Slashdot Mirror
Modders Get Nvidia's PhysX To Run On ATI Cards
stress_life writes "Following controversial allegations that Nvidia is cheating in 3DMark Vantage and Unreal Tournament 3 benchmarks, executives from Futuremark and Epic moved forward to clean any confusion.
However, the game was not over — enthusiasts from Israel ported PhysX middleware to run on ATI Radeon cards, achieving remarkable performance. Owners of ATI Radeon cards will be able to play PhysX games as well, such as Ghost Recon 2 and already mentioned Unreal Tournament 3."
Also the 48XX series ships with linux drivers.
Viral software licensing is not freedom, it is in fact GNU/Socialism.
Tom's hardware is crap. Try a decent review site: http://www.anandtech.com/video/showdoc.aspx?i=3341
And the even worse news for NVidia is some preliminary numbers for the upcoming 4870 X2 would indicate it will completely blow away anything NVidia currently has on the market.
There is no -1 Disagree mod. Slashdot.org/faq defines mod options. USE IT.
In terms of performance per $.
NVidia is still king of the hill in raw performance.
You just have to pay.
So this whole thing was kicked off by a column on the Inquirer? The same people who brought us the Rydermark "scandal"? The Inq has shown a blatant and consistent anti-Nvidia bias over the years, so why give this any credence?
Besides, the first question that popped into my head is one that is being asked a lot of places, but not answered: If accelerating PhysX on Nvidia's GPU hardware is cheating, wouldn't accelerating PhysX on Ageia's PPU hardware be considered cheating, too? Call me cynical, but I think AMD knows the answer to that, and would rather you didn't mention it, thank you very much.
This sig intentionally left blank.
The interface is a freely available SDK (for some uses). The physics is basically Newtonian mechanics (more in a moment). Physics for games are, first and foremost, an exercise in collision detection. The physics is simple. Determining collisions in a series of finite-length steps is the hard part.
Why I say that the physics is basically Newtonian mechanics, there is spring technology, although all spring technology in finite step simulations has errors (if you are not carefully, the springs increase in oscillation over time, instead of damping.) Chemical dynamics and quantum force fields are out. Classical force fields are included. The force fields operate based on propogation (distance, distance-squared, etc.) and other parameters.
The fluid/solid interaction is still being worked on, and fluids and cloth benefit most from hardware acceleration. Fluids use a number of points with mutual attraction/repulsion properties.
No sound properties.
Rope is emulated as a series of sticks with ball joints at the end, a flag as a series of springs with forces at points (cloth simulation is esentially a thing of springs), and the ball and spring, yes.
You left out an important question, which is the rigidity of objects other than cloth/fluids. The ball that deforms as it bounces. Currently, that's in the SDK, but I've not played with it yet.
Your ad here. Ask me how!
Um... it IS CUDA. Or rather, its an extension for CUDA.
From what I understand, nVidia took the PhysX engine they bought from Ageia and ported it to their own language (CUDA) so that it would run on their graphics cards, so people didn't have to shell out for a second $300 "Physics Processing Unit", thus boosting nVidia's GPU sales.
And now someones ported it to ATI.
*Nelson Laugh*
They haven't done so because they're subscribers to PhysX's competition - Havok.
AMD / ATI / Havok
vs
Intel / nVidia / PhysX
Pick your side!
(Ok so it doesn't quite work like that but dividing battle lines evenly makes it less confusing than it really is)
Yes, I did misread your comment. Nevertheless, most of my comment still stands. A 4870 in Crossfire performs significantly better than the X280 and the 9800 GX2 every benchmark I've seen except Crysis, and these cards also have the capability to be run in a quad Crossfire mode. Oh, and two of them sell for less than one of NVidia's top dogs.
http://www.bjorn3d.com/read.php?cID=1301
http://www.pcper.com/article.php?aid=581&type=expert
http://techreport.com/articles.x/14990
There is no -1 Disagree mod. Slashdot.org/faq defines mod options. USE IT.
Close, but off...
AMD/ATI vs. Intel/Havok vs. nVidia/PhysX. At least, Intel licensed code from Havok. Intel wants physics on the CPU, nVidia on the GPU, and AMD/ATI just wants to be able to use both.
Your ad here. Ask me how!
Is it real physics? Would it be good for say simmulating chemical dynamics with quantum or classical force fields? COuld I use it to model the hydrodynamics of a sail boat cutting through the water?
No. Most physics middleware provide a simplified model (collision detection, rigid body physics, etc...) which is great for visual gimmicks in games, but is too much an approximation to be used in research. You would need other engines which are optimized to do accurate physics modelling - Gromacs comes as an example.
Now about the hardware behind this : Ageia's PPU could in theory be used to accelerate research calculation. The problem is the lack of a proper API. This processor has only PhysX as available API which is specialized for gaming oriented physics. The SieveC compiler is supposed to be able to generate parallel programs for the PPU but hasn't been released publicly.
Whereas, even if the GPU port of PhysX is only oriented for gaming-specific applications, ATI Radeon card also expose the much more general purpose API "Brook+" (the usage of which is already demonstrated in Folding@Home) and nVidia card have CUDA that you know.
Unlike PhysX, those API expose generic numerical methods and can be used to calculate applications as diverse as you mention. Including calculating the game-specific Ageia PhysX.
PhysX is to CUDA what, for example, Gromacs could be compared to Fortan. The first is a specific engine which is optimised to solve some very specific problems, the second is a general purpose language that can be used to crunch numbers.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
That depends. Unless a game is amazingly well optimized for a specific card, it will not be able to use 100% of the resources. If running physics on the GPU lets it use those untapped resources then it can only be a good thing.
That's pretty much what a physics engine does, and there are already a number of open source physics libraries out there (ODE and Bullet are the most well supported as far as I know, the former has been used in a few big budget commercial titles). Someone just needs to port the back-end to CUDA and off we go... Easier said than done, I reckon.
I recall hearing chatter about CUDA bindings for Bullet but I'm not sure if anything came of that.
The PhysX system doesn't really care about heat or energy. It primarily concerns itself with force and momentum. That, as I understand it, is the principle difference between Newtonian and LaGrangian mechanics.
Your ad here. Ask me how!
My Sapphire Radeon HD4850 disagrees. I'm a reviewer and always use the latest drivers. There were no Linux drivers on the Sapphire driver CD.
Nope. Not quite :-)
Lagrangian Mechanics gives you a lot more flexibility in terms of your coordinate system, and tends to be much better for solving systems with many interacting forces. It's essentially a mathematical re-formulation of Netwonian Mechanics.. The underlying laws are all the same, but the math used to arrive at a solution is quite different.
Of course, this is all for solving problems analytically. Computers most likely do things differently.
-- If you try to fail and succeed, which have you done? - Uli's moose