View Full Version : Ageia physx chip

05-15-2006, 11:07 AM
I would like to know if SoW/BoB will support the Ageia physx chip.Also will radio direction finding (such as "Knickebein") be implemented?

05-18-2006, 01:22 AM
I'm whith zino, I am also curious about this...

As one of the first to whine about 6dof in FB/PF... don't get me started about Ageis in BOB, lol.

Seriously, I would realy like to hear the answer to the Ageia chip implementation for BOB question. Possible or no? If not now then later?

I'll settle for an unofficial answer, even an educated guess.

Has this already been addressed?

Just curious.

Thanks, D66.

05-18-2006, 03:06 AM
Oleg stated quite early, that they'll consider implementing support.

IIRC it was stated here, not long ago, that this case is still not decided.

And yet another time, please, be patient - we were promised weekly dev-updates, like we've seen them before. We'll get the answers, when Oleg thinks the time is right.

btw: I hope we'll get Ian's warstories for the BoB-dev-updates. http://forums.ubi.com/images/smilies/10.gif

05-18-2006, 03:27 AM
Oh no, yet another mandatory piece of expensive hardware...

05-18-2006, 07:04 PM
Originally posted by Vrabac:
Oh no, yet another mandatory piece of expensive hardware...

no . a possible optional piece of hardware that would give you graphics options otherwise impossible.

If you do not have it do not enable the extra graphics and you are no worse off than in a game without the option.

I do not agree with limiting the potential of a game just so low end user can get a warm fuzzy feeling.

Maddox games could of course release a version of the game where all the better options are disabled (so effectively everyone plays "medium") to avoid complaints and then sell a "version II" a year later when everyone has better machines.

Or they could sell the game with all its capabilities and people with low end machines could just be sensible and choose "medium" settigns and not complain.

Games have high end options for the enthusiasts that like to invest there money in a better machine. I do not see how removing those high end options so people with low end machines will feel better egowise actually helps anyone.

05-18-2006, 08:03 PM
Except if the physics chip provides merely non-interactive eyecandy that doesn't meaningfully change the sim, what's the point of a dedicated physics card? Or do you suggest setting up PhysX servers and non-PhysX servers? A different class of servers for each class of physics chip, perhaps, since the $300 cards can't show clouds as thickly as the $500 cards?

At some point or another Ageia's physics card HAS to be mandatory if it's to succeed. I can't see a $500 card that adds more chunks to explosions as selling well, and online games simply can't have everyone running around with different physics.

05-18-2006, 09:40 PM
Ageia does not do fluid dynamics.
It would be pointless to use for a flight sim.

05-19-2006, 12:15 AM
Hi Gotoaway,

then have a look here at the bottom of the page: http://www.ageia.com/physx_in_action/tech_demos.html
Apparently fluid physics are included, even when the precision might not be suitable for flight simulations. But I have no more details about that.


05-19-2006, 12:21 AM
I think that is the main problem - most people don't even know, what this chip is good for, including developers.

Okay, there are more and more FPShooters supporting the chip, but that doesn't mean it's for eyecandy. In GRAW or UT2007 it will be used to interact with your surroundings, move things around, etc. Collision, weight, gravity and other forces is, what the chip can calculate.

The chip can be usefull from plane-behavior (e.g. landing) over bomb-drops and effects on plane and the bomb itself (skip-bombing, for example) up to eyecandy like damage visualization (parts coming off your opponent), smoke (moving with wind and being affected, for example, if a plane flies through it) and explosions.

Coming from IL2 and LOMAC, just imagine a real looking fire on your machine or real smoketrails and contrails that dissapate considering altitude, speed, wind and temperature.

Another nice feature could be bullet behaviour and hit-effects, even propwash, maybe.

05-19-2006, 12:45 AM
Originally posted by GoToAway:
Ageia does not do fluid dynamics.
It would be pointless to use for a flight sim.

No, it doesn't do analytical fluid dynamics, blade element theory, or whatever other professional application you can think of. By default, at least -- it's probably programmable enough to accellerate those kind of programs anyway.

That's irrelevant, though. Those kind of applications are not in any way useful for real-time calculations in a flight simulator, and won't be for at least ten years.

What the PhysX chip DOES do is simulate fluids as systems of up to several hundred thousand particles. It's not accurate enough for computational work, nor is it intended to be, but it's certainly accurate enough to draw fire, smoke, clouds, running water, or whatever other fluid effect you can think up. Particle-based water still looks a bit cr@p, though. (This censor is outrageous. Remember when it censored "cockpit"?)

So, if PhysX was used for the clouds, and you had a computer fast enough to communicate all the effects on the particle system to the physics chip every (half-) frame, communicate the updated particles to the video card every frame, and then render them (what, you expected the physics card to do all the work?), it would look very nice indeed. Whether those clouds would ever be reproducable on each machine in a multiplayer game is a whole different matter.

05-19-2006, 03:27 AM
Originally posted by NonWonderDog:
At some point or another Ageia's physics card HAS to be mandatory if it's to succeed. I can't see a $500 card that adds more chunks to explosions as selling well, and online games simply can't have everyone running around with different physics.

Exactly what I thought. It's not mandatory now, but it probably will be. And it will just add more weight to the price of a complete configuration for playing games. And I just can't force myself to like it.

05-19-2006, 12:12 PM
Does the physx chip have a significant advantage over a multi-core processor? All the demos I've seen have been fairly simple particle physics. I'm curious how performance would compare to a physics thread running on a dedicated cpu.

The thing about the aerodynamic methods that NonWonderDog mentioned is that the physics isn't all that hard. Its relatively easy to setup the problem, boundary conditions, etc ... Its the math that kills you. If the chip could calculated a pde 60-70% faster than can be done on a current cpu, I would buy one today. These (http://www.grc.nasa.gov/WWW/K-12/airplane/nseqs.html) are the equations that describe aerodynamics. They are nasty, mean and have given me a lot of head aches.

The only way I can see how a dedicated physics co-processor may be successful:
<LI>The physics chips have a significant advantage over a normal processor. <LI>There is an open architecture (like OpenGL) that provides a flexible and common abstraction layer between the hardware and software.

btw NonWonderDog:
X-Plane (http://www.x-plane.com/about.html) uses blade element to calculate the flight models. Probably not all that accurately in an engineering sense, but enough to trick your mind to think its an aircraft.

05-19-2006, 04:15 PM
I recently saw some of the numbers for a new driver for the card and I was less than impressed. The comments where "yeah there is more stuff going on but it made the system crawl"

I think there is some time before this tech is fully flushed out. Right now doing the Physics with the CPU is faster. It's a combination of how everything is interacting together. CPU PCPU GPU/VPU

05-20-2006, 08:20 AM
Ahh, I didn't know X-Plane worked that much in real time. I just looked on their website, and it appears that they divide each wing into ten parts and compute accellerations from blade element theory based on the airfoil cross-sections specified in the plane builder. I'm pretty sure I knew this at one point, I'd just forgotten how they implemented it.

Still, I'm not sure you can actually have a formation of planes flying like this in X-Plane, and they don't use this analysis for propellers. Blade-element theory was made for propellers, so that was the implementation I was mostly thinking of.

05-22-2006, 04:52 PM
Me too, probably a natural part of any sim rig in the future. Question is if Oleg trust it or how much of the sim can be HW accelerated with this tech.

05-22-2006, 05:46 PM
They better have an open window for Physx or we may end up with something like the 6DOF for the Track IR in this game.

05-25-2006, 02:21 PM
Wouldn't you need to be running at least two graphic cards?

Where would you place it if there's no available slots left on your MB?

05-25-2006, 11:44 PM
Originally posted by Blood_Splat:
Wouldn't you need to be running at least two graphic cards?

Where would you place it if there's no available slots left on your MB?

The one review I could find said the "extra stuff" generated by the chip meant on their test machine they needed SLI to get reasonable frame rates ... but that is not to say that the next generation of graphics cards will not work fine with just the one graphics card. Also bear in mind that there are graphics cards coming out that will SLI two graphics processors on the same card in the one PCI-express slot.

As far as the Ageia physx card goes .. the current versions are old style PCI not PCI-X so you just need one standard old style PCI slot free.

The unit takes up a standard PCI slot

05-26-2006, 12:23 AM
As far as the Ageia physx card goes .. the current versions are old style PCI not PCI-X so you just need one standard old style PCI slot free.
Really? That's just... stupid. If PCI hasn't been fast enough for a one-way graphics card since the VooDoo2, how on earth can it be fast enough for a two-way physics card that needs to know the locations and forces upon tens of thousands of rigid bodies or hundreds of thousands of particles? I mean, the five (rather archaic) 32-bit PCI slots on my motherboard have a 133 MB/s total transfer rate. My AGP 8X slot has a 2 GB/s dedicated transfer rate, in comparison, and PCIe slots have an up to 4 GB/s dedicated transfer rate. Installing a PhysX card alongside a sound card would likely bring my computer to its knees, even if I HAD the graphics power. It would be even worse if I didn't have a built-in ethernet port (or is that on the PCI bus, too?).

Oh, and just to avoid confusion, PCI-X (PCI-Expanded) != PCIe (PCI-Express). PCI-X was mostly a professional server thing that found its way into Macs, and it's being phased out in favor of PCIe just like every other variant of parallel PCI. The difference is actually kind of important, since PCI-X is backwards compatible with PCI; PCIe isn't.

05-28-2006, 07:29 AM
The survival of the PhysX chip depends entirely on the FPS genre as it is the biggest by far. If they take it on board for extra graphics goodies or for gameplay or whatever then it will succeed and fly and be replaced by something similar but much better is short order (compatible though!)

It would appear that extra physics power is coming in 1 form or another either on
1. dedicated CPU core, since there will soon be 4 on home PC's

2. GPU, seems less likley as modern games are crippling even high end SLA/Crossfire configurations. In this scenrio the GPU (or plural) are barely capable of the graphics much less getting extra workload with physics. See reviews of oblivion and fear for framerates of top end rigs.

3. Dedicated physics processors like PhysX. This needs to be widespread before devopers will fully utalise it, of course the developers need to have made full use of it before people will cough up for it!! Catch 22

I suspect that option 3 will win in the long run as physics problems are not what CPU's are good at, but maybe it will be option 1.

The current limited support for the PhysX chip seems a little disappointing but it is early days. See http://anandtech.com/video/showdoc.aspx?i=2759

As to the benefits for us in BoB I really have no idea!!

05-28-2006, 01:58 PM
Ah yes, I can remember buying a math co-processor for one of my earliest machines... (386dx66?).

Those were phased out fairly quickly... after the tech was integrated.

I'm going to patent the integrated cpu/pcpu and beat everyone to the punch! LOL http://forums.ubi.com/groupee_common/emoticons/icon_biggrin.gif

LMFAO, you're all my unwitting witnesses for international patent law... sweet!

I'll get insanely rich and start a global empire and then you will all bow before me!

Muuh uuh ahh ahh ahh!!! ;p

D66 out.

05-28-2006, 05:42 PM

06-05-2006, 01:16 PM
There seems to be very little hard information on the web about the PhysX chip... what little there is suggests it is pretty much like a old fashioned Cray - but on a chip.

So there are a number of vector floating-point processors that use data from local on-board memory.

How is this so different from a general purpose CPU I hear you ask? Well, a normal CPU has to get data from sloooow main memory once it runs out of cache lines. I suspect the PhysX chip tries to keep all the calculations locally in memory - not too difficult as long as you're doing the same thing over and over again. Then it hides the newly revealed bottleneck of the time it takes to do floating point ops by the vector pipelines.

So a commodity CPU probably reaches ~5% of peak power in real physics calculations because it's always trashing the cache. A good ole Cray could easily reach 40% or more.

diomedes33 has got it pretty right about the level of physics though. However, there's nothing to stop someone programming the PhysX to do anything you want, as long as you stay in local memory.

The real flaw is putting it on a slow bus. That means any data that is sent back to the main CPU has got to be low volume

Yet another contribution to a winding-down thread... http://forums.ubi.com/images/smilies/35.gif


06-06-2006, 05:10 AM
You may not need to buy a dedicated PhysX card at all...


06-06-2006, 10:14 PM
Yep, nVidia has the same thing now. The problem is that current graphics cards were never meant to write to system memory, much less the CPU cache, and are thus incapable of doing so. Any physics generated by the GPU would be graphical only, and would be invisible to the CPU. This is obviously sub-optimal.

New cards can be made with write access to main memory, but main memory is still very very slow when compared to CPU cache. It's not really the place to put lots of fast-changing physics data. Even if it was fast enough, you're still using your (second!) very expensive graphics card to do physics calculations instead of the graphics calculations that so many modern games need.