Topic: Tell me again why we need a physics card?

[idolminds]

Havok 4.0

[Xessive]

Wow, that was pretty sweet.

I think the logic behind the physics card is to remove the stress from the CPU, and let a separate unit process all the physics data. Unfortunately it fails, and in most cases the separate phyics card actually degrades performance.

I like the idea of having a unique, separate chip to handle the physics, but I think it should be a part of the CPU, not a whole separate card.

[idolminds]

Thats pretty much what they are going to do with the multicore CPUs. Dedicate a core to physics.

Though this video seems to use the GPU. If you listen to the guy, esp the part about half way through where the soldier is walking through the debris and kicking it around, and the wind blows it around, he says that the debris exists only on the video card. Its was a graphical effect, nothing more.

[Cobra951]

I don't believe that you need any additional hardware to do what we just saw.  Physics is math.  Math coprocessors are already part of any decent PCs.  And now the trick is to stack CPU cores to keep adding performance where the GHz are getting harder to come by.  So dedicate one or two of them (quad ones around the corner, right?) to help along as well.

I remember when 3D graphics cards started to appear.  The technology of 3D games at the time was to compute and rasterize 3D into 2D with the CPU, into very limited resolution and color space.  It was instantly obvious then that offloading 3D transformations and 2D projections and rasterization into a separate computer on a card was the way of the future.  Seeing something run smoothly in thousands of colors at a high res was the proverbial picture that's worth 1000 words.  It's so dramatic because it really is a handoff--after the 3D computer on a card gets it, it never touches the PC bus or CPU again.  The input to the card gets processed by the card, and goes straight to its own independent output, the screen.

This business of physics cards does not in any way come close to that feeling of immediate need, and can't be truly separate from the PC side, since the results of the physics functions have to be returned to the running app code anyway.  Unlike "create a display from these data and don't bother me", a physics engine is an integral part of the mainline code.  Its results affect the mainline events, so it can't be handed off and ignored thereafter.

I just don't see it, other than as someone's wet dream, being pushed by their propaganda machine with all effort it can exert.

[gpw11]

Earlier in the demonstration (when walking through the steam) the guy said it wasn't using any cpu power at all, so we can either assume it's using a physx card or is done entirely through shaders with the gpu.  Eitherway, you're going to have to pay to get these kinds of physics effects.  The real question is which will be more cost effective (two different cards, hybrid card, or a really really powerfull gpu capable of doing this without taking a performance hit with the effects that will become commonplace in future games).

[Quemaqua]

I second what Cobra said.

[MysterD]

All we need is a Physics cards to get popular....

Then, they're will be separate cards for every damn game trick you can come up w/ -- one card for physics, another for shadows, another for shaders, another card for AA, another for AF, and the list can go on and on and on.

Skip that crap.

I don't see Physics cards taking off at all.

[Xessive]

Here's an interesting article I found on Gamespot:
Ageia's Man of Steele

Think faster CPU and GPU's are the answer to better-looking games? One company thinks they have a better idea. Meet Ageia's Michael Steele.