Print 39 comment(s) - last by Clauzii.. on Mar 22 at 9:00 PM

Havok and NVIDIA team up to deliver hardware physics processing on GPUs through the use of SLI

At the Games Developer Conference (GDC) NVIDIA announced a strong partnership with Havok, to leap into the forefront of physics processing. However, despite the trend to offload physics processing to a dedicated processor (such as AGEIA's PPU), Havok and NVIDIA announced that using SLI, a second GPU can be used to perform physics computations. The feature is customizable and a user can set their SLI configuration in either graphics+physics or graphics+graphics (traditional SLI) modes.

Utilizing Havok FX and NVIDIA graphics technology, game developers can now implement sophisticated physical phenomenon such as debris, smoke, and fluids that add immense detail and believability to game environments. Game designers can include advanced physics effects without burdening the CPU and slowing game-play, since the effects are simulated and rendered on the GPU.

We previously talked about Havok and its HavokFX engine, which is a low level engine used in a host of games to accelerate physics and provide realistic looking animations. NVIDIA states that offloading physics to the GPU will bring greater value for both developers and gamers, since no extra hardware and extra development is required.

With extra processing power on tap through the use of an SLI configuration, users will have the ability to choose. Not all games will be physics heavy, but may be graphics effects heavy. In a case like this, the user would benefit greatly by having the ability to turn his second GPU into regular GPU mode. The cost is already spent on having an SLI configuration. The user would get the added bonus of being able to switch to physics SLI mode if the application demands so. Clearly the direction of physics that NVIDIA is taking is keeping users from having to go with a dedicated physics processor.

Comments     Threshold

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

By JoKeRr on 3/20/2006 11:50:21 PM , Rating: 2
I think ageia physX will be pretty darn successful if it comes cheap, say in the 100~150 bux range. Cause that will make SLI physix pretty darn expensive. But SLI does have the advantage of extra graphics power, so it's going to be a hard call to make.

RE: price
By Goty on 3/20/2006 11:55:31 PM , Rating: 3
But the thing is that you can't use both the physics processing and the graphics processing power at the same time, so you ultimately lose out on situations where you might need both.

RE: price
By Hypernova on 3/21/2006 2:19:33 AM , Rating: 2
Aegia takes up a PCI-E slot too so unless you got a Quad Royal it's pritty much the same.

RE: price
By Hypernova on 3/21/2006 2:23:59 AM , Rating: 2
Still very interested with how it comapres with Aegia in terms of preformance, doubt GPU's can beat a specialised PPU. But if nvedia comes close (say 80%) then Aegia could be in trouble, if it's something like 40% then Aegia still got a chance.

Quad SLI
By kalaap on 3/21/2006 9:31:47 AM , Rating: 2
I think the Quad SLI would probably give you the best of both worlds w/ gfx+phys.

RE: Quad SLI
By suryad on 3/21/2006 12:19:54 PM , Rating: 2
Technically it should and will. But we have to remember the benefits of physics on graphics chip is possible only because everything is so cpu limited right now. What happens when the cpu is able to overcome enough of its shortcomings that the complex physics can no longer be run on the graphics cards without a performance hit in fps?

Also it seems to me that the physics that the Nvidia cards are performing are not 'real' physics. It seems to me that the cards are actually just drawing the pixesl in such a way as it is supposedly simulating physics. Not the stuff Agiea's processor is doing.

For further read maybe these links would help: le.x/9610 e=30434

The authors are better at me than putting ideas into words.

As you may already know that Epic games has invested in Ageia. In fact Unreal 3.0 is going to support the Agiea PPU (physics processing unit) and right now the PPU library is given to game devs for free. I dont know if it is the same with the hardware and Havok charges game devs the price of a Ferrari to use their stuff. I really hope Agiea takes off and makes money in the long run. I think this GPU based physics is a hack personally. I would rather have my quad sli driving every game at 2560 x 1600 resolution on a 30 inch or larger monitor.

RE: Quad SLI
By masher2 on 3/21/2006 12:51:28 PM , Rating: 3
> "Also it seems to me that the physics that the Nvidia cards are performing are not 'real' physics. "

No, thats not a good way to put it. They're both doing "real" game physics, but the Nvidia approach cannot output data back to the CPU. So you can't use the results to influence future gameplay...meaning its not truly interactive.

Example. Main character shoots wall. It explodes, bricks and timbers fly everywhere, falling in natural, kinematically-accurate paths. (both solutions). One brick strikes main character, knocking him backwards (only the Ageia solution).

Main character walks through hole in wall (both solutions). Main character tries to walk through hole in wall, but gets stuck on one timber lodged partially in the way (Ageia solution).

RE: Quad SLI
By suryad on 3/21/2006 1:27:35 PM , Rating: 1
Great! Thanks for the input!

Aegia better
By DangerIsGo on 3/21/2006 11:39:02 AM , Rating: 2
I feel that aegia will prob get my vote. If I have a, lets say 7900GTX, I dont feel like spending another $500 for a gfx card which will act as a physics card. Way too expensive considering PPU's cost between $100 and $300. If its small enough, I can fit a PPU in PCIe between my SLI'd cards. Sure its a good idea but moneywise, its not.

RE: Aegia better
By jkostans on 3/21/2006 12:44:22 PM , Rating: 2
Off topic big time but with a seperate physics solution like aegia or a second gpu, will the graphics framerate and physics update rate be unrelated? Like if there is a huge explosion with lots of physics calculations so only 20 updates per second can be done, but the graphics card can maintain 60+ fps. Will the game still run at 60fps and the physics objects will just move and interact at 20fps?

RE: Aegia better
By masher2 on 3/21/2006 1:36:21 PM , Rating: 2
It all depends on how its coded. Lazy programmers will probably tie everything to a synchronous pipeline, and thus a slowdown in physics calculations would affect framerate...but it doesn't necessarily have to be true.

RE: Aegia better
By FearaFox on 3/21/2006 3:42:15 PM , Rating: 2
or just tone down the physics effects to match the graphics effects, / vice versa even

It should be seperate
By Cunthor01 on 3/21/2006 1:35:12 AM , Rating: 2
I would prefer to stick the physics card in a PCI slot seperatly, and not be bound to nvidias sli solution in order to get better performance. This should be an attractive solution to any single/sli mobo on the market, not a high-end gaming card/mobos combo

RE: It should be seperate
By FearaFox on 3/21/2006 3:38:17 PM , Rating: 2
i dont think there would be enough bandwidth for that :\

RE: It should be seperate
By beemercer on 3/21/2006 3:48:32 PM , Rating: 3
I believe ageia is launching a PCI PPU first, and then a higher performance PCI express version later.

Ah, the joys of a new technology....
By Targon on 3/21/2006 10:02:25 AM , Rating: 2
When 3Dfx came out with the Voodoo, it provided something not available through software. As a result, you saw a HUGE benefit in games with a Voodoo card. Software emulation while MUCH slower, SHOULD have been able to provide an equivilant image, but it was never done.

Now, we have dual core processors out there when the vast majority of games arn't written with a multi-threaded design. This means the second core is sitting there with very little use. A game could easily do the physics in software and come up with game play that is identical to what you would see WITH a physics co-processor. So, if a game developer decides to do it themselves, people without the physics processor won't see much of a difference. Most games arn't CPU limited as well, but instead are graphics card limited.

So, keep that in mind when you say how amazing a physics co-processor would be in games if the physics can be done well by the CPU without a major slow-down. Dual-core today, quad-core on the desktop in two years.

RE: Ah, the joys of a new technology....
By masher2 on 3/21/2006 10:06:53 AM , Rating: 1
> " game could easily do the physics in software and come up with game play that is identical to what you would see WITH a physics co-processor"

No. These coprocessors are considerably more powerful than a second cpu core. They're more powerful even than three free cores.

Second of all, that "spare" core is only semi-idle...and thats the current situation, only because most games are single-threaded. Now that dual-core systems are becoming common, expect to see that second core utilized much heavier.

RE: Ah, the joys of a new technology....
By saratoga on 3/21/2006 2:49:49 PM , Rating: 2
A CPU core is also much more flexible, has more memory with a lower latency, and is steadily getting faster. In the long run, coprocessors like this don't make much sense. They have to fight the high latency of the interconnect to the CPU as well as to the main memory. On top of that, they're just specialized processors with fast vector units. With vector performance increasing on x86 cores, and 2-4 cores likely over the next few years, theres not going to be much use for these cards.

Aegia seems to realize that. The marketing stuff on their site plugs their SDK as much as their hardware, and they show the SDK wrapping calls to an x86 core as well as to their own hardware.

By masher2 on 3/21/2006 4:08:18 PM , Rating: 2
> "In the long run, coprocessors like this don't make much sense"

I disagree totally. Yes CPUs are "getting faster"...but so are coprocessors. And a generalized design is ONLY useful when you're tring to press one piece of silicon into several different functions. You can never beat the performance of dedicated circuitry, however.

If you look at the history of the computer, coprocessors are becoming more and more prevalent. Most servers contain network coprocessors in the NICs, servers and desktops both may contain disk coprocessors, sound and video coprocessors, and many other forms of specialized silicon.

Even the old floating point coprocessor still simply got sucked directly onto the CPU die. A similar thing thing will happen with a physics coprocessor I believe. In ten years when your average CPU has a hundred cores, many of them will be specialized for certain tasks.

By DigitalDivine on 3/20/2006 11:32:46 PM , Rating: 1
i like the aegia solution better

RE: aegia
By PLaYaHaTeD on 3/20/2006 11:40:05 PM , Rating: 1
What an unbelievably elegant and well thought out response. You've definitely made me a believer in aegia's solution.

RE: aegia
By Xenoterranos on 3/20/2006 11:43:31 PM , Rating: 2
I like his response better.

RE: aegia
By czarchazm on 3/20/2006 11:52:54 PM , Rating: 2

Slow solution...
By Clauzii on 3/21/2006 2:38:00 AM , Rating: 2
Well, Ageia and Clearspeed is not like 10-50 times faster - more like a couple of thousand times, so I see NO reason bying a fully equiped graphics card with connectors, DACs, expensive memory etc. for physics calculations for which the GPU is not specifically made.

But for those who have a nVidia it´s nice enough - when/if it gets out.

RE: Slow solution...
By masher2 on 3/21/2006 8:37:09 AM , Rating: 1
> "Well, Ageia and Clearspeed is not like 10-50 times faster - more like a couple of thousand times"

Oops, incorrect. Clearspeed's solution is a generalized math processor. It is roughly five times more powerful than the fastest AMD and Intel CPUs.

Ageia, on the other hand, is selling an ASIC customized specifically for gaming physics. On those particular tasks, its more like 30-100X more powerful.

RE: Slow solution...
By Clauzii on 3/22/2006 9:00:17 PM , Rating: 2
Hmm, why do I see a zillion more things on-screen then?

Helpful for Aegia?
By Wilco on 3/21/2006 3:29:28 AM , Rating: 3
This may even be helpful for Aegia. One of the main concerns with uptake of the product is the Chicken and Egg lack of upgrade path. The extra physics can't have a big impact on gameplay etc, until the majority of people can use it, and people won't want to spend a lot of money on something that won't give them a good return (obviously some people buy 7900GTXs but they're a minority :) ). With another upgrade path, albeit an expensive one, it will help the uptake of the software and hardware. ALso NVidia will provide a nice bit of competition to keep the improvements coming.

RE: Helpful for Aegia?
By masher2 on 3/21/2006 8:39:49 AM , Rating: 2
Except that the APIs between the Havok and Ageia solutions are totally different. So support for the NVidia approach doesn't mean more games compatible with Physx, and vice versa.

not substantial
By neihrick on 3/21/2006 10:25:29 AM , Rating: 2
thing is about havok fx, it will be nice, but only for a while, since the physics from it do not effect gameplay, well, not in the literal sence. you could argue the smoke would but not substantially. whereas the physX will allow for destructible environments and bodies(dead) could be stacked, or like on BLACK, crumbling walls killing enemies. havok fx could be capible of this but on a much smaller scale compared to physX

RE: not substantial
By Wwhat on 3/21/2006 12:30:46 PM , Rating: 2
Exactly right, the two things are not the same, the havok thing is to make pretty non-gameplay optical splendor, the PPU however is interactive with the actual player and gameplay.
Obviously you can do both at the same time too if you have the hardware, that is a PPU card and a solid performing SM3.0 graphics card
Some of the havok claims are a bit silly btw, in the sense that games already use shaders to do graphic effects that interacts with itself, say fire or smoke or even grass, remember the grass in the old 3dmark test that waved with the wind?

Holy @&%$!
By ToeCutter on 3/22/2006 4:05:47 PM , Rating: 2
This story prodded me into to looking into these PPUs and I found this link while looking for info. This definitely changed my opinion of PPU cababilities. The Aegia solution looks pretty sweet, and apparently, support for PhysX can be patched into existing titles.

Simply mind-boggling real-time game engine demos: l

RE: Holy @&%$!
By ToeCutter on 3/22/2006 4:07:11 PM , Rating: 2
And, I just noticed that FS has a story up that BFG intends to release their first Aegia board in MAY. And, it's PCI (!).

Agiea is better
By sotti on 3/21/2006 12:29:17 AM , Rating: 3
So since most people will crank the resolution up till a game becomes GPU limited you have to sacrafice you're frame rate and turn the resolution down a couple of notches or give up the physics.

sounds like a bad plan to me.

Wait 6-8mo and watch nvidia announce a dedicated pcie 1x card.

By Lonyo on 3/21/2006 3:53:11 AM , Rating: 2
This STILL isn't nVidia only, it's for SM3.0 hardware.
Misinformation FTW.

RE: Gah
By z3R0C00L on 3/21/2006 10:27:13 AM , Rating: 1
I think most people know it was ATi who pioneered the idea and have been working hard to implement the idea into there VPU's architecture in such a way that it does not impact the VPU's performance to the extremes it will on nVIDIA's cards.

Ultrathreaded architecture and the Ringbus memory controller were mean't to be used in conjunction with new ways of using the VPU's floating point power. Hence why we have the AVIVO video transcoder (soon to be fully VPU hardware based) and the post processing power for DVD, DIVX and not HDTV all pioneered by ATi and later copied by nVIDIA.

Not turning this into an nVIDIA vs. ATi debacle but when was teh last time nVIDIA pioneered something? I seem to recall SLI and Quad SLI already being used by another company whom nVIDIA bought out.. (3Dfx).

By themusgrat on 3/21/2006 9:12:31 AM , Rating: 3
I like this. Since putting 2 GPUs together is not really a cost effective solution, this will be a better reason to get SLI. The sweetest thing, though, would be being able to have a high end card, like a 7900GTX, then a 6600GT or something for physics.

Mismatched Cards?
By johnford64 on 3/20/2006 11:45:32 PM , Rating: 2
This would only matter to me if the second physics card could be a different card, say use a 7900GTX for Graphics and a 7800GT for physics. Otherwise this is a waste of money and i would definately go with the Aegia

AMD with Math Co-processor
By Kryptonite on 3/21/2006 1:50:34 AM , Rating: 2
so wat happens if AMD decide to use the extra HTT link straight into a co-processor like the CSX? (or Ageia for that matter)

linky: spx?newsid=1276

bring on K8L

Bad reporting IMO
By Lonyo on 3/21/06, Rating: 0
"A politician stumbles over himself... Then they pick it out. They edit it. He runs the clip, and then he makes a funny face, and the whole audience has a Pavlovian response." -- Joe Scarborough on John Stewart over Jim Cramer
Related Articles
AGEIA Continues To Gain PhysX Support
January 19, 2006, 2:34 AM

Copyright 2016 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki