News Your games might get smarter...

[Firehed]

Originally Posted by wafflesomd

Well hey, the PPU sold really well, I'm sure this will follow

I'd hold off until UT07 comes out before saying that... or really any hot-item game that supports the thing.

That said, this thing really seems like a solution looking for a problem. FEAR's AI is incredible, and it's pretty darned impressive in SC:CT as well (though given that it's not really FPS, that gets largely overlooked).

If it comes out on PCIE for a fairly low price with decent adoption rate, it's got half a chance. A big if.

[will.]

i was all excited about the physx, silly name n all, but tbh, these seperate cards for seperate processes are not going to become popular. There is only so much you can stuff in one pc, and the money is obiously a factor, like already mentioned, the dual core (soon quad core) can handle it and eventually, i'm predicting the all in one chip that will just do everything and probally be cleverer than me.

[Sinner666]

Goody! Another add on card.

Can you even fit quad core graphics, PhysX, AI Card, X-Fi Pro and TV Tuner all on the same board?

[aggies11]

Yep, FEAR had good AI, and didnt' need a special chip.

Is the problem of AI in games, one of "not enough horse power", or "not enough development resources devoted to it". Ie. Can current games "do" good AI, but they are just too lazy to implement it?

AI is tricky, ignoring all implementation questions, just comming up with a good Behavior Model for your NPC's is a tricky business. That doesn't mean it can't be done, just that it's difficult to do right, and most developers typically don't put in the effort (and maybe prefer allocating resources to something else, eg. graphics).

Unless the SDK makes it easier (read: faster, cheaper) to make good AI, no matter what the hardware can do, it likely won't get used :/

Aggies

[Rich_13]

Originally Posted by ch424

I wish these companies would just produce IP cores and sell them to nVidia/ATI. Then we (as consumers) wouldn't have to worry about where we're gonna plug in our extra physics/ai/sound/network cards...

ch424

exactly what i was thinking. Prob get quite a bit as well if they started a bidding war. Intel might prob want a peice of that to spite the ATI/AMD love nest. lol

[EQC]

Originally Posted by otispunkmeyer

but seriously though why bother, we have dual core chips now, that in al honesty arent that expensive, and very few games make use of the second core.

quad core is just around the corner too, so even with new games there'll still be 2 cores left doing nowt... why cant they just do the AI?

Didn't the article say the new chip can do AI calculations 200x faster than a CPU? That would mean that even if that "200 times faster" was the result when compared to a single core, 500 MHz CPU, then using 3 cores of a 4-core CPU running at 3GHz would still be 15 times slower than using a specialized AI chip. Think about that. Then think about how much that 4-core CPU will cost. If history tells us anything, the 4-core Kentsfield will start out well over $1000, while an AI chip (if it's like the PPU) will be around $300 including it's own RAM supply, etc.

Also: better AI doesn't necessarily mean "harder enemies," it means more realistic action with computerized enemies. They can still be easy or hard to defeat depending on your settings, they'll just behave in a more human-like fashion.

And while there's plenty of arguments about why you won't buy a separate add-on card, chances are it'll be a long while before any add-on cards are actually required. At best, they'll be optional if you want the more realistic physics/ai/etc. Even after, what, 10 years of availability, graphics cards still aren't required per-say...they're just there if you want the more realistic graphics. Same story for the PPU and AI -- but it's still in the 'new and scary' stage for most people.

Eventually, every add-on card (graphics included) will likely find they're way back inside a multi-core super-cpu...but until then, why complain about the option of making things more realistic for a bit more money?

[cebla]

The thing is that in something like FEAR they probably went through all the levels putting way points saying here is a good spot for cover and such (I don't know that for sure). If this can do it without having to put that effort in then it will be much better

[Tyinsar]

Originally Posted by EQC

Didn't the article say the new chip can do AI calculations 200x faster than a CPU?...

Interesting theory but calculations may be just a little off - the CPU only spends a fration of its time on AI and this chip would only take care of some of those caculations (mainly path-finding from what I've read). It's like if you could get out of your driveway at 3000km/hr but drove at normal speeds the rest of the way to work or school, how much quicker would you get there?

Originally Posted by EQC

Also: better AI doesn't necessarily mean "harder enemies," it means more realistic action with computerized enemies. They can still be easy or hard to defeat depending on your settings, they'll just behave in a more human-like fashion.

That part I agree with.

Originally Posted by aggies11

Yep, FEAR had good AI, and didnt' need a special chip.

Is the problem of AI in games, one of "not enough horse power", or "not enough development resources devoted to it". Ie. Can current games "do" good AI, but they are just too lazy to implement it?

AI is tricky, ignoring all implementation questions, just comming up with a good Behavior Model for your NPC's is a tricky business. That doesn't mean it can't be done, just that it's difficult to do right, and most developers typically don't put in the effort (and maybe prefer allocating resources to something else, eg. graphics).

Unless the SDK makes it easier (read: faster, cheaper) to make good AI, no matter what the hardware can do, it likely won't get used :/

Quoted For Truth (especially the last line)

[stevie1556]

Did anyone play UT2004 in pratice mode? There was an option in the game that made the computer oponents match your skill level so you always improved. As you get better, the computer gets better, then when it starts kicking your arse it starts to play more noob-ish so you can catch up.

As far as I know, UT2004 was the first game to incorporate that feature. It made playing the game on your own feel more like playing it online.

[nemesis80]

seems they are trying to follow a model similar to the cell architecture, but making it modular?...

[metarinka]

I don't think this is a problem of processing power per se' we've been working on AI for what 20+ years now and things still aren't passable. I think the problem is not so much processing power anymore but an exponentially rising wall of difficulty to climb. Luckily for NpC AI in video games the challenge is much less than say your Irobot personal assistant type. But the wall is still unsurmountable.

I used to make maps in my spare time for UT2K4, pathing and AI implentation on any map was the last and usually most convoluted stage. A lot of the time the AI is dumb, it's just scripted to smart behaviour as in all the secret spots and rocket jumps etc were programmed into the path's so the Ai would know when to turn down and fire a rocket at this location in order to jump to another cliff. Now add in another factor of difficulty (destructable terrain) etc and it's hard to program path's for the unexpected. So now we are onto conditions, say when Item X falls over if it's larger than our duck height use it for cover. but that doesn't preclude to all the situations such as the fact that you recently strafed just enough to hit him through the gap in item X.
I think the biggest gap to better AI program is the exponential increase in difficulty of programming. It's easy to make path a bot and have them appear smart, it gets harder to program in conditions to say duck or coordinate and it gets just so complex for them to factor in every condition that humans do.

sorry I'm rambling my point is that it's a battle of diminishing returns the more you try to get realistic Ai the more time it takes for diminishing returns. Same thing happens in the 3d world the more realistic a completely 3d model is the more ways the human eye tries to see if it's false

[DLoney]

man... lol, Im still saving up for that new network card...

[otispunkmeyer]

Originally Posted by EQC

Didn't the article say the new chip can do AI calculations 200x faster than a CPU? That would mean that even if that "200 times faster" was the result when compared to a single core, 500 MHz CPU, then using 3 cores of a 4-core CPU running at 3GHz would still be 15 times slower than using a specialized AI chip. Think about that. Then think about how much that 4-core CPU will cost. If history tells us anything, the 4-core Kentsfield will start out well over $1000, while an AI chip (if it's like the PPU) will be around $300 including it's own RAM supply, etc.

Also: better AI doesn't necessarily mean "harder enemies," it means more realistic action with computerized enemies. They can still be easy or hard to defeat depending on your settings, they'll just behave in a more human-like fashion.

And while there's plenty of arguments about why you won't buy a separate add-on card, chances are it'll be a long while before any add-on cards are actually required. At best, they'll be optional if you want the more realistic physics/ai/etc. Even after, what, 10 years of availability, graphics cards still aren't required per-say...they're just there if you want the more realistic graphics. Same story for the PPU and AI -- but it's still in the 'new and scary' stage for most people.

Eventually, every add-on card (graphics included) will likely find they're way back inside a multi-core super-cpu...but until then, why complain about the option of making things more realistic for a bit more money?

well yeah, but remember the physx was quoted at being many many more times faster than the CPU at physics, and lets see.... in the two games i know of that use it (GRAW and CoV) using the card is slower in GRAW, and faster in CoV but not many many times faster..... maybe 50% or a little more

http://www.anandtech.com/video/showdoc.aspx?i=2828

Even given the chance to cheat some, we still can't match the performance of using a PhysX card with a dual-core processor. The spread does drop to only 20%, but the setup using the PhysX card clearly takes the lead. It is worth noting however that the quality between these two modes is extremely close, even with the CPU using only physics approximations.

The highest quality mode still offers a superior amount of "stuff", and in turn the PhysX card does a better job of running that mode. The two modes are not easy to tell apart, though, and even less so in the middle of combat where such minor differences can easily be missed. If you compare the high-quality software PhysX scores on the dual core processor to the maximum quality hardware PhysX scores on the same setup, high-quality software PhysX ends up being slightly faster. Dropping the physics quality down a notch instead of buying a PhysX card is a perfectly viable alternative if extra performance is desired without spending any more money.



i jus dont see the need for it. anand even points out that running the Physics in CoV on a high quality mode which isnt the top mode (it uses approximations in the calculations) and just using CPU power theres bearly noticable difference in physics and while it is faster with the physx by 20% its still totally playable. I'd rather put the £££ toward a faster CPU than a physx because ill get more use out of it.

this AI card, if it ever comes to fruition, will just be the same.....lots of hype, bearly any support, and threatened by multicore CPU's and the possibility of doing the same stuff on GPU's

[yahooadam]

the thing everyone seems to miss when reviewing the physix card is not only do the physics calculations get offloaded, but there are many many more calculations when it gets offloaded, because they can do it

and then you strain the GFX card more, and of course the CPU still has to interpret the physix calculations

[DougEdey]

Hence why when the PhysX card is in use the framerate drops.
But it "looks" nicer.

No need for that in online games IMO.

[TheEclypse]

Originally Posted by EQC

Admittedly, you do have a good point wiht respect to solving hte chicken-egg problem...but isn't that going to increase the bandwidth and power and size requirement on the graphics card and possibly hurt graphics performance? With high-end graphics cards already requiring more power than high-end CPU's, and taking up so much space that an entire motherboard (slots above, below, and connectors near the slot) has to be planned around what giant card might come out next, seems to me like a separate card would be better. And don't forget that the GPU, PPU, and probably the AI-PU use their own ram, which takes up space on a board too.

Some good points there, didnt really think about it all that much. As for wanting to upgrade the processor..well, 200x that of a CPU will be fine, wouldnt want my computer becoming a sentient being now would I?

[EQC]

Originally Posted by otispunkmeyer

well yeah, but remember the physx was quoted at being many many more times faster than the CPU at physics, and lets see.... in the two games i know of that use it (GRAW and CoV) using the card is slower in GRAW, and faster in CoV but not many many times faster..... maybe 50% or a little more

Don't forget about Cell Factor (yes, it's just a demo version...and yes Ageia has a big part in it's production, so they could be tweaking things to their benefit...)

Specifically, I'd ask you to take a look at this Tom's Hardware article:

http://www.tomshardware.com/2006/07/...physx_failing/

and especially the images on this page:

http://www.tomshardware.com/2006/07/...ing/page5.html

note that in the original Cell Factor demo, somebody figured out how to run it without a PPU card...and didn't lose any framerate. However, what they didn't notice was that without the PhysX card, the most complex objects to model with physics (the fabric and the fluid) were completely missing.

In a newer version of the Demo (that didn't disable items when there was no PhysX in use), toms hardware ran an AMD FX-60 (a high-end dual core, of course) with a GeForce 7950GX2 card. On page 7 of the article:

http://www.tomshardware.com/2006/07/...ing/page7.html

they say this:

Originally Posted by tomshardware.com

However, when we switched off the PPU and ran the demo in software mode, we could not get near any of the simulated cloth. The second we turned the camera in the direction of the cloth and moved physically close to the hanging cloth strips near the tunnel entrances or the flag hanging on the upper deck, the frame rate would plummet and the game would slow to a crawl...

...Below is an image showing the dual cores of the AMD Athlon 64 FX-60 instantly being given large amounts of work to do. The frame rates suffer from a smooth data flow from the CPU. Based on what we have seen, the cloth simulation inside Cell Factor is brutal and will need hardware acceleration.

Admittedly, this is Ageia's demo...so there could be some funny business going on. And Fabric Simulation isn't exactly a selling point. But if the TomsHardware experience is indicative of exactly what the PPU is capable of relative to a Dual Core CPU, maybe Ageia's estimates of relative speed aren't so far off....and perhaps one day, the physical simulation of objects can be elevated (ie: more objects, more complex objects, more deformable objects) to the complexity that is already present in a fabric simulation.

So, what I take from this is that perhaps the other games (CoV and GRAW that you mentioned) aren't yet taking full advantage of what the physics chip can do...so of course you won't see much difference when you're comparing 1% of a PPU's capabilities to 50% of a CPU's.

Of course...whether or not the PPU will take off is still to be seen, and its capabilities may or not play a role in that...but it does seem that there could be a significant benefit to hardware acceleration of Physics, and by extension perhaps AI too.

[Veles]

Originally Posted by LoneArchon

Then there will be nothing for the cpu to do. With Intel and AMD moving toward muticore CPUs these cards seem more and more lacking value as the cpu now has mutiple core to share the Load. with quad cores coming out soon and muticpu support in games it seems like the bottle neck will crop up in other parts of the system such as the hard disks and system buses not in the cpu.

But according to the makers, you'd need 200 cores to beat this AIPU.

All these add on chips arn't just CPUs stuck on a board, they're designed to do a specific task so can blow a CPU out of the water when it comes to that specific task. CPUs are more or an all purpose number cruncher, but it's a jack of all trades but a master of none.

I think an AI chip is much more needed that a PPU, but I can't really see it working.

If I'm getting this right, this is basically like an AI engine, so the dev does a little tweaking but most of the stuff is already there, making developing AI much easier. But the problem is, to use the engine I'd imagine you'd have to have this chip. if you don't, your stuck with either no AI at all, or some really bog standard stuff.

I don't really see developers buying into this when they'd have to make a basic AI engine anyway for all the people without the chip.

I suppose this would really come in handy in games like supreme commander, where you have thousands of units, not just a 5 man squad.

[jezmck]

someone said that the AI enemies will act more humanly - that's not true, unless the game programmers have encoded human behaviour.
<geek>Vulcan behaviour is more likely tbh.</geek>

I'm sure there are non FPS-style games where improved AI would make a big difference, especially if they look into true heuristics.

[xion]

Every time i think about going back to a mATX or smaller, somthing like this rears its ugly head... smacks of retroactive evolution back to the days of expansion daughter boards... I am interested in the tech and can see its got legs (i'm such a geek, i want all the new toys) but i wont be buying into this until its matured, hopfully be then the progress of Gfx cards will be alligned with PPU and AIPU and hopfully merged into a single solution... Tho that could quite easily batter the bus!