NVIDIA Physics card (rumour... again!)

[nutball]

Saw this over at The Reg, followed the link to Asus and indeed that is what they say...


Taipei, Taiwan, September 20, 2006 – ASUSTeK Computer Inc. (ASUS), the worldwide leader of motherboards, today introduced the P5N32-SLI premium/WiFi-AP, which leveraged Quad-SLI technology, Intel's Core2 Extreme support, and a third PCI Express x16 slot for Nvidia's upcoming Physics card to deliver superior gaming realism. Teamed with SupremeFX and DTS connect, P5N32-SLI Premium/WiFi-AP-powered systems provide gaming experiences that look, feel and sound real.
IThe P5N32-SLI Premium/WiFi-AP is a gamer's dream up true. It is Quad-SLI ready and supports the Intel Core2 Extreme processor, providing superior processing power for the most graphics resource-hungry applications. The motherboard incorporated three PCI Express x16 slots (two for SLI-ready dual-GPU graphics cards and one reserved for Nvidia's Physics card) to enable unparalleled graphics performance and gaming realism.
http://www.reghardware.co.uk/2006/09/21/asus_spills_nvidia_beans/

http://www.asus.com/news_show.aspx?id=4329


I know NVIDIA have hinted at the 3 PCIe slot solution before, but the wording of this makes me wonder whether they're preparing something which is less like a re-used GPU than previously suspected.

[LeStoffer]

Everything seems to suggest that they wont make a dedicted PPU, but just use one of their GPU's for the job:

http://developer.download.nvidia.com/presentations/2006/cedec/2006-cedec-gpu-physics.pdf

Edit: See page 28.

[nutball]

Well that's what I thought was the plan, but why refer to it as "upcoming physics card" then? I mean graphics cards aren't really "upcoming". Maybe it's referring to an "upcoming marketing push" to rebrand graphics cards as physics cards.

Weird.

[Kaotik]

Everything seems to suggest that they wont make a dedicted PPU, but just use one of their GPU's for the job:

http://developer.download.nvidia.com/presentations/2006/cedec/2006-cedec-gpu-physics.pdf

Edit: See page 28.

Even though this should be good and all, do I remember completely wrong when I say that ATI's GPU's are a LOT better suited for physics (and other "GPGPU processing") than current GF's are?
So one should rather take a look on 3 ATI card solution than 3 nVidia card solution?

[tEd]

Even though this should be good and all, do I remember completely wrong when I say that ATI's GPU's are a LOT better suited for physics (and other "GPGPU processing") than current GF's are?
So one should rather take a look on 3 ATI card solution than 3 nVidia card solution?

It may change with next iterations of nvidia gpu's plus the whole physics on gpu's is still in a early state.

[SugarCoat]

early? it doesnt exist yet. All we've seen is paper.

[trinibwoy]

early? it doesnt exist yet. All we've seen is paper.

The lack of consumer products isnt an indication that it doesn't exist. That's probably what he meant by "early". It's obviously being worked on somewhere in some lab by somebody - won't just appear out of thin air.

[SugarCoat]

I know what he ment but there should be extreme caution when talking about something like this. Its not early as in Nvidia SLI early when it launched and had little compatability. Its early in that we've seen no working demonstrations let alone game compatability. Its like talking about those coming AMD/Intel Octoprocessors :).

Its very possible whats been said so far, such as in the case of ATI where any processor with dynamic branching being a capable phyisics card, may become simply words. Both companies may realize its more profitable/economical to release a third product thats more vanilla in capability. For example removing DVI/Video outputs and chips not needed for a physics card. Also arent going to need 512mb of GDDR3/4.


Even based off the very little of what i said i think its reasonable both companies will release dedicated physics boards because they'll be cheaper and thus able to cater to a wider community of gamers rather then relying on the diversity of your mid/high end cards and hoping they get retired as physics cards or that those users will buy new card releases in pairs or triplets.

Thats of course in addition to plans that their higher end parts can infact become viable physics cards, but i think dedicated physics cards from both companies are real realities.

Buy two X2800XTs, one as a physics
or one X2800XT and one stripped X2800 Physics for 2/3 of the price of another XT but with all the physics performance.

See my point? Also gives the buyer the choice of if they do infact want both to be high end vid cards so that they may use CrossFire without physics sometime.

[Kaotik]

early? it doesnt exist yet. All we've seen is paper.

Didn't ATI demo it already with several demos?

[SugarCoat]

not a game demo no, they have had their own inhouse stuff, for example heres an X1600 doing Havok FX with CF R580s...

http://www.thetechzone.com/?m=show&id=557&page=2

But when we start seeing a game demonstration or two then its time to call it early ;)

[_xxx_]

And then we'll have the questions like "Can I mix the ATI physics card with the nV SLI setup?" or vice versa, head explodes already... :sad:

[^M^]

Buy two X2800XTs, one as a physics
or one X2800XT and one stripped X2800 Physics for 2/3 of the price of another XT but with all the physics performance.

See my point? Also gives the buyer the choice of if they do infact want both to be high end vid cards so that they may use CrossFire without physics sometime.

Yes but the goal is also to reuse a old videocard.
You buy a 2600 Pro and when you upgrade to 2900XT, you use your old card for physic.

[Sunrise]

German site golem.de (http://www.golem.de/0609/47972.html) did an interview with Michael Hara, VP of NV. I did a translation of the related excerpt:

It won´t be a dedicated physics card. It´s rather something we´ll slowly grow into. With the next generation Geforce and it´s successors, the prospects for physics calculations are looking very good.
So that should definately answer any questions.

NV will therefore find themselves in the position to sell even more cards based on their ASICs, which continues their strategic plans they have been following throughout the years.

[Geo]

Well, I think he pretty much just promised much better (rather than just moderately better) dynamic branching in G80, didn't he?

[Razor1]

thats what it sounds like, but they mentioned unified again?

http://translate.google.com/translate?hl=en&sl=de&u=http://www.computerbase.de/&sa=X&oi=translate&resnum=1&ct=result&prev=/search%3Fq%3Dcomputerbase.de%26hl%3Den%26lr%3D

same article

[Geo]

Shazbat! :shock:

Did Hara just confirm ps/vs is unified in G80?

Michael Hara continued to explain that nVidia about it think, the new „Unified Shader “- architecture, in which pixel and Vertex Shader are summarized, to use also for physics computations. By the unification of these two computation units it is very simple to compute also physics within the GPU. With nVidia one believes firmly in this concept and sees also opposite processors, which will have four cores, an achievement advantage.

[Razor1]

hmm I don't know the translation is quite bad but it does sound like that. Where are you german speaking people! ;)

[Sunrise]

hmm I don't know the translation is quite bad but it does sound like that. Where are you german speaking people! ;)
It definitely sounds like a confirmation that PS and VS are indeed unified, because otherwise it wouldn´t have made a lot of sense for him to outline their strengths regarding possible physics calculations, which are done in the GPU itself.

Next week they´ll post the full interview.

[Razor1]

thx Sunrise:smile:

[psurge]

So the original quote is this


Im Interview mit Golem.de sagte Nvidia-Vizepräsident und Kommunikations-Chef Michael Hara: "Das wird keine eigene Physik-Karte sein. Vielmehr ist es etwas, in das wir langsam hineinwachsen. Mit der nächsten GeForce-Generation und den dann folgenden werden die Möglichkeiten für Physik-Berechnungen sehr gut sein." Hara spielte damit auf den bisher nur unter dem Codenamen "G80" bekannten GeForce-Chip der achten Generation an, der nach unbestätigten Angaben noch 2006 vorgestellt werden soll.

Wie Hara weiter erklärte, denkt Nvidia daran, die "Unified Shaders" der neuen GPUs auch für Physik zu nutzen. Durch die Zusammenführung von Pixel- und Vertex-Shadern sei es sehr einfach, auch die physikalischen Eigenschaften eines Objekts gleich im Grafikprozessor zu berechnen. ...


Translates to

In an interview with Golem.de, NVidia VP and Communications Driector Michael Hara stated that :
"... the prospects for physics computations in the next generation and next-next generation GeForce will be very good" , referencing the 8th generation GeForce chip codenamed "G80", which will be released in 2006 according to unconfirmed rumors.

As Hara further explained, NVidia is planning to use the "Unified Shaders" of the new GPUs for physics computations as well. The merging of pixel and vertex shaders apparently makes the additional computation of an objects physical properties on the GPU very easy. ...


So I guess you could argue that it might mean that G90 has unified shaders (not necessarily G80). Could a native speaker confirm?

Edit: they also say that they will post the full interview in the "coming weeks", not necessarily next week.

[_xxx_]

hmm I don't know the translation is quite bad but it does sound like that. Where are you german speaking people! ;)

Ever heard of the weekend, fresh air, not spending my life in front of the screen? :razz:

Give me a few mins.

[Sunrise]

Give me a few mins.
You don´t need to, i´ve already taken care of it. :wink:

Enjoy the weekend.

[_xxx_]

So here we go...

As we're nearing the release day of a new product, the more rumours are making the rounds. This goes especially for GFX-chips, in this case nVidia's upcoming G80 chip. It will probably be capable of physics calculations as well, thanks to it's completely reengineered architecture.

This was announced by nVidia's PR-chief Michael O'Hara in one of the upcoming interviews by the colleagues from Golem.de and helped to clarify (some of) the rumours which made their rounds in the last couple of days. The initiator was a press release from the mainboard and GFX-card manufacturer ASUS about their „P5N32-SLI Premium“ mainboard.

The mainboard features the „nForce 590“ chipset for Intel-platforms and offers three PCI-Express 16x slots for GFX-cards. According to the PR the thirs slot should host an nVidia physics card. This statement yielded rumors about nVidia developing their own physics accelerator card. This seems to be a misuderstanding, as M. O'Hara explained to Golem.de.

So he said that nV does not develop their own physics card, but works on something else at the moment in which they have to slowly grow into. Further he said that with the next gen and the one after "the possibilities for physics calculations will be very good". This would be a reference to the upcoming G80 chip which should probably offer support for physics calculations.

Further he explained that nVidia thinks about using their new architecture in which the pixel and vertex shaders will be unified for physics calculations as well. Due to the unification of these two processing units, it's very simple to calculate physics within the GPU. NV firmly believes in this concept and is confident that they'll even outperform the upcoming quad-core CPU's.

[_xxx_]

You don´t need to, i´ve already taken care of it. :wink:

Enjoy the weekend.

Ooops, missed that! Just plugged in. Oh well... :oops:

[_xxx_]

That's what you get by being a nice boy and reading the thread in the chronological order :lol:

[R300King!]

As we're nearing the release day of a new product, the more rumours are making the rounds. This goes especially for GFX-chips, in this case nVidia's upcoming G80 chip. It will probably be capable of physics calculations as well, thanks to it's completely reengineered architecture.

This was announced by nVidia's PR-chief Michael O'Hara in one of the upcoming interviews by the colleagues from Golem.de and helped to clarify (some of) the rumours which made their rounds in the last couple of days. The initiator was a press release from the mainboard and GFX-card manufacturer ASUS about their „P5N32-SLI Premium“ mainboard.

The mainboard features the „nForce 590“ chipset for Intel-platforms and offers three PCI-Express 16x slots for GFX-cards. According to the PR the thirs slot should host an nVidia physics card. This statement yielded rumors about nVidia developing their own physics accelerator card. This seems to be a misuderstanding, as M. O'Hara explained to Golem.de.

So he said that nV does not develop their own physics card, but works on something else at the moment in which they have to slowly grow into. Further he said that with the next gen and the one after "the possibilities for physics calculations will be very good". This would be a reference to the upcoming G80 chip which should probably offer support for physics calculations.

Further he explained that nVidia thinks about using their new architecture in which the pixel and vertex shaders will be unified for physics calculations as well. Due to the unification of these two processing units, it's very simple to calculate physics within the GPU. NV firmly believes in this concept and is confident that they'll even outperform the upcoming quad-core CPU's.

This seems very unclear to me. (the part in bold)

It says NV thinks about using their "new architecture" (which could mean ANYTHING, not neccissarily it being unified) in which the PS and VS will be unified for physics calculation as well.

Is it possible they can unify PS/VS only when it comes to the physics calculating part but yet not have unified PS/VS use for normal graphic card calculations?

In this bold text he is talking about physics ONLY throughout. This is my thought. Plausable?

[Razor1]

This seems very unclear to me. (the part in bold)

It says NV thinks about using their "new architecture" (which could mean ANYTHING, not neccissarily it being unified) in which the PS and VS will be unified for physics calculation as well.

Is it possible they can unify PS/VS only when it comes to the physics calculating part but yet not have unified PS/VS use for normal graphic card calculations?

In this bold text he is talking about physics ONLY throughout. This is my thought. Plausable?



Hmm don't think thats plausable, if its unified its unified, there would be no reason for un unification during graphics works. Well unless the schedular overhead is high?

[CarstenB]

Any attempt at running with 3 graphics cards will require new chipset and motherboard support with 3 x16 slots. Even if they are not all fully x16 capable. I've yet to see such a configuration anywhere.

[INKster]

Any attempt at running with 3 graphics cards will require new chipset and motherboard support with 3 x16 slots. Even if they are not all fully x16 capable. I've yet to see such a configuration anywhere.

Oh, you mean like this ?
http://www.asus.com/news_show.aspx?id=4329

[R300King!]

Hmm don't think thats plausable, if its unified its unified, there would be no reason for un unification during graphics works. Well unless the schedular overhead is high?

I understand your point, however, I don't mean un unifying for gfx support. I mean only unifying for physics support. Maybe the genericness of the physics proccessing would somehow be "unified"..not caring if your using PS or VS units. Possible?

[Razor1]

I understand your point, however, I don't mean un unifying for gfx support. I mean only unifying for physics support. Maybe the genericness of the physics proccessing would somehow be "unified"..not caring if your using PS or VS units. Possible?


Hmm interesting I think it would be possible as long as vertex shaders can get access to the pixel shader data after its does the calculations which is something that is supposed to be in the g80

[Rys]

Oh, you mean like this ?
http://www.asus.com/news_show.aspx?id=4329
And let's not forget the P5W64 WS Pro, with 4 x16 slots

[INKster]

And let's not forget the P5W64 WS Pro, with 4 x16 slots

Well, if you're talking about having 4 *physical* (pun not intended) PCIe x16 slots, then Gigabyte got there first a while back:
http://www.gigabyte.com.tw/Products/motherboard/Products_Overview.aspx?ProductID=1957&ProductName=GA-8N-SLI+Quad+Royal

Too bad it's for P4/PD only...

[CarstenB]

Oh, you mean like this ?
http://www.asus.com/news_show.aspx?id=4329

Doh, watch at my donkey ears grow longer.

Does the PCIe interface deliver separate powerlines of up to 75W for each of these slots?

[Kanyamagufa]

Doh, watch at my donkey ears grow longer.

Does the PCIe interface deliver separate powerlines of up to 75W for each of these slots?

Something tells me that if you're going to be running any combination of three next gen cards on one board, a few molex connectors will be in order. :wink:

[lopri]

Something tells me that if you're going to be running any combination of three next gen cards on one board, a few of this (http://www.beyond3d.com/forum/showthread.php?t=33984) will be in order. :wink:
Fixed.

[aeryon]

from dailytech http://www.dailytech.com/article.aspx?newsid=4444

Documents Leak NVIDIA's Quantum Physics Engine
Kristopher Kubicki - October 5, 2006 1:21 AM

NVIDIA is ready to counter the Triple Play

With the release of the G80, NVIDIA will also release a new engine dubbed Quantum physics engine. Quantum Effects Technology is similar (at least in spirit) to NVIDIA's PureVideo Technology -- a dedicated layer on the GPU for physics calculations. A few documents alluding to this new engine appeared on public FTP mirrors late last week.

Quantum utilizes some of the shaders from NVIDIA's G80 processor specifically for physics calculations. Physics calculations on GPUs are nothing new; ATI totes similar technology for its Stream Computing initiative and for the Triple Play physics.

NVIDIA and Havok partnered up this year claiming that SLI systems would get massive performance gains by utilizing additional GeForce GPUs as physics processors. Quantum may be the fruits of that partnership, though NVIDIA documentation clearly states that Quantum will work just fine without SLI.

NVIDIA's documentation claims Quantum will specifically compete with AGEIA's PhysX, yet does not mention who is providing the middleware. Given that there are only two acts in town right now, it would be safe to say Havok has a hand in the Quantum engine.

[Geo]

Yeah, I saw that piece. The whole "some of the shaders" language bothered me tho. Possibly it was just an inelegant turn of phrase on Kristopher's part, rather than an indication of special function physics units in G80.