The Intia Processor (AI Acceleration)

[XSAlliN]

Are you tired of the word "Stupid Bots", NPC (Non Player Character) or not a chalange for old gamers, at the begining it was nice, we played and played until we got bored, that's when Online Gaming replaced the FUN whit NPC, most of the latest games look nice (Eye Candy) but not very interactive, now, the only games where a gamer can have some fun in SP (Single Player) are story based games like HL2 (Episode), the full action based games are to easy and even if you finished them you don't get satisified like in the old days, well, Intia Processor (http://www.aiseek.com/Intia.html) might be a good solution for changing that, we'll just have to weight and see, here are some specifications from the main site, plus some Demos.



Game technology is continuously evolving to help developers build better games. Unfortunately, recent advances in graphics quality and physical effects have not been matched by a similar increase in the sophistication and believability of artificial intelligence (AI). In the words of one leading developer, “characters still act like cardboard cut-outs”. Indeed, the vision of a “living, breathing” world is yet to be fulfilled – a world where thousands of non-player characters (NPCs) move and act intelligently at all times, within a large, constantly changing environment.

The AI acceleration technology developed by AIseek addresses this fundamental challenge. By accelerating CPU-intensive, low-level AI routines, AIseek frees developers to create sophisticated, higher-level AI on a massive scale. Moreover, with accelerated AI, game worlds can be large, complex and constantly changing – without compromising on the number or intelligence of the NPCs inhabiting those worlds.


Introduction. To address today’s AI challenges, and to enable the creation of entirely new game worlds, AIseek developed the Intia processor. The Intia processor is the world’s first AI acceleration chip. The Intia processor is accompanied by a software development kit (SDK) for game developers. Using the SDK, studios can empower their AI modules and games with accelerated AI. The Intia processor’s accelerated AI functionality includes movement, sensory simulation and terrain analysis.

Movement. The Intia processor features pathfinding functionality that is both optimal and extremely fast. Unlike today’s software-based approaches (e.g., A*), the Intia’s pathfinding uses no heuristics, thereby guaranteeing that the optimal path will always be found. This optimality also means that the Intia processor avoids the common pitfalls of A*, including failures to find a path when one exists, and the generation of “artifacts” (e.g., weird, unrealistic paths). If a path exists, the Intia processor will always find it

The Intia’s pathfinding is not only optimal, it is also extremely fast. Processing time for each 100 nodes of path depth is only 10s, making the Intia processor about 100-200 times faster than A*. Most importantly, this very significant speed increase gives the Intia processor the adaptability to support large, dynamically changing maps. Unlike traditional approaches, which use relatively static, pre-processed maps, the Intia processor supports maps that change continuously, with no pre-processing required. By removing this limitation, the Intia processor enables the creation of new game worlds that are based on large, rapidly changing environments.

The optimality of the Intia processor’s pathfinding includes excellent support for tactical considerations. Thus, in finding the required path, the pathfinding algorithm can take full account of any tactical information, such as the need to find a path that passes through certain locations (e.g., hiding points, enemy positions).

To facilitate the integration of the Intia processor with existing studio AI modules, all common game graph formats are supported. These formats including grids, navigation meshes and waypoint graphs.

Sensory Simulation. The sensory simulation capability of the Intia processor features fast, highly accurate line-of-sight functionality. Unlike current methods, the Intia processor does not rely on simple point or box approximations for the bodies viewed. Instead, the actual visible area is computed. Again, as in the case of pathfinding, this increased accuracy eliminates “artifacts” (e.g. seeing through walls, failures to see an object that should be seen, etc.).

In terms of speed, the line-of-sight checks are several orders of magnitude faster than current methods: 512 agents can be checked against 512 agents in 0.02s. Once again, by leveraging this immense speed, very dynamic maps are naturally supported.

Alongside the speed and accuracy of its line-of-sight checks, the Intia processor makes available a sophisticated vision model. Under this model, the probability of detection depends on the size and characteristics of the area actually visible. Thus, as in the real world, small objects or objects that are farther away have a smaller probability of being detected. On the other hand, bright or fast-moving objects have a higher probability of being seen. Excellent support for hiding and camouflage is also included.

Terrain Analysis. One area of AI that is particularly underexploited is terrain analysis. In today’s games terrain analysis is usually performed statically during the game design process; strategically important locations are identified once for the benefit of NPCs and remain fixed throughout the game. However, the advent of dynamically changing maps calls for a leap in terrain analysis speed and sophistication. The Intia processor answers this challenge with advanced terrain analysis capabilities that can be run in real-time to enable truly adaptive decision making.

To support such strategic planning at both character and game-wide levels, the following built-in topology analysis functions are provided:

* T-connectivity: the area consisting of all points that can be reached at a given cost T.
* Critical Points: strategic locations, sometimes called “choke points”, that form the bottleneck for getting in or out of an area.
* Controlling Regions: areas that permit strategic domination of the terrain.

Moreover, via the SDK provided, the terrain analysis abilities can be easily extended to encompass custom functionality (e.g., wall building, ambush points etc.). In particular, tactical information may be overlayed procedurally on the game map.

Demos (http://www.aiseek.com/Demos.html) - Have Fun! ;)

[nn_step]

umm didn't we get this posted in the news section a little while ago?

[BeyondSciFi]

http://img367.imageshack.us/img367/633/unitsdx6.jpg

THE HARDWARE IS FINALLY CATCHING UP!!! WOOT

JUST IMAGINE 1000's if not millions of insects or animals in a game all acting realistically

If you ask me, I would predict that the first REAL game will be out just a little while after this new hardware (PPU and AIPU) prove themselvs to the public, around 2009 I would guess. Think about it: fully real physics, photo-realistic graphics, amazing sound quallity (like 22.4 or something as opposed to 5.1) and mart AI, almost at a human level, basically... think The Matrix. Scary stuff, CANT WAIT WOOT!!!

[Lightman]

umm didn't we get this posted in the news section a little while ago?


hmmmm...... long time ago :rolleyes:

PS. I hate this idea of AI offloading, this should be done on CPU where brilliant coder can make whatever he wants. On dedicated silicon, using dedicated API you will be somehow limited/same as others. :slapass: :slapass: :slapass:

[Repoman]

completely rediculous.. especially with quad cores coming out. We don't need 12 processing cards to play a damn game..

[Cooper]

I think software developers might be needing more programmers to utilize those APIs and alot more to patch the bugs :D

And yes , related article was posted month ago (http://xtremesystems.org/forums/showthread.php?t=114280&highlight=intia)

[BeyondSciFi]

We need a AIPU, PERIOD. It will prove itself just like the PPU did. Now everyone wants real physics.

BTW this should be the last add in card, totaling 4, the GPU, APU, PPU and AIPU.

[DTU_XaVier]

completely rediculous.. especially with quad cores coming out. We don't need 12 processing cards to play a damn game..
If this continues at the same pace, the term "CPU" will be completely obsolete for gaming requirements in but a few years... Sad really, as having a "gaming machine" now means "something not even able to process word", as there would be no CPU... Unless they re-wrote word for an AIPU :p:

Best Regards :toast:

[CBONE]

What is the point of an AI add-in card? Is there really a pressing need for more advanced AI than what is available today?
Who cares if an NPC can go to take a dump when he has to take a dump? Or that they can, in fact, find their ass with a map? Several games manage to not be populated with retards that walk into gunfire, and companions that can follow the character without getting hung up on a tree.

[BeyondSciFi]

http://www.youtube.com/watch?v=VeSQI2hinp0 link to video

You can see the need for hardware acceleration when the number become too great. A cpu might do fine with maybe 50-100 NPCs but how about 50,000? This would bottelneck the entire systems... the AIPU is needed... Its not just about more units, but smarter, realtime adaptable NPCs.

[uOpt]

I really hate how they abuse the term "AI".

There is really nothing intelligent here. It is about pathfinding and line-of-sight computation which is very straight search. While these guys at least allow for changing maps they still don't consider enemy positions on map, only deformable terrain. That means AI controlled figures will still happily run into enemy line-of-sight even from known enemy positions (they might react after getting hit but it shouldn't happen in first place, and this "AI" coprocessor doesn't help with the reaction at all).

I would much prefer to reserve "AI" to decision making processes or any kind of machine learning or knowledge building (very short term of course, just what happened in the previous minutes in this level). Using the term "Intelligence" at the very least requires a basic self-preservation instinct based on past events on the map. But we are far away from it.

While some people applaud that you can at least offload of "sub-AI" things line pathfinding, flocking and LOS computations to a different processor that still doesn't give gamers what they want. To act reasonable you still have to go through decision-making (or at least instincts) on your main CPU, or one of your main CPUs, so the offload is really not that great.

[DTU_XaVier]

Did SkyNet suddenly take over the hardware scene or something?? :p: :D

Best Regards :toast:

[BeyondSciFi]

uOpt has a point, but maybe he didnt think far ahead enough (no offence). Just as the NPCs are programmed to pathfind, they can also be made to make decisions. After all dicisions are just paths themselves. How you want them to act is up to the programmers. You can make the same kind of NPCs take some different paths/decisions, say agressive, passive, lazy, crazy, so on, and all can be determind by the decisions they make when confronted by obstacles. If thats not enough, you can always throw in a little random...

True AI is when you give the power to the NPC... instead of programing it to act one way or another...

[awdrifter]

I think with the upcoming C2Quad and AMD 4x4, it would be a better idea to just integrate this into the CPU. 1 cpu can run the game, 1 can run the physics, and another 1 for AI, you would still have 1 CPU left for all the other OS processes.

[Repoman]

We need a AIPU, PERIOD. It will prove itsself just like the PPU did. Now everyone wants real physics.

BTW this should be the last add in card, totaling 4, the GPU,APU,PPU and AIPU.

Wait, where was I when the PPU proved itself? I only remember no one buying it and there being a total of what, 4 games that support it :confused: Then again, I've been out of the hardware game for a few months, so what do I know..

And last add in card? That's what we all thought about PPU..

I seriously don't see the need for it.. with quad core developers could develop two whole cores to it. Going by your numbers, that would be 100-200 "things" being processed, if not more with CPU power today.. why do we need absolutely realistic AI for anything more than that? While we're on the subject, why do we need absolutely real physics? I like my CS:S/HL2 Havok physics, not completely realistic but fun (and run fine on AXP's and slow P4's and the like). And that is what games are all about.. if we wanted real we'd go out and do it ourselves.

[BeyondSciFi]

I think with the upcoming C2Quad and AMD 4x4, it would be a better idea to just integrate this into the CPU. 1 cpu can run the game, 1 can run the physics, and another 1 for AI, you would still have 1 CPU left for all the other OS processes.

The main problem with this idea is that a CPU is many times worst at any single task then a dedicated processor. This is why we offload the data to other processors, not because it cant do it but because it is to slow at doing it. The CPU is a multitasker, and can do many more things then say a GPU or APU, but it cant do them all well. Maybe in the future we will make a chip that can do everything well, but it would probably cost too much and be impractical.


Wait, where was I when the PPU proved itself? I only remember no one buying it and there being a total of what, 4 games that support it :confused: Then again, I've been out of the hardware game for a few months, so what do I know..

And last add in card? That's what we all thought about PPU..

I seriously don't see the need for it.. with quad core developers could develop two whole cores to it. Going by your numbers, that would be 100-200 "things" being processed, if not more with CPU power today.. why do we need absolutely realistic AI for anything more than that? While we're on the subject, why do we need absolutely real physics? I like my CS:S/HL2 Havok physics, not completely realistic but fun (and run fine on AXP's and slow P4's and the like). And that is what games are all about.. if we wanted real we'd go out and do it ourselves.

So by your last sentence, you would go out like you were playing GTA and steal cars, kill people, and cause all kinds of mayham? Yes, games dont have to be real to be fun, but them again Supermario Bros and Pong had their limits no? I think games are made to entertain, and whats more fun then giving a near real enviroment to play in and do whatever you'd like, because if you did the stuff you do in games in the real world, there are very high consequences. And I think the PPU proved itself the moment it actually did what it was designed to do, simulate physics. How many units sold is not a real measurement of its capability. Like all new techs, they need time to go main stream. What would it be like if the GPU was never invented... see my point. Also, the AIPU would probably be the last addin card mainly due to the fact that there are no other major tasks that can be offloaded the CPU.

[Repoman]

The main problem with this idea is that a CPU is many times worst at any single task then a dedicated processor. This is why we offload the data to other processors, not because it cant do it but because it is to slow at doing it. The CPU is a multitasker, and can do many more things then say a GPU or APU, but it cant do them all well. Maybe in the future we will make a chip that can do everything well, but it would probably cost too much and be impractical.

Yes but can't the CPU do it "good enough?" It's not like completely realistic physics are that important for gameplay, at least IMO. An example being the Havok physics I mentioned earlier. AI is obviously more important, but with power of CPU's it seems like they will be good enough once again.

Like I said, quad core in a year, 8 core and more coming soon.. I'm pretty sure we can put those to good use..

[BeyondSciFi]

More powerful CPUs will still be needed to control all sub processors and other less important tasks. As the GPUs, PPUs, APUs, and AIPUs become more powerful, so too will the CPUs needed to feed them with data. Well all know how a weak CPU can effect performance, best example I can think of is in 3dMark. The adding of more cores is just a responce to the inability to clock them faster, yes more cores = more power, but so would higher clocks. What we really need is more processing power, regardless of how its done. What we should really care about are the end results, the technicalitys shouldn't really bother us as we are not the ones in charge of making them work. I could care less if physics can be done on a GPU or if AI and be done on the CPU, I just want them to be done, and done well, and if dedicated processors are the best way, then by all means bring it on.

Also, I dont really understand why so many of you are oppossed to more addin cards. Yes the overall price will go up, but wouldn't the end result be worth it? After all prices go up all the time, a GPU a couple of years ago was around 300 bucks, now we have some costing over 600 dollars, not to mention SLI and crossfire. Besides after everything goes mainstream, prices always decrease. So say 5 years from now the average system cost around the same as today, but you also get more capabilitys from the cheaper addin cards the future has to offer.

I think many of you are thinking about the now instead about the near future. All these new products like the PPU and (hopefully) the AIPU will have a hard transition, but think 5 years back from now... Would you then have things go back to the way they are now? Progress is change... adapt or get left behind.

Ok I'll shut up now, I have ranted too long. Hopefully I have gotten my point across without sounding rude. I thank everyone that has read what I have spent some time writing. :)

[Repoman]

Also, I dont really understand why so many of you are oppossed to more addin cards. Yes the overall price will go up, but wouldn't the end result be worth it?

This is my point.. the end result WOULDN'T be worth it imo. When we can achieve nearly the same thing with less expensive CPU's that is what should happen. Not everyone has the money to buy all this stuff (good on you if you do) or would rather spend it on something else.

Not trying to sound rude or anything, just my opinion :)

[[XC] leviathan18]

why use add in card when you can make a highly efficient api to work with a cpu quad cores are coming just use 2 cores for ai if you think it wont handle it with 1 core

[uOpt]

There is one big problem with stuffing your compter with specialized processors for AI, physics and even graphics (the GPUs we have today).

The same problem you have with developers using additional cores exclusively to do AI or physics computation.

==>

It limits what you can do. What you really want in a multiprocessor system is that any CPU can do anything, and that one single task can be distributed to all CPUs.

Lets consider a very simple situation: at the current point in time during frame calculation all physics processing is done. You are currently waiting exclusive for the pseudo-AI to finish. The pseudo-AI either runs on a specialized processor or it runs on one core put aside for it, but never two or more cores (with the current level of multithreading in games). Wouldn't it make more sense to now have all CPU cores work just on the AI simultaneously?

That is why high-performance applications (better than games) always split up single tasks into threads, instead of picking cores for tasks. Doing the latter is just very wasteful and leaves you with idle cores all the time. Doing the former is hard programming work. It is so hard that current game developers can neither afford it from a cost-of-work standpoint nor from a time-to-market standpoint, in particular since most gamers have one core, less have two cores and almost none have more than two (Hyperthreading is useless for games).

%%

And it gets worse than that for these specilized *PUs. They are one-trick ponies that cannot be upgraded and that cannot be re-programmed to do their thing differently or better. As I mentioned earlier, the "AI" that is presented to us here in the "AI" chips is just a minor part of what you need for a challenging computed opponent. Granted, today's games do not really need challenging computed opponents since games are balanced by just giving the human player more enemies shooting at him.

But that will not continue, game AI will improve considerably. If you have a multi-core CPU, just a general-purpose CPU, then you write a new program to do a programmed opponent and the CPU runs it. These specialized processors can not.

In my opinion all this is a waste of perfectly fine silicon. And don't get more started on the "physics" processors. Did anybody ever look up what kind of limited physics they do? No way you can use them for flight path computation of a winged airplane, of a missile (as opposed to rockets) or for realistic armor penetration computation. They don't even try.

This is like buying software that you can never, ever upgrade even one bit.

Game programmers will have to make proper use of multiprogressor/multicore systems, and proper means not dedicating cores to specific tasks, splitting up tasks into parallelizable units, all cores working on the same thing. They won't like it but no game developer will survive without it.

[xenolith]

I could care less if physics can be done on a GPU or if AI and be done on the CPU, I just want them to be done, and done well, and if are the best way, then by all means bring it on.

Well said. :toast:

I don't know how people can be for or against something like this based on personal opinion, when they don't yet have anything empirical to base that opinion on. Unless they have a time machine. :confused:

[[XC] hipno650]

i belive this has real potenail!!!!!! because most people have to admit ai are stupid!!!!!!!! and inorder to get the same effect as an aipu would you would need one heck of a cpu because the cpu is not made for that and that alone so it is in effient at it. also going by what you guys are saying the cpu still needs to feed the graphics card, do the super real physics that everyone wants it to do, do the super real sound that everyone want's and the network calculations. as you can see even the best cpus would die trying to do all that. as a result things get dumbed down alot resulting in bad sound, bad physics and sub par ai. if you want the quality dedicated prossesors are the way to go!!!! the only problem is lack of pci slots and elevated cost. but the solution would be to build the xfi and bigfoot nic into the motherboad.

[mr_knowitall15]

F all this crap. With multicore processors becoming mainstream, and more and more cores being available, why the hell would we need a dedicated card for physics, AI, even better sound (even though most modern mobos have decent sound integrated). So given the fact that most new mobos have 1 or maybe 2 PCI slots, which may or may not even be avail. if u use dual graphics, this should never work.

[Thorry]

http://img367.imageshack.us/img367/633/unitsdx6.jpg

THE HARDWARE IS FINALLY CATCHING UP!!! WOOT

JUST IMAGINE 1000's if not millions of insects or animals in a game all acting realistically

If you ask me, I would predict that the first REAL game will be out just a little while after this new hardware (PPU and AIPU) prove themselvs to the public, around 2009 I would guess. Think about it: fully real physics, photo-realistic graphics, amazing sound quallity (like 22.4 or something as opposed to 5.1) and mart AI, almost at a human level, basically... think The Matrix. Scary stuff, CANT WAIT WOOT!!!

To get perfect sound you don't need more speakers. You actually need only 2 (and a subwoofer for those nice bass sounds). It's all about timing, you only have 2 ears, the 3D feeling is because of small timing differences between the left and the right ear (as well as filtering, for example sound from the rear gets filtered by the ear, about a 1000hz low-pass filter produces the same effect).

To get true 3D sound you only need 2 speakers at the right location (or preferably in your ear) and some perfect timings (with calibration for each user). Then some processing hardware to make sure the timings are perfect within the game and you're sorted.

More speakers only complicates things whilst you still only have 2 ears.

[Thorry]

@uOpt:

A FPGA would be a solution. A high performance high gate count FPGA connected to the hypertransport bus could be just the thing we need.

When you play a game the game programmes it to be a AI accelerator, when you run photoshop it's a filter enhancer. When you run Vista it becomes additional memory :P

Just add 1 or 2 good FPGAs to the system together with a Cell-like CPU and we'll be sorted. But untill then expansion cards are our best bet.

[Order]

A FPGA would be a solution. A high performance high gate count FPGA connected to the hypertransport bus could be just the thing we need.

Torrenza/Intel's response to Torrenza

[CBONE]

That video was pathetic. What games are moving in this direction? Who wouldn't be pissed if the Nazis won because their troops got stuck taking the long way around some bricks that got knocked over? Multi-core processors at multi-gig speeds aren't fast enough to do simple searches on the fly? I suppose if you wanted to make your potential pathways stupidly complex, but why would you do that?

What they need is to forget about this expensive add-in card foolishness because vastly more important video is taking up all the cash and space. Put these things on a USB or firewire device for less than $100 and we can talk.

[BeyondSciFi]

We all have to agree that all of todays games suck ass... We all pretty much get bored of them rather quickly and look for other games. Why is that? I think this is mostly due to the fact that you basically do the same thing over and over again. A truely great game is one that changes as the user does. For this we need better AI (among other things, graphics, physics, etc). We want to feel like we have free will in the game, instead of following preplaned paths. Just think about it, say you are playing a FPS on a Martian City, but rather then shooting the monsters, say you wanna take a ride on a Space Ship around the planet. Can you do this? NO! (unless is a mission, which ruins the whole point) What im getting at is; it shouldn't really matter HOW AI, graphics, sound, physics, etc are done, as long as they done so that they give you greater degrees of freedom to do whatever you feel like. THIS IS WHY EVERYONE LOVES GTA! You can do the missions, or do anything else you want, to a certain degree... IF dedicated processors or if multicore CPUs are the best way to achive this kind of freedom, why fight it? After all we all just want more freedom, like in the real world, why should games be any different? Everyone always comes up with the price or incompatabilty argument, but theres are just the first hurdles that the new technology has to surpass. After it has done so, price wont be an issue since you can always buy the low-end or high-end solutions. Incompatabitly will be a thing of the past as the industry has adapted.

Just my opinion, but I think ultimately most of us feel the same way.

The end result should matter most in this case, arguing about the details is just silly, unless you are part of the development process.

[marauder16]

BeyondSciFi:

as long as they done so that they give you greater degrees of freedom to do whatever you feel like. THIS IS WHY EVERYONE LOVES GTA!
- this is so true, I also love GTA cause you have three big cities to explore and you can do whatever you want. :D
It would sure be better if all these AI and physics tasks would all be performed by the CPU but if the add on cards are the only way I'd accept it. freedom and realism is what games need today.
But also I think that with new technology you should be careful and not jump on it ASA it launches, leave people to test it and then buy it, just my .02 :)

[uOpt]

i belive this has real potenail!!!!!! because most people have to admit ai are stupid!!!!!!!! and inorder to get the same effect as an aipu would you would need one heck of a cpu because the cpu is not made for that and that alone so it is in effient at it.

But it is the other way round, hipno :(

The physics cards and AI cards only do extremely limited (read: naive) AI and physics and cannot be changed to do more clever AI or physics.

If you would run pure software on a general-purpose CPU then you write new software and you run it. There is nothing you can do about the software limitations of these AI and physics cards.

So they lead AI and physics to be more stupid, not less. The only advantage is that you can more of the same stupidity, which is good to drive a lot of things on the map and the like. But stupid it will stay.

[uOpt]

@uOpt:

A FPGA would be a solution. A high performance high gate count FPGA connected to the hypertransport bus could be just the thing we need.

When you play a game the game programmes it to be a AI accelerator, when you run photoshop it's a filter enhancer. When you run Vista it becomes additional memory :P

Just add 1 or 2 good FPGAs to the system together with a Cell-like CPU and we'll be sorted. But untill then expansion cards are our best bet.

But do you think a FPGA could offer the performance we want? A single one might be in the noise when the box has a dual-core ninja-macho CPU anyway.

It is an interesting thought. But I have no idea how fast they can operate these days.

I am just saddened that the gaming industry refuses kicking and screaming to do the right thing and make full use of multicore systems. And that is after the PS3 wants the same thing, too :(

[XSAlliN]

umm didn't we get this posted in the news section a little while ago?

Ooops sorry about that, I have found it on a IT Magazine and i foght it would be nice to share, wasn't here when the last news was posted, in the future I'l use "Search" (if I remember).

Hope they don't make another hardware component like they did whit Ageia.

[Thorry]

But do you think a FPGA could offer the performance we want? A single one might be in the noise when the box has a dual-core ninja-macho CPU anyway.

It is an interesting thought. But I have no idea how fast they can operate these days.

I am just saddened that the gaming industry refuses kicking and screaming to do the right thing and make full use of multicore systems. And that is after the PS3 wants the same thing, too :(

Well at the moment it won't offer the performance we need (although the performance is quite good). Also there aren't a lot of very big FPGA's so the complexity is a bit limited.

However this is just a matter of scale, if you start producing these things in great quantities and put some R&D into it it can surely be at a level acceptable.

[uOpt]

However this is just a matter of scale, if you start producing these things in great quantities and put some R&D into it it can surely be at a level acceptable.

They are produced in great quantities, just not for PC use. I seriously doubt that you can speed them up to even 1/10th or a modern CPU core.

[hawkeyefan]

dumb question from the gallery;)

So instead of zillions of coprocessor add-in cards, why don't we just cram the whole works into the CPU:confused:

say with native 8 core CPUs, 4 could be PPU,AIPU,APU,GPU...rest remain general CPU:stick: Is this practical, or even possible?

[BeyondSciFi]

dumb question from the gallery;)

So instead of zillions of coprocessor add-in cards, why don't we just cram the whole works into the CPU:confused:

say with native 8 core CPUs, 4 could be PPU,AIPU,APU,GPU...rest remain general CPU:stick: Is this practical, or even possible?

We "could" cram them all on a single CPU die but the problem it that its way too inpractical. Besides, the power required from the GPU and other dedicated processors is much more then the CPU can offer. Let me put it in perspective; a core 2 duo has about 20-30 GFLOPS while the average "good" GPU (x1900 or G7900) have 400-600 GFLOPS of power. The thing about the CPU is that its a multitasker, it can do pretty much everything, but nothing well. Raw power if need for highly paralleled tasks like graphics, sound, physics, and AI.

[phi|os]

Who cares at this point. It's not out. And if it behaves like our PPU's that are out now, we really should disregard this processor. But hey, if it makes :banana::banana::banana::banana: better, I'm all bout it.

By the way, that little orgy diagram of all the processors linked together might work a little better if PPU's, APU's, and AIPU's usually didn't amount to piece's of crap. Quad core " one core to unite them all" lol. Isn't this why we are putting mulitple cores on one die...and eventually DAMMIT's marriage of GPU and CPU, on a multi core processor should really clear all these little wannabe's away. Less clutter = efficiency. Same goes with everything in this world. Keep it simple.

[eXceeded]

Who cares about any of these please tell me any of that you that have enough space for this card (PCI), an AGEIA PhysX card (even the unreleased PCIe version) and a sound card (PCI) and a dual graphic scard combination (PCIe) and now even the KillerNIC card (PCI). You don't even need to have a dual graphics card system to see that you won't be able to use all of these add in cards... So from my point of view the only possible way around all this is to put it all into the motherboard/cpu. I think its time that the motherboard became the motherboard again tbh.

[[XC] hipno650]

physcis will now be taken care of by and only using part of the gpu power. i would like to see a universal spefication ai and physics that could run on the cpu, gpu or the most high end would be dedicated add in card and you be able to control the realness or amount just like on graphics how you can adjust the detail to suit your system. that would be good!

[Thorry]

They are produced in great quantities, just not for PC use. I seriously doubt that you can speed them up to even 1/10th or a modern CPU core.

1/10 speed is certainly possible, but not needed at all.

Remember the FPGA can be programmed by each application to do exactly what is needed by the application. All those physics cards don't really have a fast processor (compared to the CPU) but because they are designed to do one task they are very fast (much faster than a CPU for that one task).

A CPU is designed to do everything, that makes it very slow compared to chips that can do only one task, but also very usefull.

At the moment FPGA of the size we are talking about aren't produced in great quantities. FPGA chips themselves aren't produced in the quantities needed for global widespread use in computers.

For my work I fool around with Xilinx FPGA so I know what they can do.

Mark my words, we will see a FPGA with a hypertransport or PCI-E connection in the near future for use in computers. On the long term they will become standard.

Just check out this beast: http://www.xilinx.com/products/silicon_solutions/fpgas/virtex/virtex5/capabilities/index.htm

And those prices are just for now, turn it into a high performance consumer product and we are talking around 75$, that's not bad.

[XSAlliN]

I don't care how they do it but they need to improve NPC's AI - you have to admit that most of the NPC's from games act real stupid, so instead of 3 smarter NPC you get an army of stupid Bots with limited moves + the rest like nature and weather.

[STEvil]

I find it rather amusing that NPC's today are no smarter than they were when we started.

Is it really so hard to have an NPC analyze vertex data?

[Entity_Razer]

JUST IMAGINE 1000's if not millions of insects or animals in a game all acting realistically

If you ask me, I would predict that the first REAL game will be out just a little while after this new hardware (PPU and AIPU) prove themselvs to the public,
Think about it: fully real physics, photo-realistic graphics, amazing sound quallity (like 22.4 or something as opposed to 5.1) and mart AI, almost at a human level, basically... think The Matrix. Scary stuff, CANT WAIT WOOT!!!

Euh... NO

Seriusly... this is "just" a downscaled CPU with a builth in AI engine behind it. The thing will still suffer from bad code if the programmers don't code the AI deep enough and will still be bad if the code isn't made to support the scene :s

I know why a PPU would be good since Physics math and normal rendering math isn't really the same, but a AI Processor? WTH?

AI is just a way of coding, nesting codes, and using a good randomizer/prediction unit, thats about it, slap those on a few memory modules, slap'm to a PCB, add a low budget 500Mhz CPU to it, and BAM Instant AIPU....

[Thorry]

Euh... NO

Seriusly... this is "just" a downscaled CPU with a builth in AI engine behind it. The thing will still suffer from bad code if the programmers don't code the AI deep enough and will still be bad if the code isn't made to support the scene :s

I know why a PPU would be good since Physics math and normal rendering math isn't really the same, but a AI Processor? WTH?

AI is just a way of coding, nesting codes, and using a good randomizer/prediction unit, thats about it, slap those on a few memory modules, slap'm to a PCB, add a low budget 500Mhz CPU to it, and BAM Instant AIPU....

Neural networks ;)

A good deticated specialized processor does help a lot in calculating trough large neural nets.

[Entity_Razer]

Neural networks ;)

A good deticated specialized processor does help a lot in calculating trough large neural nets.

Agreed but what use is a AIPU with a good neural networks chip when, a lot of company's do not want to invest the cash/resources to have a few people work on it or at least 1 good programmer on it? NPC's don't always need to be THAT smart and i don't really see people taking advantage of it any time soon. They want NPC's to be smart but not THAT smart. I see people here complaining about NPC only having a few moves

---> Dev team's fault, not AI's

NPC's not smart enough

---> again dev team's fault, not AI's. A good processor optimized for it WILL speed up the calculations for it but again, you still need the dev team to take full use of it, resources which in the industry are often shifted towards graphics or gameplay needs, not really AI needs.

I agree that in case of simulation traficjams for instance and finding bottlenecks/way to solve trafficjams, like the "smart lights" in downtown london use the use of a "AIPU" is a good thing, but for things like a computer game i really don't see the need for it.

Not to mention, a neural net processor, optimized for just that sounds pretty "new" the cost for R&D to make it slim/downscaled for normal comps, yet enhanced enough to do anything a game could need sounds quite expensive still so i do wonder about the price.

Thats why i really am afraid this will just be a co processor with a AI engine solderd to the PCB and slapped on a PCI card... With optimization in the prediction unit and probability calc etc ofc but.... the idea of developing a neural net processor for gamers sounds quite expensive and pretty far fetched imho

[Thorry]

Ah well, it's just a matter of time. If the market demands it there will be a supply for it automatically.

The nice thing about neural networks is that it can be trained to emulate group behavour. So a single NPC would be completely dumb and stupid, the group on the whole acts like you would expect them to.

That way it doesn't require much resources but will still supply a great gaming experience.

For now people seem to be focussing on more vertexes and more shader effects at higher resolutions ... :(

[uOpt]

Neural networks ;)

A good deticated specialized processor does help a lot in calculating trough large neural nets.

Neural networks actually have a very limited range of capabilities, they can never come up with new tricks. Not to mention you have to train them. For a game there's no time to have a 1000 cpu datacenter run a months of AI versus AI battle to train the neural network to the maps.

And even then the results will be disappointing since a humaan player can most likely quickly identify a systematic weakness in the mapping of the neural network (means: what it uses for input and what it can do on the output) or a weakness in the traning.

Example: there's a map with a corner and while training the neural network in AI versus AI mode the attacker that has to move over the map (that is who will later play against the humans) figures out that this corner of the map is never covered by enemy fire. So the neural network heavily biases it's behavior for that map and always moves through that corner. The human player sees that for the first time, sets up a machinegun/ninjablaser/massdrive/gizmogun covering the corner and the whole effort is going to the dump.

I rate neuural networks as not practical for game AI in general. There will be specific areas where it does better, though, but then it is questionalble whether it's worth the trouble.

BTW, even back in the 80ties we had nubus/sbus/ISA cards with neurral network coprocessors. History always repeats itself :)

[uOpt]

I find it rather amusing that NPC's today are no smarter than they were when we started.

Is it really so hard to have an NPC analyze vertex data?

No, but you have to hire a person who can do it.

I play wargames (as in lousy-graphics games simulating historical conflict) and in that community the battle is never ending. Can you make the AI just a little smarter? Pretty please?

The developers are very clear on this: it is not worth it.

For the majority of customers the dumb AI is good enough, either because they are not much smarter themself, or because they use the game as a slideshow watching 3D models and explosions.

For the minority of customers who has a shelf full of lectures on the Soviet Operational Art of War the AI is not good enough, granted. But even if you would put in a massive effort into making the AI better they would still think it is too stupid. They play human opponents. They have unrealistic expectations about what an AI can do.

%%

Considering the cost of AI, it is considerable. The gaming business is a special place for programmers. Very stressful, sometimes not guaranteed payed well, lots of unreasonable publisher pressure. No vacation for two years, then 6 months forced vacation etc.

Game developers do it anyway, usually because they lobe games, and they like leaning back Friday evening (or Sunday morning 3 a.m.) and watch a bunch of stuff happen in a 3D worls that they put together that week.

The type of programmer who can write a good AI is often not in that mindset. He's not overly thrilled about games, so why pick that job over the google offer?

So you'd end up hiring somebody for a lot of money who might bail out. Replacing him would be even more painful. And all that for no gain in game sales. No way.

%%

Personally I think at least wargame developers are stupid. They should write an AI in a DLL and let crazies like me implement it. But I've been preaching this to gamegame developers for years and they just don't get it. They need to "protect" their engines from outside inspection and they have the wrong idea about what of the game mechanics needs to be exposed to a competent AI (all of it, but they don't believe it because their AI only uses a minor fraction).

[[XC] leviathan18]

i dont need smart ncp i want smart storylines.....

[uOpt]

I don't care how they do it but they need to improve NPC's AI - you have to admit that most of the NPC's from games act real stupid, so instead of 3 smarter NPC you get an army of stupid Bots with limited moves + the rest like nature and weather.

You put the point I was making into much better words than me.

It is what the customters wants according to the marketing uncles, though.

[uOpt]

dumb question from the gallery;)

So instead of zillions of coprocessor add-in cards, why don't we just cram the whole works into the CPU:confused:

say with native 8 core CPUs, 4 could be PPU,AIPU,APU,GPU...rest remain general CPU:stick: Is this practical, or even possible?

I think that is where we are going.

Right now the technology isn't that advanced.

Another "advantage" of using coprocessors only for AI or only for physics is that the demand for data transfer is limited. So you can connect them via PCIe or even PCI. With a true coprocessor you cannot because the amount of data transfer volume is too big.

But that also limits these on-card coprocessors to do really stupid things on their own without really working with the game engine.

[XSAlliN]

I think that is where we are going.

That's for sure (I hope) ;)


Right now the technology isn't that advanced.

Not quite, actualy the technology is more advanced then we think but it's to BIG and to expensive for home users, I'v seen and advanced AI System for traing pilots but it's to big to fit in a home and costs milions of $, in conclusion even you have a point because it's not that small (advanced) to fit a normal PC, that's why I personaly see a future in Intia Processor or something similar, just hope I can see a result in near future ;)