J. Dzhugashvili writes "Nvidia completed its acquisition of Ageia yesterday, and it has revealed exactly what it plans to do with the company's PhysX physics processing engine. Nvidia CEO Jen-Hsun Huang says Nvidia is working to add PhysX support to its GeForce 8 series graphics processors using its CUDA general-purpose GPU (GPGPU) application programming interface. PhysX support will be available to all GeForce 8 owners via a simple software download, allowing those users to accelerate games that use the PhysX API without the need for any extra hardware. (Older cards aren't CUDA-compatible and therefore won't gain PhysX support.) With Havok FX shelved, the move may finally popularize hardware-accelerated physics processing in games."
...what will be calculating my 3D images, if the GPU is already working on the physics? It is not like there is so much spare capacity left over in modern games anyway...
The GeForce 8800GTX, for example, has 16 stream processors, each of which can run up to 8 identical commands per clock (SIMD). They're not the same as the main graphics processors; they're a separate part of the chip AFAIK.
On the CUDA forums, we've gone back and forth about this, and the diagrams that people base this statement on are backwards. There are 16 multiprocessors (to use the NVIDIA terminology), each with 8 stream processors per multiprocessor. The 8 stream processors on each multiprocessor run the same instruction at once, but on separate register files. Multiprocessors, however, are completely independent, so in principle, one could imagine partitioning the resources between physics simulation and 3D rendering. This sort of partitioning has not been made available through CUDA yet, but hopefully this means we will see it soon.
You are correct that these 128 stream processors (however you slice them) are the main compute engine. There is additional circuitry to do hardware accelerated video decoding, but NVIDIA has not exposed that functionality to 3rd party programmers, and it isn't used during 3D rendering.
They're not the same as the main graphics processors; they're a separate part of the chip AFAIK.
You know wrong. OpenGL, Direct 3D and CUDA all share the same stream processors on the chip.
(Think! Why would NVIDIA waste expensive chip real estate for stream processors if they weren't useful for 99.9% of the applications running on these chips?)
"3D acceleration" is no longer a meaningful concept, as the majority of the "acceleration" happens on general purpose stream processors. A modern GPU is a parallel processor which just happens to have a rasterisation engine built-in. I'm guessing we'll eventually move the stream processors closer to the CPU, and integrate the graphics handling parts in the motherboard.
Not to mention a GUI isn't useful either, since everything can be done on the command line anyway. In fact, all you need is a bank of LEDs to indicate the state of the registers!
...what will be calculating my 3D images, if the GPU is already working on the physics? It is not like there is so much spare capacity left over in modern games anyway...
FTFA
Our expectation is that this is gonna encourage people to buy even better GPUs. It might--and probably will--encourage people to buy a second GPU for their SLI slot. And for the highest-end gamer, it will encourage them to buy three GPUs. Potentially two for graphics and one for physics, or one for graphics and two for physics.
I hope the NVIDIA acquisition and now this news will drive the adoption of the PhysX Engine. Right now, if you look at the list of titles [ageia.com], the PhysX Engine is not used by many games (namely, mostly Unreal3-Engine titles).
If the adoption picks up, maybe Havok (which is now Intel property) will not remain the only physics engine in town, but right now, this news will not affect a whole lot of games...
Isn't the real problem that the games that DO incorporate PhysX hardware support don't really showcase the technology in any carnal desire type manner. There's no equivalent of GLQuake, that drove adoption of the original 3D cards.
I agree, hardware physics needs a killer app. I was playing around with the idea of adding PhysX support to Quake 2 and modifying some maps to have real liquids instead of the fake water it normally had, but after making some tests apps I realized how incredibly slow the physics are on a CPU. I'm not holding my breath, but I hope the GPU is capable of at least playable frame rates.
One might think reluctance to adopt PhysX would be knowing that a large number of your customers don't use NVIDIA cards and therefore wouldn't be able to take advantage of the technology. It's almost the same reason why game companies aren't making their games Vista only.
so now that my vid card is processing the 3d graphics
and the physics (which is really only eye candy)
how about we make it -the gpu- run the O/S tooo!!!!
ooo ooo my next summer project!
have linux run on just the video card! (openmosix is still around right:-D?):-p
what?!?!?! it runs on everything else. right now i'm typeing this on my old 700mhz laptop running the latest debian:-p
Any physics done on a PhysX card is only eye candy.
The latency to get the results of the calculations back from the card is high enough that your frame rate would cut in half (or worse) if you waited for the results. So games use it for particle effects, and render the results a frame or two behind. It doesn't matter at all for pure eye candy stuff, but it's just not useful for anything affects gameplay.
Physics covers a lot, from gravity, inertia, particles, collisions, IK and various other bits and pieces. Not everything lends itself to acceleration. So what will be accelerated by this?
Obviously, gravity and other kinds of non-steady motion are good targets for acceleration. And because of NVidia's evil closed source drivers, the best way to accelerate your GeForce is at 9.81 m/s**2.
From what I understand of this (and I could be wrong), the physx accelerator is primarily used to add eye-candy -- so things like showers of sparks, sprays of blood, geysers and clouds of dirt or water or snow on an impact (whether a footfall or a weapon strike...), leaves falling when you shoot trees, better hair and clothing, clouds, rain drop impacts, etc, etc.
All the physics processing for all those particles can be offloaded to the physx engine, allowing more particle effects to be going on at higher level of detail and realism (e.g. incorporating 'wind' etc..) without dragging down the cpu.
Its cool... but not earthshattering. And its a logical step to incorporate it into a video card.
I don't honestly know if it it can really be used to assist with the trajectory calculations of the interactive players tank or fighter plane or whatever, etc... but I doubt it. And it probably doesn't matter either. That is a minor part of the scene...each shower of sparks by itself probably requires more physics calculations than an entire squadron of planes... more independant particles in the shower.
I don't honestly know if it it can really be used to assist with the trajectory calculations of the interactive players tank or fighter plane or whatever, etc...
It won't unless you can get the data back from the card. It's useless for some calculations and I much prefer the way a dedicated card works that feeds the data back to a program.
Why? Well say you're running an MMOG server (or any server for that matter), you could have all sorts of crazy physics running on the server through a dedicated chip, or e
You can't do that with physics on a graphics card because it's a one way pipeline, from your program to your monitor.
I don't think that's the case. Graphics cards work on the same PCI-X buses that acceleration cards probably use lately. They use DMA to communicate with main memory without involving the processor. The VRAM might be optimised for writing, but it should be very possible to do calculations on the card, and get the results back. That's the whole point of the generalised GPGPU techniques.
I don't think that's the case. Graphics cards work on the same PCI-X buses that acceleration cards probably use lately. They use DMA to communicate with main memory without involving the processor. The VRAM might be optimised for writing, but it should be very possible to do calculations on the card, and get the results back. That's the whole point of the generalised GPGPU techniques.
Nitpicking: PCI Express is not PCI-X. PCI-X was a derivative of the parallel PCI bus and never found in mainstream machines.
Nitpicking: PCI Express is not PCI-X. PCI-X was a derivative of the parallel PCI bus and never found in mainstream machines.
Nitpicking your nitpick... it's not worth pointing out that PCI-X is different from PCI Express unless you also point out that PCI Express is usually abbreviated as PCIe or PCI-E.
I confused you with an MMO example. What I am looking at here is consumer grade hardware that allows physics computation but not only for graphics, I want to use physics for much more then just what you can see on the screen.
Yeah, that's what City of Heroes uses it for. When a villain smashes a mailbox, a cloud of letters goes flying everywhere. Things like that. It really is eye-candy, doesn't improve the game in any majorly meaningful way. Certainly not worth the $150-$200 extra that an Ageia PhysX standalone card would run you. Of course, you don't have to have the sepaate card even now to get some of the benefits; the Ageia engine will run in software, too, just not as well. It will be interesting to see what happens when nV
Nope, the eye candy was done b/c it's harder to re-architect games to use more extensive physics in primary gameplay. You have to shove all the physics models into the application code for all that advanced simulation. It's a lot easier to run 10,000 copies of a simple parabolic arc. Rigid body physics, constrained motion, etc all take up some decent CPU. As does collision detection. So far, game developers have had to do with simplified collision geometries, simplified models, etc. As a first stab at
Bah I just woke up so this may come across clearly. The reason why it's only for eye candy at the moment is because developers do not want to fork the gaming experience. Since accelerated physics would create a have and have not situation for gamers, where the non accelerated experience would be too slow to be acceptable, developers choose to only fluff up the eye candy portions because you could not make the game play experience identical between the two.
This means that you could fork development and have t
Naw, it's much easier than you think. All of Physics can be expressed by just one equation, the Grand Unified Theory, the computation of which is accelerated by PhysX. The Grand Unified Theory was first discovered when programmers at Valve tried to optimise the physics engine of HL2 [bbspot.com]. From the link:
Game Engine Software Engineer at Valve, Jose Garcia discovered the theory. "The game engine ran too slowly. I was assigned the job of speeding it up," he said. "I started out by combining some of the gravity equations with some of the other force equations and found it all started to fit together. After a day, I had fine-tuned the entire physics-animation functions down to four lines of code, which ran a bit faster," he added.
If existing cards can be upgraded thrugh a software patch, NVidia should have been able to do this all along. Are the PhysX people just much better at coding physics, or is there another reason this haven't already been added?
In other words, did NVidia just buy some clever code?
Because they don't get developers to plaster things like that "en-veeee-diar" voice over the start of games?
Having said that, I use Linux so my next card probably will be an nVidia because of the better drivers. Unless ATI get better in the one/two/three years until I buy a new card.
It'll be interesting to see what they can do to really exploit this PhysX and make it worth its while, though.
Better is very subjective. We have both nVidia and ATI based thinkpad laptops running Ubuntu at work. And what I've noticed is that the ATI ones can do a kernel update with out screwing up the gfx drivers and they can switch between single and dual monitors (necessary when going on and off dock) without restarting X. On the other hand the nVidia ones have a pretty lil graphical config tool, while the ATI ones use a somewhat arcane and unreliable command line program.
Personally I wouldn't trade my ATI one for an nVidia one any day, I very much like being able to unplug from the dock and switch down to single screen without closing and restarting all my apps.
See my other post for what doesn't work quite right with my ATI card. I guess there might be a difference between desktop and laptop, but most of those things aren't an issue for me. Can't say a kernel update has ever screwed up the graphics drivers on my work machine with an nVidia card, but then I use the Livna repositories for Fedora to download the RPMs for the graphics along with the kernel update.
Whereas for me, ATI hasn't had usable XV support since 8.35.5 (or thereabouts), and the 3D rendering is buggy as hell... which kinda defeats the purpose of using a dedicated GPU. Go look at the known issues in the release notes - it reads like an alpha dev-snapshot. I regret fitting ATI to my laptop for the sake of a supposed performance advantage over the Nvidia option, and my next machines will absolutely be NVidia Quadro (Sun workstation, yes Nvidia even provides drivers for Solaris x86) or Intel integrated (ultraportable notebook).
Having said that, I use Linux so my next card probably will be an nVidia because of
the better drivers. Unless ATI get better in the one/two/three years until I buy a new card.
AMD has open sourced their Radeon drivers. What more could you ask for than that?
Graphics drivers that run Compiz-Fusion and let my play DirectX games through Wine without having to do a force redirect with a key combo? Graphics drivers on a 256MB X800 that rotate the cube as smoothly as the low-end nVidia in my machine at work? Graphics drivers that run video at full-screen with Compiz-Fusion running without all the tweaking and tinkering? Yes, the open sourcing might be useful, but nVidia works more smoothly with DirectX, Compiz-Fusion and media played through anything other than VLC (
The last big news I saw was not that they OSed the drivers, but that they had given partial card specs and promised more.
Please note that Matrox did the same thing in 1999 - They gave partial card specs (insufficient for implementing any 3D) and promised more, but never delivered. Lots of Linux users got suckered into buying paperweight G200s (including myself) back then. I will buy a card that performs as advertised NOW (whether or not it is with an open source driver or not), not a card that the manufacturer promises will eventually perform as advertised but can't at the moment.
The ATI/AMD guys responsible for releasing docs are also FOSS devs working on X.org. The next data dump will be tcore code, which is used to program 3D shaders on R500 and R600 GPUs, and will probably be relevant to the R400 engine, which is similar to the engine in certain R500 cards. This is not a bait and switch; ATI needs the market share too badly.
Please note that Matrox did the same thing in 1999 - They gave partial card specs (insufficient for implementing any 3D) and promised more, but never delivered
Bull! I used to routinely play Quake3, as well as TuxRacer (full version) with a matrox g200 card in my Linux box.
See this site [ntlug.org] for instance, the documentation may not have been the best, but it was enough.
I know they had problems getting an OpenGL driver out for Windows, I'm not sure they ever got it right, and a lot of people were pissed, but
The RadeonHD driver *is* GPL'd, and all the specifications necessary for writing your own drivers from scratch are in the process of being released. Significant amounts have already been released after being checked out by AMD's lawyers.
And the closed-source, binary module is still making progress while all that other stuff happens.
Not quite true. They said they'd release the specs so said driver could be created but they only released partial specs for the 2D aspects of their chips. They still haven't released updated doc's for 3D/Video rendering etc
Exactly. I support ATI now primarily for that fact. After using them I realized that they are just more capable all around cards. Who cares which one is minutely faster? In the end it doesn't matter what card I play my games on, so why should they be shoving ads for nVidia down my throat?
As an OpenGL developer, I can say that I will never touch an ATI product when I have a choice.
Their driver support lags behind nVidia by years, and when they "support" a feature, it will often be in software with no warning that it is - so instead of failing with a useful error message, all you know is that *something* you did causes your system to render at 1 frame per minute and be completely unusable.
I have spent weeks bending over backwards and through hoops to get our ATI test card to agree with me, just because it is so darn unresponsive when anything goes wrong. Non power of two texture in one of your models because the modeller apparently ignored your instructions? No warning, no error - just a hung machine that will take 5 minutes to kill the process.
No it wasn't a waste of time.. the PhysX-card is much better at calculating physics than the 8800 which is already busy enough doing 3D.. So the combo of 8800 with Physx-card is much better than using a dual 8800..
They are not as they did put in the market a product that has a place. Okay they did fail to sell succesfull a product but the first graphic accelerators we'rent the most lucky either! What matters is that the idea sticks and that now we might see Physics being offloaded more. So it has a place but the one putting it first on the market well, they didn't fare well!
I am certainly not a lawyer, but here's how I understand the situation: GeForce 8 cards have had CUDA support from day one. nVidia bought Ageia, and with it all they need involving the PhysX API. This upcoming download to enable physics acceleration will be a PhysX-to-CUDA wrapper that is in no way locked down to the Geforce 8 architecture (which is the point of CUDA).
By my understanding of SarbOx (which admittedly is not great) this falls under the same category as programs being written for an Intel processo
Nice! But... (Score:5, Insightful)
Re: (Score:2)
Re: (Score:3, Interesting)
Re:Nice! But... (Score:5, Informative)
On the CUDA forums, we've gone back and forth about this, and the diagrams that people base this statement on are backwards. There are 16 multiprocessors (to use the NVIDIA terminology), each with 8 stream processors per multiprocessor. The 8 stream processors on each multiprocessor run the same instruction at once, but on separate register files. Multiprocessors, however, are completely independent, so in principle, one could imagine partitioning the resources between physics simulation and 3D rendering. This sort of partitioning has not been made available through CUDA yet, but hopefully this means we will see it soon.
You are correct that these 128 stream processors (however you slice them) are the main compute engine. There is additional circuitry to do hardware accelerated video decoding, but NVIDIA has not exposed that functionality to 3rd party programmers, and it isn't used during 3D rendering.
Parent
Re:Nice! But... (Score:5, Informative)
(Think! Why would NVIDIA waste expensive chip real estate for stream processors if they weren't useful for 99.9% of the applications running on these chips?)
Parent
Re: (Score:2)
3D acceleration itself is not useful for 99.9% of the applications running on these chips, if we include computing activities that are not gaming.
Re: (Score:2)
Re: (Score:3, Funny)
Re: (Score:3, Informative)
...what will be calculating my 3D images, if the GPU is already working on the physics? It is not like there is so much spare capacity left over in modern games anyway...
FTFA
Hopefully, now PhysX adoption will become better.. (Score:3, Insightful)
If the adoption picks up, maybe Havok (which is now Intel property) will not remain the only physics engine in town, but right now, this news will not affect a whole lot of games...
Re:Hopefully, now PhysX adoption will become bette (Score:4, Interesting)
Parent
Re: (Score:2)
Re:Hopefully, now PhysX adoption will become bette (Score:2, Insightful)
It's almost the same reason why game companies aren't making their games Vista only.
now that the gpu is doing 2 things lets do 3 !!! (Score:3, Funny)
Re: (Score:2)
The latency to get the results of the calculations back from the card is high enough that your frame rate would cut in half (or worse) if you waited for the results. So games use it for particle effects, and render the results a frame or two behind. It doesn't matter at all for pure eye candy stuff, but it's just not useful for anything affects gameplay.
So, what's actually accelerated here? (Score:4, Informative)
Re:So, what's actually accelerated here? (Score:5, Funny)
Parent
Re:So, what's actually accelerated here? (Score:5, Interesting)
All the physics processing for all those particles can be offloaded to the physx engine, allowing more particle effects to be going on at higher level of detail and realism (e.g. incorporating 'wind' etc..) without dragging down the cpu.
Its cool... but not earthshattering. And its a logical step to incorporate it into a video card.
I don't honestly know if it it can really be used to assist with the trajectory calculations of the interactive players tank or fighter plane or whatever, etc... but I doubt it. And it probably doesn't matter either. That is a minor part of the scene...each shower of sparks by itself probably requires more physics calculations than an entire squadron of planes... more independant particles in the shower.
Parent
Re: (Score:2)
It won't unless you can get the data back from the card. It's useless for some calculations and I much prefer the way a dedicated card works that feeds the data back to a program.
Why? Well say you're running an MMOG server (or any server for that matter), you could have all sorts of crazy physics running on the server through a dedicated chip, or e
Re: (Score:3, Interesting)
I don't think that's the case. Graphics cards work on the same PCI-X buses that acceleration cards probably use lately. They use DMA to communicate with main memory without involving the processor. The VRAM might be optimised for writing, but it should be very possible to do calculations on the card, and get the results back. That's the whole point of the generalised GPGPU techniques.
On p
Re:So, what's actually accelerated here? (Score:5, Informative)
Parent
Re: (Score:2)
Nitpicking your nitpick... it's not worth pointing out that PCI-X is different from PCI Express unless you also point out that PCI Express is usually abbreviated as PCIe or PCI-E.
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Of course, you don't have to have the sepaate card even now to get some of the benefits; the Ageia engine will run in software, too, just not as well. It will be interesting to see what happens when nV
Re: (Score:2)
Rigid body physics, constrained motion, etc all take up some decent CPU. As does collision detection. So far, game developers have had to do with simplified collision geometries, simplified models, etc. As a first stab at
Re: (Score:2)
The reason why it's only for eye candy at the moment is because developers do not want to fork the gaming experience. Since accelerated physics would create a have and have not situation for gamers, where the non accelerated experience would be too slow to be acceptable, developers choose to only fluff up the eye candy portions because you could not make the game play experience identical between the two.
This means that you could fork development and have t
Re:So, what's actually accelerated here? (Score:5, Funny)
Parent
I dont quite get it (Score:3, Interesting)
In other words, did NVidia just buy some clever code?
Compatible cards (Score:3, Interesting)
Re: (Score:2)
Having said that, I use Linux so my next card probably will be an nVidia because of the better drivers. Unless ATI get better in the one/two/three years until I buy a new card.
It'll be interesting to see what they can do to really exploit this PhysX and make it worth its while, though.
Re:It's the "Ray" experience. (Score:5, Interesting)
Parent
Re: (Score:3, Interesting)
Re:It's the "Ray" experience. (Score:5, Informative)
Parent
Re: (Score:2)
AMD has open sourced their Radeon drivers. What more could you ask for than that?
Re: (Score:2)
Yes, the open sourcing might be useful, but nVidia works more smoothly with DirectX, Compiz-Fusion and media played through anything other than VLC (
Re:It's the "Ray" experience. (Score:4, Insightful)
The last big news I saw was not that they OSed the drivers, but that they had given partial card specs and promised more.
Please note that Matrox did the same thing in 1999 - They gave partial card specs (insufficient for implementing any 3D) and promised more, but never delivered. Lots of Linux users got suckered into buying paperweight G200s (including myself) back then. I will buy a card that performs as advertised NOW (whether or not it is with an open source driver or not), not a card that the manufacturer promises will eventually perform as advertised but can't at the moment.
Parent
Re: (Score:2)
Re:It's the "Ray" experience. (Score:4, Insightful)
So did Matrox...
Parent
Re: (Score:2)
Bull! I used to routinely play Quake3, as well as TuxRacer (full version) with a matrox g200 card in my Linux box. See this site [ntlug.org] for instance, the documentation may not have been the best, but it was enough.
I know they had problems getting an OpenGL driver out for Windows, I'm not sure they ever got it right, and a lot of people were pissed, but
Re: (Score:3, Informative)
And the closed-source, binary module is still making progress while all that other stuff happens.
Re:Open Source != Holy Grail (Score:5, Informative)
Parent
Re: (Score:3, Insightful)
Re: (Score:2)
No, the way it is meant to be is a game that I play on my computer, not an advert for a specific card manufacturer!
Re:It's the "Ray" experience. (Score:5, Interesting)
Their driver support lags behind nVidia by years, and when they "support" a feature, it will often be in software with no warning that it is - so instead of failing with a useful error message, all you know is that *something* you did causes your system to render at 1 frame per minute and be completely unusable.
I have spent weeks bending over backwards and through hoops to get our ATI test card to agree with me, just because it is so darn unresponsive when anything goes wrong. Non power of two texture in one of your models because the modeller apparently ignored your instructions? No warning, no error - just a hung machine that will take 5 minutes to kill the process.
Give me nVidia any day.
Parent
Re: (Score:2)
Re: (Score:2, Interesting)
Re: (Score:2, Insightful)
Re: (Score:2)
GeForce 8 cards have had CUDA support from day one.
nVidia bought Ageia, and with it all they need involving the PhysX API.
This upcoming download to enable physics acceleration will be a PhysX-to-CUDA wrapper that is in no way locked down to the Geforce 8 architecture (which is the point of CUDA).
By my understanding of SarbOx (which admittedly is not great) this falls under the same category as programs being written for an Intel processo
Re: (Score:3, Funny)