I am saying whatever it is, the hardware is guaranteed to be inferior to an x1800+ or a 7800GTX.I'll agree it's the idea that's bringing in the cash, but isn't that how it always works anyway? People don't buy a chip because of how many transistors or threads it can handle, they buy it for what it can do. I'll also agree that as a new company, they're having to charge more than it's worth. Their uphill battle opening up this new market is extremely steep.

When you say guaranteed to be inferior, inferior in what exactly? Math calculation? No doubt the R5x0 and G7x are superior (faster) in calculating the type of math that they're made to, but Ageia's chip is simply capable of calculating many more types of math than ATI and NVIDIA. It's more flexible, so in regards to physics, no, it's not inferior, technically speaking.

The complex workarounds required to get SM3.0 cards to run physics creates tremendous overhead, and even still it's not capable of doing as many things (such as interactivity), like the article talked about. Ageia's card does it natively, so the only overhead is inefficiencies in drivers or game implementations, or possibly the PCI bus (a bottleneck).

[ahmad]

When you say guaranteed to be inferior, inferior in what exactly? Math calculation? No doubt the R5x0 and G7x are superior (faster) in calculating the type of math that they're made to, but Ageia's chip is simply capable of calculating many more types of math than ATI and NVIDIA. It's more flexible, so in regards to physics, no, it's not inferior, technically speaking.

The complex workarounds required to get SM3.0 cards to run physics creates tremendous overhead, and even still it's not capable of doing as many things (such as interactivity), like the article talked about. Ageia's card does it natively, so the only overhead is inefficiencies in drivers or game implementations, or possibly the PCI bus (a bottleneck).

Overhead will depend on the complexity of the implementation ATI and Nvidia decide to go with, but it will be very little. Why? Because physics is pure calculations. No matter how optimized the Ageia PPU is for providing these "advanced physics", it all comes down to raw processing and number crunching power.

And even if there is overhead, it will be overcome by the awesome power of GPUs and I highly doubt it will perform slower than a PPU. It was SM3.0 that opened up many possibilities by allowing programmers to easily take advantage of a GPU's processing power. They call it GPGPU now:

http://www.gpgpu.org

And if you look here:

http://www.gpgpu.org/cgi-bin/blosxom.cgi/Scientific%20Computing/index.html

You can see Physics on a GPU is something that was being worked on for sometime now, so I imagine no one is going to be wasting time trying to figure out how its done. All they need to do now is optimize it further and implement more features, and its a done deal (if they haven't already done that).

And look at it like this: if the current CPUs can handle these physics tasks, but they just get bogged down from the bandwidth issues and weak multithreading capabilities, imagine what something 10x more powerful in processing power and is all about multithreading, taking over. To me, its a no brainer. You are going to tell me something that has 125million transistors with 25W of power consumption will outperform something that sucks up 130W of power and holds over 300million transistors? Just does not add up in my head. Think of the PPU as this: an add on CPU+RAM in your PCI or PCI-E slot that will perform special computations to remove a lot of CPU overhead. Can't give it much more than that.

Ageia is pushing hard to keep this on their side of the fence, but ATI and Nvidia won't pass up a chance like this.

[Cybercat]

Hmmm, well, I really don't have anything more to add that I haven't essentially already said. That's your opinion and I'll accept it as such and leave us at odds.