AIPU

Klintor said:
How can something so long...be so horribly written?

I will re-iterate what I said on the previous page: I see (somewhat) of a need for a PPU: physics are really intensive calculations, and could benefit from an actual card (though, I do hope to see it integrated on the video card some time in the future). For an AIPU, I don't think we really need one. As people have said before, it sounds like a solution for a need that doesn't exist.
Click to expand...

Not to mention that physics are based on a well known phenomenon. Physics. Where as AI, you just cannot make an all-purpose "AI" chip. AI is made up of clever programming, not fancy chips. The "AIPU" is a dodo even before its hatched. The best all-purpose chip you get for AI is a cpu or other general purpose number cruncher.
 
Exactly, Immacolata. Unless AI processing becomes somewhat standardized (like physics can be with things like the widespread Unreal Engine(s) ), I don't see how an addon card will help things more than a dual-core CPU.
 
I just want to chime in real quick, that there IS a standardised programming language for AI. LISP. And this is not the first time that hardware architecture has been designed solely to run LISP type instructions.

I don't know if the card will ever take off, but if a standard API is developed for AI, and this card will accelerate that API by a lot, then it may suceed.
 
What could possibly be better than the intergration of all the best aspects of MMOs and single player games. A world that behaves realistically, but is completley under you control. Of course, I doubt that we will able able to do truly lifelike AI on the consuemr level for some time, but even little (or in this case, perhpas a lot) helps.

Although I do have an extra core lying around that really does nto do much in gaming yet.... :)
 
spugnor said:
I just want to chime in real quick, that there IS a standardised programming language for AI. LISP. And this is not the first time that hardware architecture has been designed solely to run LISP type instructions.
Click to expand...

LISP is a nifty language, but dear god did I ever hate coding in it!!
 
Immacolata said:
Not to mention that physics are based on a well known phenomenon. Physics. Where as AI, you just cannot make an all-purpose "AI" chip. AI is made up of clever programming, not fancy chips. The "AIPU" is a dodo even before its hatched. The best all-purpose chip you get for AI is a cpu or other general purpose number cruncher.
Click to expand...
Rendering and Physics is not one single fenomena.
Ghapics and Physics and AI consist on different kinds of sub fenomena's. Like the lot of units a GPU have.
Physics is not one thing it are many. So a PPU is a bit of general parralel computing optimised to support hardware acceleration to a large part off the Physics fenomena.

A GPU doesn't consist on one type off parralel unit. It consist out off many.
Pipeline
ROP
Rasterizer
TMU
All those different type of shaders
RamDacs
Decoders

The shaders makes it be able to do more then Ghapics alone.

While AI is largly more a Desision making routine.
If what then this else that branching but not serial but very parralizing form.
For some off the lower level AI problems.
For higher level there can be extra or general purpose logic to accelerate that better.

Like once they the did put T&L on Gpahics chips and call them GPU. Later they put shaders on them.

The question is what can excisting hardware mean for AI.
Can PPU accelerate AI to
Dito GPU.
What makes a AIPU differ?.

They are massive parralising general purpouse Copro PU's
ATI architectore have more branching power in it. AI could use that to.
A AIPU would be a controler PU managing shader unit with massive branching power.
 
You don't have much of a clue do you, supergee? Physics and graphics rendering are running off well documented principles. You can make a whole library of routines, say, DirectX or OpenGl, that does graphics for you. When it comes to physics processing, you got the law of physics. Ever heard of them?

When it comes to AI what have you? The laws of intelligence? The DirectAI API? Why do you think no one has made a standard AI library yet? Coz it wouldn't work. Cant retrofit an AI routine from Quake 4 to run with Civ 4. You have a very murky situation about what is needed. A grand strategy game like Civ 4 and Quake 4 needs rather different AI programming.

Of course you need lots of calculation of pathfinding and other similar algorithms, but the needs for each game is so dissimilar that you will probably just find a few, very general principles that they share in common. Such as a-star algorithms and other clever techniques. However they are very theoretical and the implementation in each game will differ a lot.

Branching off AI is also going to be challenging, and I think coding the AI is hard enough as it is without having to bother with an actual separate unit to do it. The AIPU will never sell, simply because AI is calculated quite well by a CPU already, and really stellar AI needs something short of a miracle in programming, not necesarily more hardware thrown at it. A dual core processor would probably be just as fine for the job. Soon everone will have one, whereas this phantom AIPU will be sold to one in 10.000 if that many.
 
Lord of Shadows said:
I hope all these little add in cards eventually give intel/amd enough incentive to develope a one for all solution into the cpu.

/me wonders whats next ;o)
Click to expand...

Torrenza My friend. Torrenza! :D :D
 
Immacolata said:
You don't have much of a clue do you, supergee? Physics and graphics rendering are running off well documented principles. You can make a whole library of routines, say, DirectX or OpenGl, that does graphics for you. When it comes to physics processing, you got the law of physics. Ever heard of them?

When it comes to AI what have you? The laws of intelligence? The DirectAI API? Why do you think no one has made a standard AI library yet? Coz it wouldn't work. Cant retrofit an AI routine from Quake 4 to run with Civ 4. You have a very murky situation about what is needed. A grand strategy game like Civ 4 and Quake 4 needs rather different AI programming.

Of course you need lots of calculation of pathfinding and other similar algorithms, but the needs for each game is so dissimilar that you will probably just find a few, very general principles that they share in common. Such as a-star algorithms and other clever techniques. However they are very theoretical and the implementation in each game will differ a lot.

Branching off AI is also going to be challenging, and I think coding the AI is hard enough as it is without having to bother with an actual separate unit to do it. The AIPU will never sell, simply because AI is calculated quite well by a CPU already, and really stellar AI needs something short of a miracle in programming, not necesarily more hardware thrown at it. A dual core processor would probably be just as fine for the job. Soon everone will have one, whereas this phantom AIPU will be sold to one in 10.000 if that many.
Click to expand...

AI libraries have been around since the 70's... There aint nothin new about them. I'm no expert on the topic though. And as was already said, there are several programing languages out that are designed strictly for processing AI.... As a matter of fact any language that claims to be "object oriented" has been based on the lessons learned from prior AI-like languages.

That said... You need to have a standardized library. Trust me if this hardware becomes popular MS --WILL-- write a DirectAI or something similar.

Just as MS --IS-- writing a physics library as we speak.
 
Ozymandias said:
I want to play against something that beats me because of its skill. At this point, the only place to find that is by playing against other people, but this isn't always an option (SP-only games, those lonely hours when you have no web access, etc.).
Click to expand...

I got out of online play a while ago.... I'm tired of all the cheaters. It is impossible to get a leg up today on some of these server becouse the cheaters are too many. I cant stand it... I hate cheaters.

AI on the other hand.... Sounds interesting.
 
This is a very interesting post... See, the PPU section got so much better without Terra flaming everything :D

Personally, I dont think using an add-in card is the right approach to the AI problem. It's a very interesting one, but most imporantly it shows the gaming industry that consumers want better AI...
 
Immacolata said:
You don't have much of a clue do you, supergee?
Click to expand...
Oh you do?
I wonder. Do you think. AIseek bussness model insane? I wonder how AI game coders think about it. They know more about the topic but also not all but a lot more then the average mortal. Like AIseek people to. the would eb specialist. To even try it this adventure to start yet again a new hardware market.
The AIseek software and hardware engineers have had some very professional thought about it and domain expertise to think it could be made real. Maybe they ar elakin in the marketing research thing. But the are going for it to make game more emersive due to enhanching Game AI with a large margin. They are in the busness not to flop but to enhance games more on a specifick field and that AI. First step is Decent Dev support. Gamers can decide to go for it if there are games avaible that put it to good use.

There lot of different Physics Features wich can be done more or less in hardware.
Same for AI. It's the first AIPU. The first HArdware acceleration solution.
Like voodoo1 compared to R600 it has more high level units before the pipeline and ROP and TMU's . A lot was done in software like the T&L part. Same for AIPU first a API and introducing solution. then it evolutes to a AIPU 4 generation wich can do a lot more dissaplines in hardware adding higher level units.

You think GPU are so basic. It's easier to imagine by gamers. So it stands more out.
Physics the game play one is not so obviouce and AI competly not. A game picture tells litle about AI, Physics or gameplay.

Well some day the day voodoo came they went then for geometry model with texturing.
Wich the other solution path would be
Raytrace PU solution
or Voxel PU

I think they will choose the best in there view solution for every AI problem wich can be put in hardware. They will make the best choice based on this current situation.
Now it are pure the low level AI features. More will come. But they are hadding in a specifick way. With it con's and pro's.
That fine ther emust eb making a choice.. With 35nm in the future there is more possibel for nextgen Copo's
But the problem is there choice for a solution could force the Hardware AI market in one direction. IF the Next generation build op on the previouse choices.
Like Rayrace and Voxel are left behind. In the early day the could go either way. Novologic droped Voxel. Novalogic software solution game voxel engine competed agains 3DGeometrie texture beast at that time. Novoalogic games where very CPU dependant.
Same with AI. AI software coders can go any way but are restricted to CPU power. CPU power is shared. While other do AI with a fix choice solution with a Hardware accelerated high level API.
Physics and graphics rendering are running off well documented principles. You can make a whole library of routines, say, DirectX or OpenGl, that does graphics for you. When it comes to physics processing, you got the law of physics. Ever heard of them?
Click to expand...
I know about some know AI technieks also a lot of books are about that topic. not allone game specifick. Then again, Physics is basic learning stuff on schools. That way there i more a Physics awere ness with more people and have certificates of that speciality. And so a lot of school books about it.
AI has large of it spores in the technical reachurge. When CPU become powerfull to simulate AI like for robotics. TU give the need for it.
Well thinkin of this there would be more Physics specialist compared to AI specialist.
Where Physist are well spread on more collage intititutes. AI specialist ar more around TU.
And AI is runing behind as it get its large support of CPU decade while Physics predates its largly.
When it comes to AI what have you? The laws of intelligence?
Click to expand...
Fuzzy logic. comes to mind.
Pathfinding,
Decision Making,
Finite State Machines,
Neural Networks:

There ar books about these topic.
But for education AI is mostly a tech direction
There is a lot of though in AI on tech universities.
AIseek put some of those "laws" in silicon.
The DirectAI API? Why do you think no one has made a standard AI library yet? Coz it wouldn't work.
Click to expand...
Well DirectX is more HArdware aimed. All directX components are low or high level Hardware abstraction and emulation layers. AI is still software so no DirectX.
Physics is not to it's starting to be also a hardware solution field. But with GPU and PPU do it in hardware MS wakes up.
If AIPU are breaking to. and there come some competion from who nows where DirectXAI API would be requiested by dev's and MS act after DirectX PhysX is introduced.
depends wicht new market will stay.
Could be also because the focus is on Graphics. Even gameplay get less attention due to Graphics focus of some Dev teams and there publishers there marketing bussnes model.
AI but also Physics get in the last years much more attention. They are just running behind. Just like a good Game User Interface is very important. Dev team dont have often a well balanced focus. Also the expertise for GFX is no problem but AI and PhysX specialisme for game coding is very variable with different teams.
Think of Crysis did AI and Physics on there own inhouse. Not all teams have guru avaible or high level AI coder or Physix for games specialist.
With High level api's the low level grunt work is put away. More devs get highe rlevel AI and PhysX avaible to there team. with out the need of a guru.
Cant retrofit an AI routine from Quake 4 to run with Civ 4. You have a very murky situation about what is needed. A grand strategy game like Civ 4 and Quake 4 needs rather different AI programming.
Click to expand...
Some is different some is common. So Pahting different?
In FPS you are a unit. In RTS you are a Observer commanding Units from a high view and inter acting. the cammere point is different. Wich can be done in FPS style to, Commanding AI mates. Like very small scale RTS but from the middle in it.
The Pathing AI would be the very be the same for RTS and FPS. The diference is scale and unit types. It even posible like a view games to switch from commanding observer view. to a unit in FPS style and versa visa.
There are games Wich have RTS and FPS in realtime and you can swith back and forth.
This means there is a lot Pathfinding logic common to these kind of games. On other isues they could differ much. Like for RTS there is a need for a commanding AI opponent. Highlevel decision making AI.
So the difference is more for the Player not the NPC or units in pathfinding case.
Pathing is more larger in scale. with more pure RTS or FPS games.

So What can be done by the AIPU wil be pathfinding. Low level AI. For a RTS with many units and lot's of atributes it means lot of computing power needed. Instead making trades off.

Of course you need lots of calculation of pathfinding and other similar algorithms, but the needs for each game is so dissimilar that you will probably just find a few, very general principles that they share in common. Such as a-star algorithms and other clever techniques. However they are very theoretical and the implementation in each game will differ a lot.
Click to expand...
Sure the difference between
RTS and FPS will be there
The differs between a Flight sim and RTS is larger.
Also a space game between a FPS.
But there is always some common ground and some difference.
RPG games will have more ALive. More complex AI.
Space games aimd at treading a economic model adn AI dealing with that.

Well there will be more way's to Rome. But the API engineerd will pick the best solution to a AI problem that suit there needs and will force AI to a specifick High level solution. Low level AI coders can code easaly with.
As AI guru are rare. Not enough of them for every Dev team and expensive.
Branching off AI is also going to be challenging, and I think coding the AI is hard enough as it is without having to bother with an actual separate unit to do it.
Click to expand...
Well the API deal with the hardware and come up with already pre made solution throught the API.. The Ai coder must set it up. And do the highlevel part of his AI engine on top of that. More game specific stuf then low level AI coding. Focus more on game details then low level code. Faster time to market. A API eases the job. Of course there are AI coder with this atitude of I can it better and do a inhouse solution. Where they can make there own choice for a solution to a specific AI porblem. Well. Such dude can do that only in a large Procect where a complete inhouse game engine is developed so there is time for this chocie and the money for it also for this expensive AI specialist.
The common dev's go for a API solution and even a Hardware accelerated combo. A if not made a trade off stressing people with stupid AI.
The AIPU will never sell, simply because AI is calculated quite well by a CPU already, and really stellar AI needs something short of a miracle in programming, not necesarily more hardware thrown at it.
Click to expand...
So you take a look in to the future?
Well the complaines about stupid AI still surfice even in this time. I think the AI expertise must be very high to do it right in a game. So there are some who CPU sufice. Who implement the basic decent level of AI for a game. Do some Asembler optimisation in a time critical AI code routine or algoritme. Well most Dev's laks the time money and expertise in a team to pull that off. So AI will not be often not all that and in worse case be stupid. But that's not all. More AI power means more is possible. Game can get more better AI. Dev's teams with desenct AI expertise could extend there AI features and life likeness to a larger degree. Wich make games richer in emersion as AI behaves more live like. and not so predicatble and scripted. So like way beyond the stupid AI problem. better AI but also more AI features.
A dual core processor would probably be just as fine for the job. Soon everone will have one, whereas this phantom AIPU will be sold to one in 10.000 if that many.
Click to expand...
Well maybe but people buy a PPU not because HL2 can do nice Physics or crysis. But want much more of it. in Nextgen games.
As enhausiatic gamer I alway's want more then is possible now. I look forward to game that put Physics and AI to much higher levels then possible.
Sure it will cost some.
A AI killer game is needed to excellerate AIPU exception. And higher volume sales.
 
Hmm, SuperGee sounds a lot like Terra...

Anyway, I have an email pending with AIseek to request more information. Two days in the queue now, so I hope they'll respond as I am a valid game developer...
 
Once again the wish list fills with joyish glee. Accellerated AI--Now that has me salavating for AVP3 again! Damn they are missing the chance with all the new hardware out now.
AVP1 1999, then AVP2 in 2001. Now on the cusp of 2007 we have physics hardware that will allow more realistic distruction, and now AIPU for more a deadly single player game. WHERE ARE YOU SIERRA???
Why AIPU for me, simple. It takes time to learn the game and no I dont buy Beta software. So I wait about 2--3 months. By then someone is outthere cheating, Great. I spend $60 for a game I have waited 6 years for and with a 2 year old kid it leaves me a miserable hour or two to play and I can look forward to getting my ass smeared by Aliens doing the patented "dolphin jump" if I play online. Screw that. Give me a LAN party of close nit friends or single player. Period.
Now is it too hard to ask for a single board PPU/AIPU ?
 
merlin704 said:
SuperGee, you're posts make my eyes bleed...severely. :(
Click to expand...
Thanks for that informative, if gramatically incorrect post.

I think that an AIPU would likely benefit games in the long run or far future (2010), however I do not have sufficient knowledge about whether most games use third party AI packages nowadays, which may be the only way to harness the power of such a chip. Additionally, I do remember from my 'high-level' machine learning class that certain algorithms require different resources from the underlying architecture. Nonetheless, I would assume that a safe bet would be that most advanced AI processing would like two things:
  1. lots of RAM
  2. lots of 'processing' power
 
Could this be used in a server side only sort of way?

Example:
Game X comes out and has computer AI as part of the multi player (support or somthing). Could the server have the AIPU to do the calculations and everyone gain the benifit?
 
An AI card now? First the GPU, then the PPU, now the AIPU? That is just way too many cards!
 
Well some people like to play it directly on the man. Well no argumenting detected no need to reply:) As I use argument fully explaind but in broken english or something.
So manny people disagree with me but where is there argumentation about the topic?
---------------------------------------------------------------------------------------------------------------------------

Darakian said:
Could this be used in a server side only sort of way?

Example:
Game X comes out and has computer AI as part of the multi player (support or somthing). Could the server have the AIPU to do the calculations and everyone gain the benifit?
Click to expand...
So far as I know. AI could be a server side solution. For Mplay a Client joins so his rig depends on the AI computing power of the rig that runs the server PAckage.

Wich means. the problem of PPU Gameplay Physics synch interactive debri bandwidth flooding is not there.

NPC and unit are dynamical in amount but better controlable limit wich is much smaller in amount then generated debri or particles and amount of boxes. Like say RTS games but a max limit of units like 100 to 400. For FPS NPC or bots it is also fix limit 100 is a amout I think off.
While they put couple off 5000 interactive boxes in a map that must be sync with clients.
So the AI would not create much more bandwith. Depend how many AI objects may be in a game session.
No need for a game with 10000 NPC or units.

So for me it makes sense that only the non dadicated Server. needs a AIPU.
While for Effects Physics de Client needs to have a PPU and that would be optional.

Gameplay Physics is a bit tricky. Could be either way. would expect also server side. makes more sens but CF:CT suppose to not requier a PPU for dedicated server.
Wich is very rare because wenn I run dedicated. the client don't fire up becaus PPU in use. Maybe they have change that. with a algoritmens to spread the Physic task over the clients.
must try that out with R36 of CF:CT
 
I think I need a SuperGeePU card :D

Sorry SuperGee, no disrespect meant, just that your posts are really really really long :p
 
SuperGee said:
Well some people like to play it directly on the man. Well no argumenting detected no need to reply:) As I use argument fully explaind but in broken english or something.
So manny people disagree with me but where is there argumentation about the topic?
Click to expand...

I don't care for the broken english, it is the broken arguments that is the problem. I don't need 3 pages to make my point.

AI is not something you need to make extra cards for. The calculation needed is sufficently taken care of by the central processing unit. The "problem" with AI is much more about programming routines that works in a very specific situation.

The dry facts are that there is no DirectAI API and no one is about to make one. I am rather sure that none of the AI libraries written in the 1970s are used today in games. Perhaps theorems and principles about expert systems and fuzzy logic and what not is implemented, but the actual programming is new. Much of AI code is menial sorting of data, which a CPU does just fine.

If you look at the practicality of it, you have games where you need relatively simple AI. In those situations a AIPU is a complete waste. Using the CPU for AI means you can throw what ever spare cycles at other tasks. Having some people have an AIPU and others not, means that you cannot just write one set of AI routines, but have to write more. I'd probably slit my wrists before trying to do that if it was my game.

Just look how much of a roaring succes the Ageia PhysX has turned out to be. I do not believe that the AIPU will ever be more than a press release. It is an absurd idea.
 
Immacolata said:
Just look how much of a roaring success the Ageia PhysX has turned out to be.
Click to expand...
It's way too early to be calling it a failure. It's been out four months. Things may well change once the games start rolling out, especially with Epic integrating PhysX with the UE3 SDK.

cf. 3D cards in the mid '90s. Didn't have support from dozens of titles the second they hit the shelves, didn't sell millions overnight, and yet you would have ended up looking pretty silly talking about 3D acceleration in the past tense...
 
Nice PDF on AIseek site.

AIseek Intelligence for New Worlds.pdf

It explains what a AIPU could mean.

It explains the status of AI for Game developers before and now and yet to come.

It looks like the low level AI is very computational and algoritmic rich. And the High level AI depends on that.
Where Low level AI is very commonly in games even in different genres.
The High level is more game specifick. But the input for that is often the low level AI results.

Of course the source is from AIseek. But roaming Develor forums on AI topics give you more insigt what Dev's think off it.

AIseek Bussness model and marketing reseach would incorporate a lot of polling by the dev's.
 
Sovereign said:
Once I see benchmarks, then I'll know it's not just ghostware or another fancy-looking card that does nothing...eventually we won't have high-power CPUs, just a bunch of add-in cards for a low-power CPU to coordinate...
Click to expand...

One of the few people with realistic foresight, thank you.

I completely agree, the progression towards dedication for any classifying action in a program is obviously apparent in the history of computer design. An ai unit was unexpected, but I feel everything, no matter how inconsequential will have it own dedicated unit. Specifically when dynamic processors are developed and in use.
The only issue with an AIPU I can see, is that using one for all games, defines all games AI to be the same, you'll be playing in the "ai environment" game after game, all that will change is the graphics, and play mode. But then again, this implies that the the AIPU will not have a diverse instruction set, which I think it will not. Perhaps in later years game designers will be able to specifically set which thresholds are important in any entity's "mind". For example, this unit defines at the basemost level, the idea of a goal point, and the idea of finding a path to it, then it moves down a tier, and meets the next set of requirements, like finding cover along the way from enemy gunfire in an fps. Can we define alternative requirements for activity at this point? Or better yet, our own tier system for any given unit in the game?
 
I have been waiting for this component to come out. If there was anything that needed dedicated calculations, it's a bot. They have become so prevalent in online gaming as well as single player gaming that it was time people realized that if you could devote a few more cycles to the AI, they'd stop running into the wall.

Wait, don't I do that too? :eek:

I also would love to see this create a more realistic single player experience, but having been without broadband all my life until just a few weeks ago, I can safely say that the single player sphere of influence is slowly dying. Whether it be online FPSs, MMORPGs, or RTSs, people are forsaking what started many of us on our gaming binges. I don't think this is altogether bad, if you consider the congregation to be a form of socializing. However, phone lines and cable boxes can always have problems. I hope to see AI more focused upon, so we can move closer and closer to the day that SkyNet goes online. :D
 
drizzt81 said:
Thanks for that informative, if gramatically incorrect post.
Click to expand...

Thanks grammar nazi! Without you pointing out my mistake, I would have never known! :rolleyes:

Anyhow, I think the AIPU will be beneficial in the near future. I just hope it will be reasonably priced upon release. That it my biggest concern.
 
Kalabalana said:
One of the few people with realistic foresight, thank you.
Click to expand...
I don't think sol, read the PDF. A decent CPU is still needed. AIPU only accelerated the computational demanding low level AI. The AI that's most common in games.
The High level AI build on that whats is most uniek to a specifick game.
This High level AI will still be done with a CPU.
So just like a PPU need a decent game rig and today that will be a DC. A AIPU need a decent CPU to for the much more high level AI that possible, to compute because with the more and accurate low level AI more highlevel AI is possible. Like More Sensor AI features that add up. Wich makes highlevel AI the decision making with more input more complex.
I completely agree, the progression towards dedication for any classifying action in a program is obviously apparent in the history of computer design. An ai unit was unexpected, but I feel everything, no matter how inconsequential will have it own dedicated unit. Specifically when dynamic processors are developed and in use.
The only issue with an AIPU I can see, is that using one for all games, defines all games AI to be the same, you'll be playing in the "ai environment" game after game, all that will change is the graphics, and play mode. But then again, this implies that the the AIPU will not have a diverse instruction set, which I think it will not. Perhaps in later years game designers will be able to specifically set which thresholds are important in any entity's "mind". For example, this unit defines at the basemost level, the idea of a goal point, and the idea of finding a path to it, then it moves down a tier, and meets the next set of requirements, like finding cover along the way from enemy gunfire in an fps. Can we define alternative requirements for activity at this point? Or better yet, our own tier system for any given unit in the game?
Click to expand...
Well AIseek think otherwise. Apose to us they are domain field experts. That's needed to even start such undertake. In that company there must be a decent highlevel of AI understanding.

The First AIPU will only do low level AI wich is common in all games. Makes me think of Design Pattern. But then for AI. AI solution Patterns?
The high level AI would be more different in games.
But because of the more Low level AI power at hand the highlevel AI can be a lot heavier to. but that is still computed by a powerfull CPU. More AI features to choose. And how they act on the low level input is the AI designer choice.

The AI programmer doing the high level stuff the desistion make hiarchie and desision thresshold and wich sensors features are used, define how AI will react in games.

Like wenn to avoid player if player is more powerfull. the game designer deside wenn, in the desision tree tresshold.Have choices on weapon and or health and or armor and or group size. And set's a tresshold. There are a lot of choices to take. While a other trow in some more parameters and use not all of those previous mentioned. Like a Elite troopper will atack also agains a bit more powerfull unit but have also more change to sucseed and use cover way better and weapons avaible. While a main trooper will retreat and call in fellow forces before attacking and be less efficient with the same weapons. So has a lower tress hold to deside to fall back or attack.
That's where games will be Different.
Pathing done right will be indentical in apperance for most games. no need in much variation. the shortes and or fasted path is the same for every game. The problem is will the AI code solution algorithem find it in time or find it at all.

I think just like a PPU and AIPU do offload the CPU so you think then I can go along with a cheaper CPU.
That is wenn you buy $250 + $250 for PU's just to offload?
PPU and AIPU most primory task and value is for enhancing games further then is posible with a CPU.
Wich means yes it does offload. But the game enhancing give the CPU a new different larger load wich will partly be compensated by the offloading.
 
SuperGee said:
That is wenn you buy $250 + $250 for PU's just to offload?
PPU and AIPU most primory task and value is for enhancing games further then is posible with a CPU.
Wich means yes it does offload. But the game enhancing give the CPU a new different larger load wich will partly be compensated by the offloading.
Click to expand...

So why not just use the second core of the new cpus with two cores? You will end up writing parallel processing no matter what. Aiming for the second core makes life SO much easier for the devs.
 
Well you forget a CPU is multipourpose wich means, what to do with that second core? Well they can use it for any thing! You can Split a game in every task module. Or
you can also split a heavy task like rendering to!!!!. Well Quake 4 went for boost SMP for rendering power. Quake 4 is also CPU depending so it scales well. Make sense there. For there Project. While crysis might use a large part of that extra core for more Physics and better AI. Wich means It will be not so straight forward that one task gets a core exclusivly but Devs can choose how the split there game as they fit even task it selfs.
So whats get the focus. Or be more balanced.
Like I already said. AIPU would not accelerate all AI. Only the low level and do it a 100 times better. But the result off that will be used by the highlevel AI, wich runs on the CPU. That extra core comes in handy. Because a HighlevelAI load generated by AIPU could be a lot heavier.

As for Dev's Game engine aren't pure Rendering annymore for a long time. They are often a complete package. The Design level tools incorporate physics and AI.
Unreal Ed 3.0
Crysis engine tools to.

AI API is often Part in a total solution game engine. So has Epic incorporate PhysX in there Unreal engine. So there are a lot of AI API but are part of a total package over shadowed by the graphics component.

Well wen I hear about a engine.
1) Support it DX10 :)
2) Wich language is supported. C# besides C++. Oh not :(
3) It has Net API for Mplay uses, nice
4) It support rigid body.
5) it has more Physics features :p
...
n Oh it as AI to :)
m It has a lot more AI features :p

o Supports consoles

I admit the eye do want some :)

Point is lookin first to the render part.
 
Odd, I tried to leave my post as unbiased, and vague as possible.

Unfortunately, what I said may be a little argumentative but is very close to the truth. Otherwise all my profs who taught me the forefront of A.I. tech at UofT these past years are talking out of their asses (Which I doubt since I trust their professional word, and Ph.D's over the word of some people you asked on tthe net.).

Did you want me to quote you originally and say you were one of the few people with foresight? Well, probobally, but no cookie today supergie.

I hope this does not set you into a frothing rage, since I am not a troll.
 
Kalabalana said:
Odd, I tried to leave my post as unbiased, and vague as possible.
Click to expand...
Yes vague and short it is. But they are agains to long post. So I link to a PDF. Wich I think leads to the core of the problem and the solution. It make a lot of sense. AI game specialist.
Unfortunately, what I said may be a little argumentative but is very close to the truth.
Click to expand...
You think? But then again it could be totaly not. I don't say I'am Right. It's my opinion. The way I see it. Reading a lot of Game design and programming books. Have a bit more insight then the average gamer. At the core the theory of the Physics and AI it laks but very important game related is a more different cookie.

That's why there are AI books about AI theoreticly
And very much Game AI books.
Otherwise all my profs who taught me the forefront of A.I. tech at UofT these past years are talking out of their asses (Which I doubt since I trust their professional word, and Ph.D's over the word of some people you asked on the net.).
Click to expand...
Tom did the scientist inside about Physics. Well they are domain specialist. More Therotical physicist and in your case AI theroretical Proffesors. Or scientifical practical for robotics. Or experimenting with highlevel AI, simulating a human brain. But do they no much about it related to Game intergration. Gameplay realtime physics and AI for games where developers have to deal with.
On there field off expertise they are right but to the game side they could be wrong. Because. Some other very important factors come into play wich swirl it up.
Did you want me to quote you originally and say you were one of the few people with foresight? Well, probally, but no cookie today supergie.
Click to expand...
Nice PDF on AIseek site.

AIseek Intelligence for New Worlds.pdf
That's its in some way my opinion to but in good grammar and long text.

Well read the PDF at AIseek and then I want to know your critical opinion on that.
The thing wich get my attention is the AI trade-off.
I hope this does not set you into a frothing rage, since I am not a troll.
Click to expand...
Not at all. Tom uses Scientist. Domain specialist wich know there field of non game related expertise. But no jack experience for Practical games usage. The problem is those opinion are often to high rated because. Well they must know it. You would expect and the masses asumes it to. But the smartest people could be wrong espacialy if they drift away in a related specialized practical field, they have not much experience with, Practical Game AI and physics usage. They look at it at a Theoretical way or in a Practical non game related way they are right for there use in there field. But for Game usage?

In the early stage not gamer deside if it will be a sucses but the dev's. They must support it and it take years to develop a game so if Dev see something in it some will support it. And if it makes a difference, more devs will. Till the first game hit the market and more follow. Then gamers comes into play and decide to use this extra by buy'in the hardware for it.

The way I see it.
I weight the words of some established AI game programers wich could have different opinions to but with better insight arguments. more then some Prof that do AI on a University wich uses AI differently like robotics.
So dito I weight the words of established Game Physics programmers more then a physics dude doing some research on a Sientific GPGPU project, commenting on something he never used or tried.

It boiles down how Dev's Game designer and the team members wich must do AI and physics, think about the AIPU. There will be some devs lookin into it and try it out. Some Game AI research about testing what could AIPU mean for them. If they see potentional and first Dev testout impresions are positive Dev's will more willingly to support it.
AIseek sucses depends on the devs if they will support it.
It take time. And is in a much more early stage then Ageia PhysX wich have decent Dev support. They are starting to reach the gamers.
AIseek is in a stated the must convince Dev's.

Talkin about Game innerworks make me want to pick up game designing again. As a lone wolf indie Game dev'y.
My Preference is RTS and Space game.
I'am lookin forward to a RTS game where unit act life like. Where you lead your side or forces. Instead of Babysit every unit. As aresult of stupid behavior.

Well I look it as a Gamer wich apriciate more detail and life like AI more earlier the years then later. That's the point for me.
 
Hmm, while I'm still waiting for the "need" to purchase a PPU besides the "possible want", the AIPU comes along.

While I'm not going to run out and buy one right away, I'm always happy that technology is being produced that in 30 - 40 years may give us all the equivalent of a hologram room for our entertainment needs.

Look at the abilities we have now in comparison to just 15 years ago. If the capabilities accelerate at the same rate, I seriously think that we'll have Star Trek level hologram capabilities in 30 to 40 years. Granted the pricing may not allow personal home use, but that will also happen in time.

Just imagine what BF30 would be like.
 
HighTest said:
Hmm, while I'm still waiting for the "need" to purchase a PPU besides the "possible want", the AIPU comes along.
While I'm not going to run out and buy one right away, I'm always happy that technology is being produced that in 30 - 40 years may give us all the equivalent of a hologram room for our entertainment needs.
Look at the abilities we have now in comparison to just 15 years ago. If the capabilities accelerate at the same rate, I seriously think that we'll have Star Trek level hologram capabilities in 30 to 40 years. Granted the pricing may not allow personal home use, but that will also happen in time.
Just imagine what BF30 would be like.
Click to expand...
Well I'm 37 now so I can imagine that yes, games will have a true holographic room in 40 years but damn, who is going to play the 80 year old? Or will I even care about anything other than the last day I had a B.M.? Perhaps my children will be nice and choose a nice nursing home that will let me have a P.C. in my 12' by 12' room I'll have to share with Mr. Wilson, who hates noise in his room after 7P.M. Actually I am fearfull the day is comming that we will simply be taken out back and euthanized after our 70th B-day. So in 40 years, I'll pass on gaming. ;)
 
SuperGee said:
Yes vague and short it is. But they are agains to long post. So I link to a PDF. Wich I think leads to the core of the problem and the solution. It make a lot of sense. AI game specialist.
Click to expand...

the PDF you linked to was created by the company that is trying to market its product. Do you really think that it contains an objective view? Just like the Killer NIC whitepaper, I would take any information directly from the company with a large grain of Salt.
 
I already point that out with the Critical Opinion I requested. Problem is.
There are no Established AI or Physics programmers interviewed of there opinion.
They have to deal with those implementation problems and computing restriction's a CPU have. Also compete for CPU cycles to other members modules. There Opinion wich matters most are not avaible. AIseek are the clostest thing we got now but are offcourse subjective on the matter.
So Drizz did you read it and what do you think about it.
-------------------------------------------------------------------------------

griff30 said:
Well I'm 37 now so I can imagine that yes, games will have a true holographic room in 40 years but damn, who is going to play the 80 year old? Or will I even care about anything other than the last day I had a B.M.? Perhaps my children will be nice and choose a nice nursing home that will let me have a P.C. in my 12' by 12' room I'll have to share with Mr. Wilson, who hates noise in his room after 7P.M. Actually I am fearfull the day is comming that we will simply be taken out back and euthanized after our 70th B-day. So in 40 years, I'll pass on gaming. ;)
Click to expand...
That's a good one. I'am about your age to. I think I will be dead then, so I completly would mis it. We are born to early. That more a nice given for the young gamers.

For me I look forward like 5 years. what will be possible then what will Dev do with games. with the hardware avaible then. DirectX 11/12

Holographic Room's like startrek. 40 years could be a to possitive estimate could be 150 years.
 
SuperGee said:
So Drizz did you read it and what do you think about it.
Click to expand...

I read it. It sounds interesting for sure. I am wondering how much the lack of AI in current generations games can really be attributed to the lack of computation resources rather than the lack of time that developers spend on this item.

Let's look at the market. How many people are buying games for its AI features rather than the flashy graphics? After the popularization of multi-player games, there has been less and less emphasis on the AI in SP mode.

Two games that I can remember for touting their AI are FEAR and Titan Quest. And neither of them delivered. Honestly, I would not be surprised if a large amount of the consumers are somewhat turned off by promises of great AI on the box, only to find that the actual implementation is rather mediocre.

Anyway, the question that I would assume needs to be answered before we can make proper statements about the AIPU is what the underlying problem to poor AI is:
Lack of computaitonal resources: While it wouldn't be a problem to create excellent AI, such things require tons of CPU resources, therefore are not feasable.
  • Lack of human resources: While one can program good AI and it would work reasonably fast, companies do not see the value in this, hence do not use their human capital to implement good AI.
  • Lack of tools: Programming an AI is very difficult due to a lack of standardization and a
  • Lack of tools for developing AI: Programmers often have to reinvent the wheel, which makes AI development very wasteful.

While the real answer for the lack of good AI in games is most likely a linear combination of the three points, I think that unless the first point is a major hurdle, there will be little need for an AIPU in the near term.

I am generally a supporter of the idea that in the mid to long-term, there will be ASICs for most functions in a PC, since they are likely to perform orders of magnitude better than a GP-CPU at specific tasks. I have a problem with the fact that it's currently in fashion to launch an ASIC.
 
drizzt81 said:
While the real answer for the lack of good AI in games is most likely a linear combination of the three points, I think that unless the first point is a major hurdle, there will be little need for an AIPU in the near term.

I am generally a supporter of the idea that in the mid to long-term, there will be ASICs for most functions in a PC, since they are likely to perform orders of magnitude better than a GP-CPU at specific tasks. I have a problem with the fact that it's currently in fashion to launch an ASIC.
Click to expand...

Ive tried to convince supergee, but he is adamant that such a device is a good idea. And for the record, I do not believe the future will be specialised ASICs, but a redefintion of what the CPU is. You will still have one product doing your calculations, but perhaps it will have the work subdivided so that the future "CPU" will really be a grid of processors each doing something. I doubt we will see a large boost of cards and addons. It is not in the interest of the cpu makers (and they really are the ones with the power), nor in the interest of the consumers. At least not if you don't want to make building your own a computer even more of a hassle than it is today :)
 
You must log in or register to reply here.
Back
Top