Toms Hardware has a couple videos that show for a minute or so the ATI physics acceleration.

These surely look impressive -- as complex real-time behaviors were shown using complex objects colliding and being moved in real-time by different kinds of "force fields" under keyboard control.

The system used 3 X1900XT cards -- two for Crossfire and one for Physics. But, the spokesperson said one X1600 or up card could do the physics. Moreover, one X1900XT could do both the physics and graphics. Obviously, fewer or lower-power cards mean slower physics.

ATI claims their solution is faster than Ageia PhysX hardware -- and it may be -- due to the far faster PCI-E connection (vs PCI). Of course, that remains to be seen.

ATI claims this is real and they are going to have systems in the relatively near future -- months and months -- not just weeks, I presume. The games are in development.

I won't even buy two video cards for one PC, never mind three. For those with more money than sense I guess it's cool. Imagine the heat and noise from running three X1900XTX.

I like the idea of using my old ati card for something other than a paperweight. I guess I could sell it, but since this option is coming along, I can just use it as my physics card.

But could you run 2 nvidias and a ATI for physX? I can see some head aches comming......

Well, if you have to have a crossfire motherboard, I don't know if you'd be able to sli the cards anyway. (I could be wrong) I would think that ati would do their best to not allow an ati card to do physics with an nvidia graphics card, realisically it would just be a driver/software issue that someone might hack anyway. I don't know, someone who knows more about this needs to chime in.

Reportedly, Nvidia is working on the same thing. So, at some point, one will be able to use Nvidia cards.

Today, the advantage of ATI is that only ONE card is needed to do BOTH graphics and physics (x1600 or higher card). Two or three cards are OPTIONAL for more power.

Whereas, Nvidia needs at least two cards in SLI to get a good result. That will probably change when Nvidia releases the G80 series of DX10 cards.

Still, both are in the Lab now. So we'll see what finally emerges from both ATI and Nvidia (reportedly later in the summer).

Well, the industry seems to look for new money sources
The GFX card sector doesn't give 30% p.a. growth anymore. So we are told to need Ageia Physics Cards.
It's so sick.
Better CPU usage and more efficient programming could use the cpu for tasks like that. But that wouldn't be financially smart, would it?

You're drinking the marketing Kool-aid, Allen. Let's see if one X1900 can handle a game like Crysis and also process its physics. And those physics are effects-based, nothing that's really going to change gameplay.

Got to agree with Mr. Reynolds. Judging by how even something like Oblivion brings all but the absolute top to their knees... Unless the next-gen cards are going to increase GPU horsepower by an order of magnitude, or at least 3-4x... there's no way one card can do graphics and physics - unless maybe you sacrifice detail or resolution to take some off of the graphics end.

Obviously -- if a game is strongly GPU limited and has low FPS to begin with, adding physics calculations could reduce FPS too much to be useful

Yet, some games are CPU limited -- thus, in principle, the GPU has some reserve capacity. In those cases balancing the load with the GPU may improve system performance overall.

It will be a game-by-game thing, I think. How much physics matters and how well it works will depend on the game developers -- some will do it right and many will do it wrong. Right now, I am interested in the possibilities when its done right

Erm, with all due respect Allen, what modern games can you name that are CPU limited? (F4:AF doesn't count) I can't think of anything, certainly nothing very mainstream. The trend has been - and as far as I can tell it will continue - to push hard on the GPU.

Now of course you can lower the resolution to make almost any game CPU limited, but who runs at 640x480 in the real world?

Lock On 1.1 is CPU-limited is most realistic mission compositions. It's only GPU limited in simpler missions.

How much of that is due to bad programming, though?

Bad according to who? If a given game doesn't run well on a given machine under a particular circumstance, is it bad programming? Is it a limitation of the design? Is it an unfortunate side effect of trying to do more than was probably wise to attempt?


The Jedi Master

In this case, who cares? Speedo asked for a modern CPU-limited game, and I gave him one.

But LOMAC is a bit of an anomaly, and it's probably heavily CPU-limited because of bad coding. If you're going to cite something as an example with which to defend a position, it needs to be a good example.

Like I wrote at the end of the Corsair memory article that went up today, the trend within overall PC game development is definitely putting the onus of performance on the installed graphics board. Sure, there's the occasional title that's more CPU-bound than anything else, but that doesn't take away from what was being claimed.

Gatticus..."I won't even buy two video cards for one PC, never mind three."

Well then you need to get with the program...this is the age of 2x2's and 4x4's. Two processors, 2 motherboards and 4 processors, 2 video cards in SLI or 3 vid cards (one extra) to run another gadget.

In fact, in less than a year from now having your system specs displayed at the bottom of your post with anything less than 2 vid cards will be considered to be a, "faux paux"!

You are apparently mistaken in that you think I'm defending something. I actually think this whole "GPUs also do physics" thing is a horrible idea. I was just trying to be helpful, that's all.

I agree that more and more things are GPU-limited and a faster video card is the best thing you can do. Hell, we've even seen not too long ago that something as old as Jane's F-18 is GPU-limited, although artificially (it is coded for single-pipe rendering, and software modifying its render path to use multiple pipelines boosted framerates from the 20s to the 80s on my system).

I didn't mean to sound curt or anything (which, rereading my post, it does sound snippy). I just think LOMAC is a bad example when saying there are games that're CPU rather than GPU limited.

And, yeah, I'm not real sold on putting more work on the graphics board. We're still very early in the era of moving from fixed function to programmability with GPUs and I'd much rather see the graphics board spending its cycles producing better visuals than playing a mixed, however small, role.

I think Crossfire is overkill and would not buy a card for Crossfire alone because it makes no visible improvement. On the other hand, physics acceleration makes a visible improvment.

However, assuming similar performance, I would rather pay $300 to $400 for an extra Crossfire (or SLI) GPU -- to do physics and, sometimes, Crossfire -- than $200 to $300 for a PPU that can only do physics.

Actually, that will not always be the available choice. Some developers are going with Havok FX which currently requires a GPU for acceleration. Others are going with PhysX which requires a PPU for acceleration. And, the games are ALREADY being appropriately coded for Havok FX and Phys X.

In the foregoing sense, the direction of the industry is already an accomplished fact -- the games and hardware are "under construction". So, arguments against hardware accelerated physics based on games available today -- and even regarding GPU vs PPU -- are moot. In the short run of a couple years, things are set.

So, over the next couple years, I think it will still come down to the game developers -- and how well they use/implement hardware physics. We'll see if developers have the desire and smarts to produce physics-accelerated games that make us want to buy the hardware -- and which hardware (GPU or PPU or BOTH).

In theory, they can make us want the hardware, I think. But, will they make us want the hardware?