Does PhysX help?

[gs29]

I want to know this to help make a decision between 4870 and 260. My understanding is that CPUs do ALL the physics in games currently. Since better CPUs seem to hardly have any effect these days (assuming one is using C2D ~3.0 ghz+), does this mean PhysX will have very little effect? i.e. physics can already be done at speeds fast enough that even faster physics processing has little effect on FPS. Or is the jump to faster CPUs, say, a C2D at 3.0 ghz to a C2Q at 4.0 ghz, different from a jump from CPU to GPU?

 

[pillagenburn]

well, if you have a quad core system, i don't see how physx could help that much... my opinion, though.

edit: assuming the game is optimized for multiple cores anyway....

 

[R1ckCa1n]

It sure does in 3DMark Vantage!

 

[wkarma]

Once games come out with physx hardware support (there are a few out there already), then yes, it will help significantly.

 

[kllrnohj]

Until GPU physics become common place enough for developers to take advantage of it, you won't see any difference. Physics still have a very long way to go (see: http://www.tsnstudios.com/Demo_Reel.aspx for an example of what I mean ). The reason you don't see much improvement is because current game physics are generally pretty basic. Hopefully, once GPU and other forms of accelerated physics (be it quad/octo core CPUs, even) start becoming more common place (or in the case of GPU physics, something that is actually released ), developers will start to take advantage of it.

However, this shouldn't affect your decision. Both the GTX 280 and the 4870 both have hardware that can do GPU physics, and unless the physics can be accelerated by both camps, you likely won't see anything spectacular or even desirable.

 

[bitgod]

I was kinda interested in physx, but now that I'm thinking of jumping to the crossfire ATI side, I'm not worried too much. UT3 is the only game that I'd play that would take advantage of it, and I'm not holding my breath on tons of support at the moment, support has been promised for a while. It might be more attractive to game makers if all future nvidia cards can do it, but I don't know how many are going to go out of their way to do it. And if AMD/ATI start doing stuff with Havok, well there's a lot more games that I play that use that.

 

[jebo_4jc]

There's something to consider: PhysX will NOT run on the same GPU with 3d at the same time. So, if you want to run PhysX, you'll need at least two video cards.

I also wonder if you couldn't run an nvidia 9600GT or similar for physx with an ATI 4800 gpu.

 

[Belgain Roffles]

There's something to consider: PhysX will NOT run on the same GPU with 3d at the same time. So, if you want to run PhysX, you'll need at least two video cards.

I also wonder if you couldn't run an nvidia 9600GT or similar for physx with an ATI 4800 gpu.

Could you run a X2 card with one core on graphics and the other on physics? I wonder if we will be seeing something like that eventually, or even an asymmetric card with just a strapped on slightly less powerful core for physics processing.

 

[GotNoRice]

There's something to consider: PhysX will NOT run on the same GPU with 3d at the same time. So, if you want to run PhysX, you'll need at least two video cards.

I also wonder if you couldn't run an nvidia 9600GT or similar for physx with an ATI 4800 gpu.

That's incorrect, you absolutely CAN run PhysX with just one videocard using Cuda.

And it doesn't even seem to detract from the GPU's graphics abilities much, at least not in 3dmark Vantage from the early benchmarks that have been done so far: http://www.hardforum.com/showthread.php?t=1317351

 

[kllrnohj]

There's something to consider: PhysX will NOT run on the same GPU with 3d at the same time. So, if you want to run PhysX, you'll need at least two video cards.

I also wonder if you couldn't run an nvidia 9600GT or similar for physx with an ATI 4800 gpu.

Where does this rumour keep coming from? It seems like it has surfaced recently... As the person above pointed out, that is completely incorrect (or it will be an artificial restriction by nvidia - there is no technically reason it can't be on the same card, as you can already do CUDA+rendering on the same card).

 

[jebo_4jc]

Where does this rumour keep coming from? It seems like it has surfaced recently... As the person above pointed out, that is completely incorrect (or it will be an artificial restriction by nvidia - there is no technically reason it can't be on the same card, as you can already do CUDA+rendering on the same card).

interesting
It was mentioned in a review recently...either a 9800GTX+ or GTX280. I'll try to find it.

 

[jebo_4jc]

Ah I can't find it.

please tell me I didn't read it on fud or inq

 

[jebo_4jc]

That's incorrect, you absolutely CAN run PhysX with just one videocard using Cuda.

And it doesn't even seem to detract from the GPU's graphics abilities much, at least not in 3dmark Vantage from the early benchmarks that have been done so far: http://www.hardforum.com/showthread.php?t=1317351

I think physx support in 3dmark is absolutely moronic (or, at least, physx performance adding to the final score). Further proof that 3dmark scores mean very little.

 

[vicaphit]

Nvidia is about to release drivers that enables physX on all 8 level geforce cards and higher.

 

[turumbar]

PhysX might help to decrease CPU load or add eye candy on some (very small set of) games, and the number of games that use PhysX for actual gameplay (i.e. fully destructable environments, thousands of individual objects flyng around) is almost nil.

 

[Yeorwned]

That's incorrect, you absolutely CAN run PhysX with just one videocard using Cuda.[/url]

Agreed.

 

[kllrnohj]

Of course, since 99% of the games out there are GPU bottlenecked, putting more of a load on the GPU isn't really such a great idea

 

[Yeorwned]

Well when you perform pyshics at the GPU level, it doesn't have to talk to the processor to do the math and perform the action, which reduces the bandwidth/gpu usage heavily. So believe it or not, doing it at the GPU actually reduces the load.

 

[jebo_4jc]

Well when you perform pyshics at the GPU level, it doesn't have to talk to the processor to do the math and perform the action, which reduces the bandwidth/gpu usage heavily. So believe it or not, doing it at the GPU actually reduces the load.

hahahaha

 

[turumbar]

Well when you perform pyshics at the GPU level, it doesn't have to talk to the processor to do the math and perform the action, which reduces the bandwidth/gpu usage heavily. So believe it or not, doing it at the GPU actually reduces the load.

Performing physics calculations on the GPU on top of rendering will most certainly increase load. Even if the GPU were somehow able to run a game without instructions from a CPU, the actual GPU core would still have a heavier load, given that it's shouldered even more computation.

 

[Yeorwned]

Hey, it was worth a shot right? Some people pick up on the light sarcasm better than others.

Advantages

CUDA has several advantages over traditional general purpose computation on GPUs (GPGPU) using graphics APIs.

* It uses the standard C language, with some simple extensions.
* Scattered writes – code can write to arbitrary addresses in memory.
* Shared memory – CUDA exposes a fast shared memory region (16KB in size) that can be shared amongst threads. This can be used as a user-managed cache, enabling higher bandwidth than is possible using texture lookups. See example here.[4]
* Faster downloads and readbacks to and from the GPU
* Full support for integer and bitwise operations

Limitations

* Texture rendering is not supported.
* Recursive functions are not supported and must be converted to loops.
* Various deviations from the IEEE 754 standard. Denormals and signalling NaNs are not supported; only two IEEE rounding modes are supported (chop and round-to-nearest even), and those are specified on a per-instruction basis rather than in a control word; and the precision of division/square root is slightly lower than single precision.
* The bus bandwidth and latency between the CPU and the GPU may be a bottleneck.
* Threads must run in groups of at least 32 threads that execute identical instructions simultaneously. Branches in the program code do not impact performance significantly, provided that each of 32 threads takes the same execution path; the SIMD execution model becomes a significant limitation for any inherently divergent task (e.g., traversing a ray tracing acceleration data structure).
* CUDA-enabled GPUs are only available from Nvidia (GeForce 8000 series and above, Quadro and Tesla[1])


Source: http://en.wikipedia.org/wiki/CUDA

 

[kumquatsrus]

check out this thread for a couple links listing physx support for games released and curently in development

http://www.hardforum.com/showthread.php?t=1316955

 

[ZerazaX]

PhysX is in an interesting situation right now because Intel aquired Havok and AMD joined Havok's side.

Both Intel and AMD are still first and foremost CPU companies and will do everything possible to ensure that Havok can be run smoothly via CPU. Furthermore, Intel is entering the discrete GPU market w/ Larrabee and future products and in the press release, they discussed the acceleration of physics for Havok via GPU's for ATI cards (and future Intel cards).

While PhysX certainly has a lot of games, the majority of those games use PhysX drivers but do not require hardware acceleration. And in the games where actual hardware acceleration can be used, mtos of the effects are small and mean nothing in the grand scheme of things and gameplay.

It's kind of a catch 22: If developers make a PhysX compatible card a requirement, they lose out on a lot of customers but they can get lots of actual physics in the game that can make gameplay interesting. However, if they make PhysX effects negligent, they can get away without needing a card that can hardware accelerate the goods but that also means negligent effects on gameplay.

Keep in mind that ATI as of right now is ~25% of the discrete GPU market whereas Nvidia is 75%. No game developer is going to alienate 25% of their potential market base by forcing PhysX hardware acceleration - that would just make the move to consoles that much faster. Furthermore, given how many console prots there are, games might not make physics hardware acceleration a big thing until next gen consoles are out where their GPU's will have physics acceleration.

And with Havok having big titles upcoming (Fallout 3 and Starcraft 2, 'nuff said), its clear that developers that want to make big titles want those titles to run on everything.

 

[tc93]

Its a proprietary graphic thing that Nvidia added to their latest cards. Which means no other cards can use it, including all the previous nvidia cards. So I doubt very much you will see any game company using this.

 

[trinibwoy]

Its a proprietary graphic thing that Nvidia added to their latest cards. Which means no other cards can use it, including all the previous nvidia cards. So I doubt very much you will see any game company using this.

Uh, no it's not. PhysX can and has been running on CPUs for quite a while now.

It's proprietary yes, but so is Havok. There is no open physics standard out there.

 

[cybereality]

There is no open physics standard out there.

Theres ODE ( http://www.ode.org/ ). I don't know if you can call it a standard, but its open-source. It was used in Call Of Juarez, but not too many other recent titles use it.