PC: ATi Physics

Foe · 05-June-06, 11:05 PM

Quote:


	ATI set to drop Physics bomb Computex 2006 After the Nvidians, here come the Reds AFTER A LACKLUSTRE DEBUT from Ageia, there are many people who are scratching their heads, both in the world of journos and developers who feel they were left in a state of awe. That is because the PhysiX PPU actually ended up in slowing games down and showing some not-so-fancy graphics effects, the testers allege. But, every start is a hard one. After being a no-show on CeBIT and E3, ATI is returning to the wonderful world of trade shows, doing it in style with the presentation of their own "GPU can be GP GPU and do more stuff" concept. Our sources claim that the concept won't be tied into the "1GPU does GPU-stuff, 1GPU does PPU-stuff", but rather sharing a dynamic load. RingBus memory controller is of course, the guilty party for making Physics & GPU sharing a reality. Even better, it will probably not be required to use two same spanking cards at the same time - shelling out 500 dollars for a GPU and another 500 for the "GPU with identity crisis" is really not a great idea. ATI's approach is set at raising the number of sold X1600 boards and getting rid of old R520 stock. Of course, there'll be much more info from ATI when the NDA lifts and Nvidia's non-flexible SLI Physics get some competition. One thing is certain: competition is a really good thing to have. On the other hand, we have been warned by the developers that many games could end up with no physics support at all, since the publishers could decide against the expenses required in order for implementation of the feature. Someone needs to pay for those extra man-hours... µ

News Source: The Inquirer

Graaille · 06-June-06, 10:54 AM

Sooooo........:huh:

If I'm reading correctly, Ati is moving towards a double chip design on their video cards, one side going toward the actual graphics, the other going toward physics calculations? Seems to me to be a better solution than getting a second card.

Having said that: three things are causing me to wonder.

1) Heat issues w/two graphics processors on one card. Most high end cards right now need quite a bit of cooling, both passive and active to maintain performance.

2) Size. Current video cards, whether pci16 or agp are getting longer and longer, not to mention thicker w/cooling solutions (see #1), are we going to see the re-emergance of the double length cards that were prevalent in the late 80's for specialized functions? There's so much crammed on a m/board as it is w/out trying to jam an even longer video card on there.

3) Crossfire/SLI: How will this work for a dual vid card setup, or will those users be out in the cold? Somehow I doubt it, but it would seem to me to be a bit of overkill to do 2 of these newer tech cards if they each had a physics subsystem incorporated. Maybe it would work better, maybe that's the only way to have the extra physics content in a SLI/Xfire format, I don't know.

Interesting things coming down the pike for '07

droeblek9 · 06-June-06, 12:06 PM

I'm definitely interested in seeing / reading about ATI's newest offering when it's finally unveiled.
Foe, you make some excellent points, and time will tell. However, there's one aspect that about this that really makes me happy: integration into one card. If this works, and doesn't present the problems that you've listed, this is going to be a very good thing. Heck, even if it doesn't work, the simple fact that manufacturers are attempting to integrate these functions into one package leaves all us consumers on the winning end, IMO.

Graaille · 06-June-06, 03:58 PM

Gee, droe and thrasher seem to be drinking from the same glass......

NCXCStud · 06-June-06, 04:09 PM

I think they are Graaille.

Those are good points GRAAILLE

The other one I'd like to know is how much will it cost. A 'top of the line' videocard ranges between 400-600 bucks...the ageia phsyx card is what? 250?

So, does this mean we'll see a card hit the 800 mark this time? I really, really don't want to see that.

Foe · 06-June-06, 07:07 PM

Here is more information to fill in the gaps.

Read More

Quote:


	ATI had a system based on their upcoming RD600 chipset with Radeon X1900 Crossfire and Conroe CPU - they’re calling it “True-to-Life Gaming Reality”. ATI’s method of hardware physics is different to the AGEIA PhysX method in that ATI are able to program a GPU (such as the Radeon X1600) to handle the physics tasks, rather than having to produce a standalone physics processing card.

Graaille · 06-June-06, 09:43 PM

Quote:

Originally Posted by Tweaktown.com


	After taking a closer look at the back of the system, we discovered Crossfire was not actually operational. While the system does indeed have two graphics cards installed (which act as actual GPU’s), Crossfire was not enabled. The top X1900 card was connected to the X1600 rather than the other X1900 card below it - sneaky ATI!! As this is the case, it would seem like ATI still might have a little bit of work to do before their version of hardware physics is ready for the consumer. We also noticed some type of adapter on the X1600 (that explains why the card sticks our further than the rest) which probably goes from a PCI Express 16x slot to PCI Express 1x or whatever the motherboard was using - we couldn’t see.

Ok, after looking at the pictures, and looking at pictures of currently available Xfire ready m/boards, I got just one thing to say:
How in the world did they get 2 X1900X's and one X1600 on a board that only supports two pci-e 16 slots? Especially since the X1600 was sandwiched between the two X1900X's.

Yes, an adapter was involved, it had to be involved, but it would seem really flippin' difficult to work a pci-e 16 adapter into that small slot in between the pci-e 16 slots on your typical Xfire m/board, then put a full height video card in that adapter and have it physically stable -- because you couldn't have any ability to screw it down to the frame (unless they are providing the frame standoff I saw in the video in with the card).

And even though they had two X1900X's in the system shown, Xfire wasn't enabled, so . . . . . what's the point of having 2 $500 video cards in the system? They are still using a stand alone card (in this case a X1600) to do the physics processing, which just means they dropped in a seperate card that does the same thing as the Agiea card, but beats the Agiea card by a cool $100 USD. Don't get me wrong, saving $100 on a card is a good thing, but unless they can get this to work w/Xfire, it would seem to me that going w/Agiea's offering that fits in the pci-e 1 slot would be a better way to go. (God I hate saying that.)

Yep, a lot of further work/R&D is gonna have to go on before this gets off of the ground, especially as tight in there as the pictures show. Step in the right direction? Yep, no doubt. But I don't see this even hitting the pre-release reviewers until 1Q, 2007 because of moderate to signifigant card restructuring -- public release no sooner than 2Q, 2007.

droeblek9 · 07-June-06, 02:17 AM

Quote:

Originally Posted by Graaille


	Gee, droe and thrasher seem to be drinking from the same glass......

LMAO...whoopsie...i blame the meds

Graaille · 07-June-06, 08:30 AM

Article from The Register:

Quote:

Originally Posted by The Register


	Computex 2006 ATI will use GPU-hosted physics processing to "overtake" arch-rival Nvidia's SLI technology "by this time next year", Godfrey Cheng, the company's head of platform marketing said today during the Computex show in Taipei. Like Nvidia, ATI is looking to games middleware developer Havok to connect game code to its GPUs. The Havok FX API runs physics-based special-effects algorithms through pixel shader pipelines rather than the host CPU. ATI's pitch is that its graphics cards provide a better foundation for this technique than its rival's products do. Indeed, the company presented a chart to journalists showing its GPUs massively out-performing Nvidia's when running object-collision detection routines - though it later admitted the chart was actually a forecast of the performance boost it expects its chips to provide rather than measurements it has actually taken. ATI is pushing its Radeon X1600 as its physics-dedicated GPU of choice, a part it claims delivers twice the performance of Ageia's dedicated PPU chip - well, based on "estimates" derived from sphere-to-sphere collisions per second numbers made public by Ageia. The X1600's half the price of Ageia-based physics boards, ATI claimed. Games with physics processor support are expected to ship late this year or early 2007. By then, ATI expects to have drivers available that will allow a graphics card to run as a physics card, whether that's one GPU alongside a pair of CrossFire cards that handle the imaging, or two GPUs, one for graphics, the other for physics. Nvidia announced it was working with Havok in March. Keen not to be touting a "me too" offering, Cheng said ATI has been working with Havok for more than seven months. The company has said nothing about it until now, he added, because it didn't want to make an announcement without having anything to demo at the same time. Demo it did, and the acceleration of object collision and particle effects was impressive. But then, other such demos have been too. By the end of the year, Cheng said, Havok expects to have realistic liquids, cloth and even hair effects made possible through its API and GPU acceleration. In the same timeframe, Cheng said he expects motherboards based on ATI's RD600 chipset to be shipping with room for three graphics cards.

So ATI is expecting, projecting, and flexing . . . . but is it producing? Paragraph 2:

Quote:


	Like Nvidia, ATI is looking to games middleware developer Havok to connect game code to its GPUs. The Havok FX API runs physics-based special-effects algorithms through pixel shader pipelines rather than the host CPU. ATI's pitch is that its graphics cards provide a better foundation for this technique than its rival's products do.

Yep, makes perfect sense. 12-16 parallel pixel pipelines would allow for more and potentially better physics computations to be performed in real time w/a game, and would have less of a performance impact, plus it would (in theory) be scalable so when they do 24-48 pipelines, the physics eyecandy will be essentially seamless.

However, all of this is right now quite a bit of speculation/vapourware - as evidenced by paragraphs 3 and 4. Forecasts, expectation, estimations, and claims . . . but no real production.
But in short, they are planning to convert a secondary line of video cards into physics cards, and then make m/board manufacturers alter the boards to allow for either two cards or three cards. Not the innovation that I was hoping for.

Foe · 07-June-06, 08:35 AM

Good find Graaille!

Here is somemore news from Gamgurumania:

Quote:

AGEIA Responds To ATI's Physics Hardware Solution - tech
(hx) 05:55 AM EDT - Jun,07 2006 - Post a comment / read (1)
After Ageia's PhysX chip and Nvidia's Havok FX support announcement, ATI now follows suit with its own proclamation of support for accelerated gaming physics. The chaps over at FiringSquad decided to contact AGEIA to get their side on things. Here is AGEIA's response via their vice president of marketing Michael Steele:

We're glad that they've validated physics as the next big thing in gaming
The performance claims appear to be based solely on gigaflops of the Radeon chips and an assumption of PhysX gigaflops, but that is not a meaningful way to measure physics performance. That's like suggesting that the more wheels I have on my car, the faster I will go. Physics requires much more than raw gigaflops.
Graphics processors are designed for graphics. Physics is an entirely different environment. Why would you sacrifice graphics performance for questionable physics? You'll be hard pressed to find game developers who don't want to use all the graphics power they can get, thus leaving very little for anything else in that chip.
"Boundless Gaming" is actually enabled by AGEIA's Gaming Power Triangle in which the PhysX processor adds true physics to the mix instead of leaving it to a repurposed graphics processor.
In the end, what matters is who develops software for the product. There are games shipping for PhysX today and more than 20 announced for 2006. In addition, over 65 developers of more than 100 games are deep in development for PhysX. No specific games are announced even in development mode for ATI.
PhysX is here now

In related news, AGEIA also sent over a link to a CNet News.com blog which is also critical of the ATI solution:

- ATI's Radeon X1000-series 3D cards are DirectX 9 parts. By the time any games are out with Havok FX features, Windows Vista, and with it DirectX 10, will either have arrived or will be arriving imminently, rendering the current generation of Radeons obsolete. We're sure ATI will include Havok FX acceleration in its next-gen cards, but you'd be foolish to buy the Radeon X1900 XT card ATI is highlighting now in anticipation of future boosted game physics.

- ATI's two-card physics acceleration presents the same problem as Nvidia's. By using a 3D card for physics acceleration, you sacrifice 3D performance for physics performance. Yes, there's lots of new technology on the horizon promising faster computing, but if you've played the Elder Scrolls IV: Oblivion, you know that it chokes even the most advanced current-gen hardware. Smooth frame rates are still more important than more flying bricks. As for the three-card option, ostensibly you can plug in an older Radeon alongside a two-card Radeon CrossFire setup to ramp up the physics, but we'll withhold judgment until we know more about pricing of the special three 3D slot motherboards.

Graaille · 07-June-06, 09:25 AM

So Ageia is saying "We're here now, we're working out the kinks already, by default we're the benchmark that everyone else is using to compare themselves to. We have support for X# of games already, Yes it's a stand alone card that does nothing other than physics -- not a retooling of a graphics card, and our toy will be upgradable when MS forces everyone off into a Vista."

Ati is saying "We're up and coming, we've got the ability to retool our experience w/video cards into a multiple parallel pipeline physics card that should integrate better into the graphics subsystem, and we have the market share to not only 'encourage' game makers to incorporate this into their code, but to 'encourage' m/board manufacturers to rethink/retool their m/board design to incorporate the one videocard/one vidphysics card or 2 Xfire videocards/1vidphysics card that we're developing/pushing now."

NVidia is saying "Yeah, we're doing our thing, don't peek."

New technology -- ain't life grand?

Nerdz · 09-June-06, 03:48 PM

Thats pretty cool. We knew it would happen, so whats nVidia going to do about this? Im pretty sure both companies will go to Ageia, but they might have two different Ideas. The FPS loss in small in some games (3-5fps) but in other games its as large as 10% loss! Its either the software, or something is bottlenecked by the card. But how does a PPU cause loss in a game?

I think its because of latency. If the PPU has to take every frame, analize it and do the physics stuff, thats loss right there. SO theoratically, they have the make the PPU Faster than the GPU itself and be able to Pre-Process each frame.

In the future I can see the PPU being inside the GPU itself, rather than a seperate chip on the card.

Oh about the heat, it doesnt generate that much by the looks of the heatsink on the original card.

Thread Tools
Show Printable Version Email this Page
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode

06-June-06, 10:54 AM		#2 (permalink)
Graaille Apex Master Tech	Sooooo........:huh: If I'm reading correctly, Ati is moving towards a double chip design on their video cards, one side going toward the actual graphics, the other going toward physics calculations? Seems to me to be a better solution than getting a second card. Having said that: three things are causing me to wonder. 1) Heat issues w/two graphics processors on one card. Most high end cards right now need quite a bit of cooling, both passive and active to maintain performance. 2) Size. Current video cards, whether pci16 or agp are getting longer and longer, not to mention thicker w/cooling solutions (see #1), are we going to see the re-emergance of the double length cards that were prevalent in the late 80's for specialized functions? There's so much crammed on a m/board as it is w/out trying to jam an even longer video card on there. 3) Crossfire/SLI: How will this work for a dual vid card setup, or will those users be out in the cold? Somehow I doubt it, but it would seem to me to be a bit of overkill to do 2 of these newer tech cards if they each had a physics subsystem incorporated. Maybe it would work better, maybe that's the only way to have the extra physics content in a SLI/Xfire format, I don't know. Interesting things coming down the pike for '07

06-June-06, 12:06 PM		#3 (permalink)
droeblek9 Warcraft Warrior	I'm definitely interested in seeing / reading about ATI's newest offering when it's finally unveiled. Foe, you make some excellent points, and time will tell. However, there's one aspect that about this that really makes me happy: integration into one card. If this works, and doesn't present the problems that you've listed, this is going to be a very good thing. Heck, even if it doesn't work, the simple fact that manufacturers are attempting to integrate these functions into one package leaves all us consumers on the winning end, IMO.

06-June-06, 03:58 PM		#4 (permalink)
Graaille Apex Master Tech	Gee, droe and thrasher seem to be drinking from the same glass......

06-June-06, 04:09 PM		#5 (permalink)
NCXCStud Apex Tech Fanatic Supreme	I think they are Graaille. Those are good points GRAAILLE The other one I'd like to know is how much will it cost. A 'top of the line' videocard ranges between 400-600 bucks...the ageia phsyx card is what? 250? So, does this mean we'll see a card hit the 800 mark this time? I really, really don't want to see that.

07-June-06, 09:25 AM		#11 (permalink)
Graaille Apex Master Tech	So Ageia is saying "We're here now, we're working out the kinks already, by default we're the benchmark that everyone else is using to compare themselves to. We have support for X# of games already, Yes it's a stand alone card that does nothing other than physics -- not a retooling of a graphics card, and our toy will be upgradable when MS forces everyone off into a Vista." Ati is saying "We're up and coming, we've got the ability to retool our experience w/video cards into a multiple parallel pipeline physics card that should integrate better into the graphics subsystem, and we have the market share to not only 'encourage' game makers to incorporate this into their code, but to 'encourage' m/board manufacturers to rethink/retool their m/board design to incorporate the one videocard/one vidphysics card or 2 Xfire videocards/1vidphysics card that we're developing/pushing now." NVidia is saying "Yeah, we're doing our thing, don't peek." New technology -- ain't life grand?

09-June-06, 03:48 PM		#12 (permalink)
Nerdz Sir Knight of Spamalot	Thats pretty cool. We knew it would happen, so whats nVidia going to do about this? Im pretty sure both companies will go to Ageia, but they might have two different Ideas. The FPS loss in small in some games (3-5fps) but in other games its as large as 10% loss! Its either the software, or something is bottlenecked by the card. But how does a PPU cause loss in a game? I think its because of latency. If the PPU has to take every frame, analize it and do the physics stuff, thats loss right there. SO theoratically, they have the make the PPU Faster than the GPU itself and be able to Pre-Process each frame. In the future I can see the PPU being inside the GPU itself, rather than a seperate chip on the card. Oh about the heat, it doesnt generate that much by the looks of the heatsink on the original card.