[techeye] High end kepler -- 2013?

[Joseph F]

Unfortunately you didn't read it right. Non-NVIDIA processor architectures means ARM and x86. CUDA LLVM is closed source and NVIDIA has complete control over what architectures it can compile to.

What a shame. I would really like to see CUDA on AMD GPUs.

 

[boxleitnerb]

there is a reason that almost no company uses physX, bullet or havok...

it's easyer to code a decent phisics in the game engine, than build a game engine around havok, for example.

Why is that? You would have to do everything from scratch. And even then, why is nobody doing anything like this? And what can be done to improve this situation?

 

[blackened23]

Unfortunately you didn't read it right. Non-NVIDIA processor architectures means ARM and x86. CUDA LLVM is closed source and NVIDIA has complete control over what architectures it can compile to.

Unfortunately this is a no win situation for nvidia. They want to make it a value added feature for NV cards, but it is not impressive at all (IMO). It has a huge performance hit in metro 2033 and batman: AC, and generally is not usable on single card configs. It is also not impressive, newspapers flying around here and there in Batman, who cares. The other issue is, there is typically only 1 game per every 8-10 months that support it, so its not very value added.

Now if they made it open source , that would change - we could see multi platform games and console ports support it. So NV can make it more widely adopted and not such a joke, or they can keep it a value added NV only feature. Damned if you do, damned if you don't (IMO of course)

 

[VulgarDisplay]

What a shame. I would really like to see CUDA on AMD GPUs.

OpenCL is going to make CUDA obsolete in short order simply because it does work on AMD and Nvidia gpu's.

 

[GaiaHunter]

You know you could say that about almost every single graphical effect, right? Think further how that could impact gameplay. You can spot enemies by the movement they cause in fog. There could be water puzzles where you have to distribute a certain amount of water into cups to trigger some weight plates. Or having a tsunami wave that hits. Possibilities are manifold if you let your imagination fly.

And then there is also immersion, the ability to create not a realistic but believable world where you don't immediately think "oh man, this looks so ridiculous, well it's just a game"

I like the water cups example - have you never seen pressure plates in games?

Or what about the fog? Do you know in real life you use both fog and smoke for cover?

I wish I could detect cars in the fog watching its movement.

The tsunami wave - you can already cause a tsunami wave in games. You model it and then create ascript for it in game. Sure, it is going to be the same all the time, but so it would be using a fixed variables physics engine. Maybe use an engine with random variables, but why not random scripts of the results?

The difference is the ability to change it on the fly. Being calculated on the fly or scripted based on beforehand calculation has the same visual results

 

[boxleitnerb]

Yes to both, but I have never seen this in games. That's my point.

 

[GaiaHunter]

Yes to both, but I have never seen this in games. That's my point.

You have never seen places where you press and something happens in games?

 

[quest55720]

I am sure the answer is very simple that high end Kepler will be late. The die is very large with a lot of transistors and the fabs just are not up to it yet. When the fabs work out the 28nm issues I am sure we will see high end Kepler. Until then we will get mid range Keplers. Its the price NVIDIA is willing to pay for its large die strategy. It sucks for people looking for high end cards right now. But at the end of year it should be a good time for people who buy high end cards. A high end kepler vs a refreshed 7970 with much higher stock clock speeds.

 

[boxleitnerb]

You have never seen places where you press and something happens in games?

Maybe you didn't understand my example. You have a certain amount of liquid that you have to put into several containers. If the fluid were correctly simulated with weight and fluid dynamics, you could actually spill something, thus rendering the riddle insolvable. Or you can experiment, filling liquid from one container into the other until the combination is right. On top of that, it would look realistic.

Mostly, these things are for visuals, that's for sure. But you can incorporate them into gameplay and use them to enhance immersion. So why not?

 

[Olikan]

Why is that? You would have to do everything from scratch. And even then, why is nobody doing anything like this? And what can be done to improve this situation?

look at bf3, crysis series....all have good phisics

pretty much nothing can be done about it... this article may help you to understand.

yes, the article says that direcX, but the same aplies to physx, bullet and havok

 

[Gloomy]

It's over! Nvidia is finished! They should just code the drivers for silicon AMD engineers, then we'd have the best of both worlds :hmm:

 

[Concillian]

Why is that? You would have to do everything from scratch. And even then, why is nobody doing anything like this? And what can be done to improve this situation?

Consoles are a big part of the reason PhysX will never gain traction.
- You have to have some other physics coding for the 60+% of your customers using consoles,
- You have to have physics coding for the 10-20% of your customers using AMD hardware.
- You have to have physics coding for the ~10% of your customers that have nVidia hardware that isn't powerful enough to run physics AND decent graphics.

What are you left with? Using PhysX for <10% of your customers? The game engine has to code physics with something else whether it uses PhysX or not. No matter how you look at it, PhysX is additional work for developers. The only way it will ever gain traction is if it's available in one of the major consoles so then it's at least over a third of the potential customers.

Before the console explosion, PhysX had a chance. These days? There's no chance for PhysX if it remains as closed as it is now. How do you possibly convince a developer that they will make back their programming time investment on something that applies to such a small percentage of their customer base?

 

[SolMiester]

The real truth is that Nvidia waited until they got a 7970 to test before actually doing much of anything. AMD is actually the leader in GPU's, Nvidia is a follower.

Pretty much the same thing as when the 5870 was launched.

LOL, i cant work out if this guy is serious or not?....

 

[SolMiester]

Did you actually just try and say than nVidia developed SLI? Because they most certainly did not. Nor were they first on the scene with it making that point kind of moot. They did have multi-GPU before ATI, but thats not because nVidia spent all this time developing it. They simply bought the rights and the engineers from 3dFX.

I dont believe NV SLI was ever the same as Voodoo SLI....so the did develop it, just with Voodoo developers!

 

[wahdangun]

Maybe you didn't understand my example. You have a certain amount of liquid that you have to put into several containers. If the fluid were correctly simulated with weight and fluid dynamics, you could actually spill something, thus rendering the riddle insolvable. Or you can experiment, filling liquid from one container into the other until the combination is right. On top of that, it would look realistic.

Mostly, these things are for visuals, that's for sure. But you can incorporate them into gameplay and use them to enhance immersion. So why not?

so tell me what is more realistic BF3 or Batman:AA ?

i mean i can't even make a hole in batman:AA. but i can destroy the building, the tree in bf3.

 

[GaiaHunter]

Maybe you didn't understand my example. You have a certain amount of liquid that you have to put into several containers. If the fluid were correctly simulated with weight and fluid dynamics, you could actually spill something, thus rendering the riddle insolvable. Or you can experiment, filling liquid from one container into the other until the combination is right. On top of that, it would look realistic.

Trine 2 for example.

It uses a physics engine where water can bend plants under its weight and steam can be redirected.

While GPU accelerated physics can provide higher precision to real time physics simulations (in real time without killing frame rate or so one expects), the question lies in if it necessary such high precision or if an approximated result is enough. If you hit an object with a vector ->, you expect it to go follow the -> direction. If the vector as an angle (and depending of the absolute value of the vector) you might expect it go upwards and forward. An approximated physics engines will show that, an higher precision physics engine might it make go at slightly different heights and fall off at different distances, while taking account of wind or whatever. So, you can either use a high precision engine, or create several scripted animations when the collision of two objects occur. The higher precision one might give more variability but the other one will be a decent approximation.

The problem is, why don't games actually incorporate more physics in the first place?

If all the games were using every little bit of physics they could scrap it would be much easier to understand the need to require more performance, but when most games barely scratch the potential of what you can do now, the problem seems a bit odd.


I guess that is the reason that make some people believe GPU accelerated physics is required to have physics (or more realistic ones) in game.

 

[boxleitnerb]

so tell me what is more realistic BF3 or Batman:AA ?

i mean i can't even make a hole in batman:AA. but i can destroy the building, the tree in bf3.

I don't want to generalize here. Obviously PhysX (or GPU accelerated physics via OpenCL for that matter) only improves select effects. However, it is capable of delivering more realism across the board, the devs just don't use it. The PhysX destruction module would enhance BF3 big time. There are some great videos out there and it looks really good.

I think we just don't have the GPU power to have PhysX (or similar tech) applied to a whole game, to every aspect of it.

 

[GaiaHunter]

so tell me what is more realistic BF3 or Batman:AA ?

i mean i can't even make a hole in batman:AA. but i can destroy the building, the tree in bf3.

That is because BF3 use a physics engine in a tree - yes you can use physics engines without them being GPU accelerated, but for some reason developers don't use the tools available often, but I guess they will go straight to complex tools without using basic ones.

 

[Kloreep]

I think we just don't have the GPU power to have PhysX (or similar tech) applied to a whole game, to every aspect of it.

I think Concillian points to the greater problem: a proprietary standard is self-defeating, as it cannot be relied upon for anything essential when it's guaranteed only a portion of the user base will be able to turn it on.

 

[boxleitnerb]

I said "similar tech" and with that I meant OpenCL and another middleware of course. But aside from PhysX, there is nothing noteworthy at all out there, open or not.

 

[wahdangun]

I don't want to generalize here. Obviously PhysX (or GPU accelerated physics via OpenCL for that matter) only improves select effects. However, it is capable of delivering more realism across the board, the devs just don't use it. The PhysX destruction module would enhance BF3 big time. There are some great videos out there and it looks really good.

I think we just don't have the GPU power to have PhysX (or similar tech) applied to a whole game, to every aspect of it.

that why GPU physic is a joke what ever it come from nvdia or AMD. i mean if the game is more demanding then GPU physic will get in the way unless you like to settle to lowered graphic option, i mean why we need GPu to do more work beside rendering graphic, i know we need some degree of realism but by doing it on GPu is absolutely ridiculous while our CPU getting faster and more core doing nothing.

and when the time we hit the ray tracing the GPu will not have any spare performance to do Physic calculation.