DailyTech - AMD Asymmetric Physics Processing Requirements Revealed

Submit News

Hardware AMD Asymmetric Physics Processing Requirements Revealed
Anh Tuan Huynh (Blog) - January 9, 2007 12:18 PM

41 comment(s) - last by JNo.. on Jan 10 at 10:34 AM

Slide courtesy of Chile Hardware

Triple Play physics just around the corner

More information about ATI's long-awaited physics processing is finally trickling into light. ATI has named its GPU accelerated physics processing Asymmetric Physics Processing.

Hardware requirements for Asymmetric Physics Processing include two AMD ATI Radeon based graphics cards and a compatible CrossFire motherboard. One Radeon may be used as a GPU with an additional Radeon acting as a physics processor. The requirement for three Radeons to complete physics processing is only a requirement of Triple Play operation.

ATI’s Triple Play physics processing is also supported with two ATI Radeon graphics cards operating in CrossFire with an equal or lesser performing third card dedicated to physics processing.

Chile Hardware has the full presentation for Asymmetric Physics Processing in English and Spanish.

Asymmetric Physics Processing requires two AMD ATI Radeon graphics cards for baseline processing, the two graphics cards do not need a direct physical link like CrossFire. AMD is also touting the ability to mix and match different ATI Radeon graphics cards to have greater flexibility for users that upgrade their older graphics cards.

At this time it seems AMD chipsets are the only supported platform for Asymmetric physics processing. Supported chipsets include the AMD 580X, ATI CrossFire Xpress 3200 (for Intel) and the upcoming RD790. It is unknown if the CrossFire capable Intel 975X and P965 chipsets will be supported.

Comments

Threshold

Username
Password
remember me

This article is over a month old, voting and posting comments is disabled

spelling...

By slashbinslashbash on 1/9/2007 12:20:48 PM , Rating: 2

Spell-check at least the headline please! Or are you trying to make an editorial statement with the "assymetric" misspelling?

RE: spelling...

By bldckstark on 1/9/2007 12:47:06 PM , Rating: 2

What an assy9 post.

Parent

RE: spelling...

By SunAngel on 1/9/07, Rating: 0

RE: spelling...

By ahodge on 1/9/2007 1:14:19 PM , Rating: 5

LOL, dude...
Dailytech is supposed to be a reputable news site. Professional journalists should be able to spell and write correctly in whatever language they report in. It's unprofessional that Dailytech posts storys with glaring spelling and grammar mistakes. At first, I used to try to help by posting grammar and spelling corrections. I've given up now. It's obviously not going to change. That's the only reason people should stop commenting on the spelling and grammar. Not because it pisses off the staff. The staff should take some pride in their craft. It's not our fault that they chose a profession in writing and yet they have difficulty with basic spelling and grammar.
-Alex

Parent

RE: spelling...

By PAPutzback on 1/9/07, Rating: -1

RE: spelling...

By Spoelie on 1/9/2007 1:40:04 PM , Rating: 3

Apparantly so aren't you, it's spelled 1337!!!

Parent

RE: spelling...

By therealnickdanger on 1/9/07, Rating: 0

RE: spelling...

By TheMaster on 1/9/2007 4:27:33 PM , Rating: 2

It's stories , genius.

Parent

RE: spelling...

By phusg on 1/9/2007 1:17:48 PM , Rating: 3

It's you're (= you are) pissing the staff off.

Parent

RE: spelling...

By masher2

(blog) on 1/9/2007 1:52:50 PM , Rating: 5

To avoid splitting the infinitive, it should be "you're pissing off the staff, not "you're pissing the staff off". :)

Parent

RE: spelling...

By mark2ft on 1/9/2007 10:15:26 PM , Rating: 2

An infinitive is anything that has "to" + verb, like "to go" or "to eat."

Pissing off <-- this isn't an infinitive.

Either way, splitting an infinitive is a matter of taste. Why do you think the greatest show on Earth, Star Trek, expressed their sentiments thus: "To boldly go" instead of "to go boldly"?

Parent

RE: spelling...

By JNo on 1/10/2007 10:34:07 AM , Rating: 2

Yeah I had to log in just to annoyingly correct masher2 too.

If you're going to correct someone on a split infinitive masher2, it would help if you knew what one was first.

"to annoyingly correct" is a split infinitive.

There's no infinitive in "you're pissing off the staff" or "you're pissing the staff off"

Parent

RE: spelling...

By obeseotron on 1/9/2007 2:09:06 PM , Rating: 2

Spell check isn't rocket science, "assymetric" isn't a word. Little grammatical errors are excusable, but not doing a basic spell check is not.

Parent

RE: spelling...

By BladeVenom on 1/9/2007 8:41:40 PM , Rating: 2

Since it's a tech site, they should be using some super elite spell checker of death that can teabag all the other noob spell checkers out there.

Parent

RE: spelling...

By bfonnes on 1/9/2007 4:34:28 PM , Rating: 2

Have you read any of the comments on other articles? I'd say the authors of dailytech spell better than probably 60% of the people that post? lol, come on.

Parent

RE: spelling...

By bfonnes on 1/9/2007 4:37:06 PM , Rating: 2

If all we did was comment on all the posters spellings then there would be no room for actual comments about the articles. sheeeeeessh

Parent

RE: spelling...

By ElJefe69 on 1/10/2007 12:49:50 AM , Rating: 1

people do not deserve to be in any sort of public journalism if they cannot at least SPELL.

grammar is as shitty as it can

Parent

RE: spelling...

By sh3rules on 1/10/2007 3:08:25 AM , Rating: 2

Maybe their spell checker needs an update? Now seriously, perhaps they couldn’t be bothered with spell checking because they think that it’s important to get the news out there ASAP. I have experienced in schools the sad reality of growing illiteracy, so for me this is just a sign of the times. I just take DT for what it is, and I’m grateful for the news.

Parent

RE: spelling...

By KristopherKubicki

(blog) on 1/10/2007 3:56:46 AM , Rating: 3

We have two writers at CES, two editors at CES and another writer at Macworld. The article was published at 12:19 and the correction was made at 12:24. I'm not here to make excuses, but I imagined the content of the article would have a larger splash than typo that was on the front page for less than 5 minutes.

Parent

RE: spelling...

By nurbsenvi on 1/10/2007 10:33:06 AM , Rating: 2

Every time I see a spelling-Nazi I feel sorry for them...
Because after being critical at someone else's spelling mistake he or she will waste time and effort to post Perfect reply on a FORUM so that they won't look like an idiot.

Parent

Dual slot?

By Alphafox78 on 1/9/2007 12:25:44 PM , Rating: 3

Those cards all look like dual slot cards in the pic. if you have 3x of them, thats like 6 bays! hope you dont want to have any other cards in your system!

RE: Dual slot?

By MrBungle123 on 1/9/2007 12:33:10 PM , Rating: 2

Yeah it's a little rediculous isn't it?

I personally think that the PhysX approach is more elegant anyway. Not to mention that a $200 PhysX card is a lot cheaper than a 2nd or 3rd high end video card.

I've also heard that some physics functions can't be run on video cards because everything has to be converted to Direct X 9 and back again through software to make it work.

Parent

RE: Dual slot?

By PAPutzback on 1/9/2007 1:16:12 PM , Rating: 2

The third card could be old X1300 you bought or picked up cheap when someone upgraded their system. 6 bays. Now us Techie fellas know there are plenty of ATI single slot cards out there. They needed space to put the word Physics and Rendering. There is always a product griefer in every discussion.

Parent

RE: Dual slot?

By decapitator666 on 1/9/2007 2:38:33 PM , Rating: 2

A x1600 class graphicscard has more that enough calculating power to start with.. besides that.. buying a x1900 class card will put your electricity bill through the roof

Parent

RE: Dual slot?

By ElJefe69 on 1/10/2007 12:53:07 AM , Rating: 2

needing a 500 watt psu is rediculous.

what utter shit amd/ati come up with as "solutions".

I couldnt imagine more than 1% of gamers really want to get a crossfire system and spend 500+ minimum on video cards.

Parent

RE: Dual slot?

By kilkennycat on 1/9/2007 2:59:01 PM , Rating: 2

With regard to physics/parallel-number-crunching on a GPU, I refer you to Anandtech's article on the 8800GPU. Far more factual and informative than all the marketing hype and technical misunderstanding circulating at the moment.

http://www.anandtech.com/video/showdoc.aspx?i=2870

BTW, physics computations have nothing to do with Dx9c. The misunderstanding seems to come from some warble that Microsoft has made about a Physics API being incorporated into future updates to DX10. However, both Havok and Ageia market third-party physics software Development toolkits for both WinXP and Vista, with Havok FX being targeted for computation on GPUs (not to be confused with Ageia PhysX hardware -- now pretty well a dead-duck, with the arrival of multicore CPU and massively-parallel-processing GPUs)

Parent

RE: Dual slot?

By decapitator666 on 1/10/2007 5:58:22 AM , Rating: 2

I remember reading somewhere that dx10 hardware is due to their design more easy to program than dx9. This might be the reason for the confusion where people think that the dx10 api includes physics capability at this moment. A simple hardware vs software confusio.

Parent

RE: Dual slot?

By willow01 on 1/9/2007 7:37:36 PM , Rating: 2

If they implement a new driver for the physics GPU, so that they don't necessarily have to rely on using the directx API then it shouldn't really be a problem. That is until there is a directPhysx implementation.

Parent

RE: Dual slot?

By encia on 1/10/2007 4:56:29 AM , Rating: 2

quote:
I personally think that the PhysX approach is more elegant anyway. Not to mention that a $200 PhysX card is a lot cheaper than a 2nd or 3rd high end video card.

quote:

According to pricewatch.com, a SAPPHIRE Radeon X1900GT (RV570) cost around 161 USD.

quote:

I've also heard that some physics functions can't be run on video cards because everything has to be converted to Direct X 9 and back again through software to make it work.

Not with AMD's CTM ("Close To Metal").

Parent

975X and P965 a possible gold mine.

By VooDooAddict on 1/9/2007 12:50:31 PM , Rating: 1

If X1900's can work as a physics card ... And they get this working with the C2D as well then existing X1900 and X1950 users with 2 PCIe16 slots have more of a reason to upgrade to R600.

... oh wait .. AMD/ATI vs Intel chip set. This might be our first glimpse into AMD locking out Intel chipsets. True it also might be a limitation of the x4 PCIe slots on the 975x and P965.

RE: 975X and P965 a possible gold mine.

By decapitator666 on 1/9/2007 2:34:54 PM , Rating: 2

wrong.. intel locks out the ati chipset.. the present ati chipset for core 2 duo is allready capable of doing this.. but intel locks amd/ati out due to MB makers being to afraid to implement a atiamd chipset motherboard.. so...

Parent

RE: 975X and P965 a possible gold mine.

By Xenoterranos on 1/9/2007 2:37:56 PM , Rating: 2

Not only that, but for ATi to make intel-chipset boards, Intel would have to liscense it's chipset tech to AMD...which is as close to never as it gets.

Parent

RE: 975X and P965 a possible gold mine.

By decapitator666 on 1/9/2007 4:31:21 PM , Rating: 2

agreed but forgot to mention

Parent

chilehardware

By piesquared on 1/9/2007 4:18:39 PM , Rating: 2

I might have missed it, but where can I find the english version of chilehardware? I've always looked for it, but never found it.

RE: chilehardware

By hwhacker on 1/9/2007 6:54:51 PM , Rating: 2

It's all about the flops and memory bandwidth. GPU's have it, CPUs don't. The higher you set a baseline (let's say a x1600 with 86.4gflops and 22.4gb/s) the prettier physics is going to look. Look at the flop count and memory bandwidth even for something like Core2, it's not even close. I don't know the flop count of the physX (noone does) but I reckon the x1650xt surpasses it in computational power. Not only could this be awesome if there is a good API for physics, but it shows how the computational power ATi has been putting in even their low-end products can be put to good use. First folding, then physics, then perhaps GPU-accelerated decoding (that we were promised long ago) and who knows what else.

If you doubt the efficiency of multi-gpus in a system over multi-cores, just google 'Peakstream', I reckon it may change your mind.

Parent

RE: chilehardware

By piesquared on 1/9/2007 7:47:11 PM , Rating: 3

Oh but I have no doubt whatsoever! I just thought i'd post simply to see if there was an english version of chilehardware.

Parent

Interesting on paper...

By Axbattler on 1/9/2007 3:09:46 PM , Rating: 2

Now we have to see how well the implementation goes.

I'll be particularly interested in the support and efficiency/performance. What would be required to use the Radeon as a physics processor? How well would a graphic card do the job as a physics processor? What sort of penalty are we looking at when using two cards in GPU mode? Crossfire do not provide twice the performance.. Will we see further penalties (beyond the 'loss' observed on a Crossfire rig)? And if so (I would expect so, but who knows) - how much the penalty? Is it possible to seamlessly switch from the second card from GPU mode to Physics mode?

Lots of questions, but I think that all of those will attribute to the solution's attractiveness.

RE: Interesting on paper...

By qwerty1 on 1/9/2007 5:11:29 PM , Rating: 3

I don't think the penalty of using a gpu as a ppu would be any higher than the Aegia method. After all, they are both separate cards in pci-e slots doing the same thing while not linked to other cards via crossfire or the like. The most likely single penalty that would result is probably in the efficiency of the gpu converted into ppu. Now we just have to see the amount of benefit each type of gpu card provides.

Parent

Too complicated

By Domicinator on 1/9/2007 6:01:50 PM , Rating: 1

Is it just me, or is AMD making things way too complicated and going in the completely wrong direction? Who needs 80 graphics cards and 80 processors in their computer to run a game? The absolutely mind blowing physics we've all seen in Crysis videos are running on the second core of a dual core processor. It's quite clear that extra cards are not needed to achieve stunning graphical AND physics effects that the games of 2007 are going to display.

All this 4x4 technology and extra GPUs to run physics are making AMD look like they're trying to gain a little ground on the Core 2 Duo, or soon the Core 2 Quad. I've been pretty faithful to AMD for a long time, but they're starting to look stupid. I hate to say this, but if I really needed that extra graphics boost, I'd be more likely to buy a PhysX card that takes up very little space in my tower rather than a whole new GPU.

RE: Too complicated

By Acanthus on 1/9/2007 6:38:06 PM , Rating: 2

Memory bandwidth will be a serious constraint on physics in the near future. Think about how many times a second you have to alter data in physics.

Its the same reason they put on board memory on graphics cards, fighting with the CPU for bandwidth (not to mention less bandwidth to begin with) just doesnt work.

Parent

RE: Too complicated

By hwhacker on 1/9/2007 6:59:49 PM , Rating: 2

whoops, I replied above his post accidentally, but I agree, memory bandwidth is certainly an issue. CPUs don't have it, but even fairly low-end GPUs do. Many cards are available for less than PhysX with a 256-bit bus and 1.4ns ram. That's 44.8gb/s. What's that, over twice Core 2 and over 3x PhysX.

Parent

"Let's face it, we're not changing the world. We're building a product that helps people buy more crap - and watch porn." -- Seagate CEO Bill Watkins