JC Keynote talks consoles

[Titanio]

1st party games will need to operate at 1080p.

Carmack ain't a first party

But really? Where did you hear that?

 

[blakjedi]

I think he's been taken out of context and misrepresented with some of the reports and quotes going around. Some quick thoughts:

1) He did not say that IBM made a misstep with the Cell design as is being reported by some, he said takes issue with IBM's contention of how that power should be used. They say it should be used for physics and AI since graphics is "done". Carmack obviously disagrees. And if Carmack wanted to use any CPU's power for graphics, I think he'd be better off with Cell regardless. But I don't think he's not saying he wants to do that. He was simply using that comment as a jumping off point to assert the primary importance of graphics.

I think he was saying IBM made a misstep in their design because their design intentions and priorities (AI, physics) are missplaced. Basically he's saying why design for physics and AI which cant be seen but can be experienced in limited quantities, when you have not maximized the visible representation of the game world yet?

2) He also did not say that physics was unimportant or unnecessary as such for games, or at least in the way that was being portrayed. He's saying that if you want to take a pure simulation route, you're going to find it much more difficult to control what happens, and thus ensure a good game experience. Some people earlier were making the point that physics can contribute to the eye candy, so why would Carmack think it wasn't important if he thinks graphics and presentation is important? He actually does say it can be used in that manner to make things look better. Physically based visualisation doesn't have to upset the apple cart as far as game design is concerned, and he points at that - liquid water physics, smoke that behaves realistically etc. etc. So tying physics to visuals is useful as far as he's concerned. But he does tend to make it seem less important than just graphics alone, which is flatly contradictory IMO. He talks physics down to a degree as being mostly relegated to that - unless you're feeling lucky/ambitious - but graphics is "just" about presentation too and he seems keen on that. He concedes that point, although he glosses over it quickly, and he does admit he's not a physical simulation guy. He also does admit that this is something that requires a lot of power regardless of whether you go a pure simulation route or not. And of course, even if Carmack doesn't feel comfortable making physics a lynch-pin of the gameplay, others may (and others arguably have already).

I think he was saying that for all intents and purposes game coders can through game code reproduce realistic enough physics and AI in a way meaningful for the player. Who needs a chip which simulates globally correct physics when the playable game world is limited to the players vision cone at any given time (thus the turn your back on physics commentary). In other words..."Once your not looking at it, who cares?" and then why go through the trouble of extra programming for stuff the player cant/wont see?

3) With regard to AI, I think he's right in that it's as much about what the player perceives as what the characters are actually doing. But purely directorial approaches just don't work, or at least his own examples don't. Doom3's "directed" AI was horrible IMO. Maybe he thinks most people don't notice, but I do, and I'm sure I'm not alone.

He's right IMO. I notice bad AI cause I'm a person. But because I accept a videogame for what it is - simulacra - I suspend my disbelief such that I'm willing to accept NPC "stupidity." In many cases if the designer thinks like me he can place and direct bots in certain situations such that I really have to think in order to beat them in a way which isnt pattern/time based.

4) His comment about perhaps being in a better position next-gen with multi-core etc. is of course true, but if the current systems were all OoO as he ponders, that wouldn't be the case. You gotta start sometime.

Agreed.

I also thought his comments on HD were interesting. He flat-out said that while enforced minimum resolutions may be OK for now, with Quake4 etc. with his next-gen rendering tech, he'd prefer to do more complex per-pixel rendering with a lower resolution vs having to cut that to meet a higher resolution. It'll be interesting to see Sony's policy on enforcing minimum resolutions or not. Nintendo might also get some credibility out of that aswell

Very good analysis Titanio... thank you for posting this excerpt.

 

[Titanio]

I think he was saying IBM made a misstep in their design because their design intentions and priorities (AI, physics) are missplaced. Basically he's saying why design for physics and AI which cant be seen but can be experienced in limited quantities, when you have not maximized the visible representation of the game world yet?

I have to completely disagree. He doesn't mention design anywhere, heck I don't think it even matters that it was Cell engineers he was talking with. He was addressing the use of that power. And it wasn't a point specifically about how the CPUs will be used - obviously 2 big things they're going to be doing is physics and AI - but how power in general is used. Unless he believes CPUs should be doing graphics work too, which I don't think is his opinion - but if it was, Cell would be better from that perspective anyway than other CPUs. So I certainly don't think his issue was with the design of Cell. Just with the notion that graphics has now levelled out - he simply used the IBM comment as a springboard to discuss that.

I don't think his issue is with CPUs being biased toward physics and AI, I don't think it's his claim that these things won't take more power. I think his issue is with the notion that the focus in power *generally* should switch away from graphics to them (i.e. within systems, for GPUs to become less important, and CPUs more important or whatever). He doesn't expect CPU designers to go make their chips better for graphics (although that, ironically, is something that IBM did with Cell).

I think he was saying that for all intents and purposes game coders can through game code reproduce realistic enough physics and AI in a way meaningful for the player.

True - although not everyone would agree - but then he recognises the value of physics to presentation and making things look better. Which is something he started out saying was so important, and hence iD's focus on graphics. That dilutes the argument a little, though he seems to indirectly recognise that contradiction.

Who needs a chip which simulates globally correct physics when the playable game world is limited to the players vision cone at any given time (thus the turn your back on physics commentary). In other words..."Once your not looking at it, who cares?" and then why go through the trouble of extra programming for stuff the player cant/wont see?

His point is actually that if you want to do fully correct simulation that fundamentally affects gameplay, you actually HAVE to process everything, all the stuff going on behind your back etc. If the physics were simply there to make things look nicer though, without a really big effect on how things play out, you could turn down the detail on things you're not looking at. But in that instance, it just leaves even more to use for what you can see

In terms of using physics to affect gameplay etc. - something he's cautioning about going overboard with - it's interesting that he talks about "directorial" AI but doesn't think the same could apply with physics. He talked about one bad thing is that free physics could end up blocking passage i.e. getting in the players way of going somewhere. I'm sure you could also "direct" the physics to a degree such that that isn't possible, or in that specific example, so that it is not possible to block key paths - localise the physics etc.

He's right IMO. I notice bad AI cause I'm a person. But because I accept a videogame for what it is - simulacra - I suspend my disbelief such that I'm willing to accept NPC "stupidity." In many cases if the designer thinks like me he can place and direct bots in certain situations such that I really have to think in order to beat them in a way which isnt pattern/time based.

The game needs to do enough to suspend your disbelief however. I can't just accept anything that happens just because it's a game. This type of approach did not work so well in Carmack's games unfortunately, but this possibly speaks more for iD's implementation than the value of the method generally (this method can work well in some games, particularly those of iD's ilk if done properly, but other games I think could benefit from more intensive/complex AI). Spawning bots into dark rooms behind you and watching enemies run parallel to walls before very belatedly turning their attention to you, despite clearly having seen you for a relatively long time, is bad AI imo.

 

[blakjedi]

I have to completely disagree. He doesn't mention design anywhere, heck I don't think it even matters that it was Cell engineers he was talking with. He was addressing the use of that power.

And it wasn't a point specifically about how the CPUs will be used - obviously 2 big things they're going to be doing is physics and AI - but how power in general is used...
So I certainly don't think his issue was with the design of Cell. Just with the notion that graphics has now levelled out - he simply used the IBM comment as a springboard to discuss that.

Umm on the one hand youre right.. How the power of the new CPUs is utilized is his main focus. However, I get the impression that in his conversation with IBM engineers they tried to convince HIM that the design of the Cell was best utilized for advanced physics and AI. The design in their opinion reflected that and JC has a problem with physics and Ai as a CPU design focus when Graphics arent where they should be yet. In the end we may be saying the same thing, but i think that Carmack DOES have a problem with the design of Cell and XeCPU because their design reflects a focus on accelerating things he isnt necessarily worried about yet (physics and AI).

I don't think his issue is with CPUs being biased toward physics and AI, I don't think it's his claim that these things won't take more power. I think his issue is with the notion that the focus in power *generally* should switch away from graphics to them (i.e. within systems, for GPUs to become less important, and CPUs more important or whatever). He doesn't expect CPU designers to go make their chips better for graphics (although that, ironically, is something that IBM did with Cell).

True - although not everyone would agree - but then he recognises the value of physics to presentation and making things look better. Which is something he started out saying was so important, and hence iD's focus on graphics. That dilutes the argument a little, though he seems to indirectly recognise that contradiction.

His point is actually that if you want to do fully correct simulation that fundamentally affects gameplay, you actually HAVE to process everything, all the stuff going on behind your back etc. If the physics were simply there to make things look nicer though, without a really big effect on how things play out, you could turn down the detail on things you're not looking at. But in that instance, it just leaves even more to use for what you can see

Well stated.

In terms of using physics to affect gameplay etc. - something he's cautioning about going overboard with - it's interesting that he talks about "directorial" AI but doesn't think the same could apply with physics. He talked about one bad thing is that free physics could end up blocking passage i.e. getting in the players way of going somewhere. I'm sure you could also "direct" the physics to a degree such that that isn't possible, or in that specific example, so that it is not possible to block key paths - localise the physics etc.

I think localising physics is all he was talking about. In that case having the extra processing power available in order to simulate physics properly is great but it shouldn't be a focal point of your programming. Frankly I think the issue is a non-starter... At most you might want to process the physics of weather and night and day progression universally...because you are always subject to their effects... Much like what the programmers behind Resident Evil 5 were alluding to.

Beyond that if your not looking at the waterfall or at the river... the fact that their physics is propagated accurately is immaterial... but if the water in the sewer you happen to be in does pool and splash properly that can dampen your experience... or not! It all depends on whats happening at the moment on the screen.

The game needs to do enough to suspend your disbelief however. I can't just accept anything that happens just because it's a game...Spawning bots into dark rooms behind you and watching enemies run parallel to walls before very belatedly turning their attention to you, despite clearly having seen you for a relatively long time, is bad AI imo.

True... but most AI is not really "interactive" in terms of intelligience but rather interactive in terms of fighting. Simple moves like ducking behind objects to avoid fire have been staple for a while. Circling an enemy and doing coordinated attacks are also there... which are basic normal moves of an opponent in any melee or ranged weapon attacks. Thats what REAL people (and hyenas! ) do.

Evasive, retreat maneuvers, ambush tactics which are counterintuitive but effective, have been mildly successful in both games and real life which is why they are seldom used if at all over brute firepower... these tactics however would add greatly to the impression of extensive/complex AI.

It would be great imo if games stopped the "unlimited enemy" syndrome and actually set limits to the amounts and nature of enemies that could be encountered in a game world. The idea of enemy communications and reinforcements would be a great opportunity to increase challenge instead of "enemy generators." So many ideas so little skill!

 

[seismologist]

After playing Doom 3 I really wonder if Carmack is qualified to speak on what's best for gameplay.

"Carmack notes with some irony that in the same way id Software raises the graphical bar for games, other companies are raising the standards for physics, so that id has to integrate simulations into their games or they look dated."

That's a pretty telling indicaion that ID's games are already falling behind the curve...

 

[scooby_dooby]

i don't know why more processing power is really required to make AI that improves gamplay.

i think alot of games could take a huge lessons from Halo 1 and Halo 2, especially on legendary settings.

The AI was extremely smart, they would hide, attack in groups, have scouts, sneak up on you, it was by far the funnest part of the game to me.

Games like Brothers in Arm 2 are also showing very cool AI where the enemy counter attacks as you attack and reacts to your actions, i.e. if you rety to flank him he'll flank you at the same time and you'll actually end up reversing positions on the battelfield. or if you fail to suppress properly they will flank you themselves.

Both these games are the two most advanced AI system, I've ever seen in action and they are both xbox1 games. I don't think it has much to do with "power" as it does with programmer skill and the priority AI is given in the development process.

i mean, the power would come in handy when you have many enemies on screen at once, but even in current gen with the 5-10 people on screen you typically get, most games had horrible AI, it wasn't from lack of power just lack of quality programming.

it seems with the next gen now they can take that same crappy AI and give it to 1000 characters on screen at the same time, but it's still going to be crappy until they start writing GOOD AI routines.

 

[London Geezer]

i don't know why more processing power is really required to make AI that improves gamplay.

i think alot of games could take a huge lessons from Halo 1 and Halo 2, especially on legendary settings.

The AI was extremely smart, they would hide, attack in groups, have scouts, sneak up on you, it was by far the funnest part of the game to me.

Games like Brothers in Arm 2 are also showing very cool AI where the enemy counter attacks as you attack and reacts to your actions, i.e. if you rety to flank him he'll flank you at the same time and you'll actually end up reversing positions on the battelfield. or if you fail to suppress properly they will flank you themselves.

Both these games are the two most advanced AI system, I've ever seen in action and they are both xbox1 games. I don't think it has much to do with "power" as it does with programmer skill and the priority AI is given in the development process.

i mean, the power would come in handy when you have many enemies on screen at once, but even in current gen with the 5-10 people on screen you typically get, most games had horrible AI, it wasn't from lack of power just lack of quality programming.

it seems with the next gen now they can take that same crappy AI and give it to 1000 characters on screen at the same time, but it's still going to be crappy until they start writing GOOD AI routines.

You know what the problem really is,

It's that whatever super duper advanced AI engine routines extra plus they use, computer characters will always feel stupid to us. If not "always" then "for a long time".

So i think that for now developers choose to focus on the eye candy, because whatever new advanced AI games have, they will always feel too simplistic to our brains. We're just too clever compared to computer generated stuff - even non-realtime, so realtime stuff is even in a worse position.
Obviously things will get better, and they are getting better, but think about it, why spend lots of processing power for AI to generate characters that ultimately will still feel unnatural, when you can use that power for other things like better animation and physics and have just slightly more unnatural characters, which will move better though...

Proper AI, like proper anything else, takes a lot of processing power.

 

[scooby_dooby]

that's definately a good point.

So do you think anything has really changed this generation? It sounds like they still have limited power for eye candy, and even more constrained budgets(much higher art costs) so they will have to make the same decisions that they made last gen.

 

[Titanio]

However, I get the impression that in his conversation with IBM engineers they tried to convince HIM that the design of the Cell was best utilized for advanced physics and AI. The design in their opinion reflected that and JC has a problem with physics and Ai as a CPU design focus when Graphics arent where they should be yet. In the end we may be saying the same thing, but i think that Carmack DOES have a problem with the design of Cell and XeCPU because their design reflects a focus on accelerating things he isnt necessarily worried about yet (physics and AI).

The thing is, what would he rather the CPU have accelerate? Hardly graphics. I don't think he has a problem with the CPU design being focussed on AI or physics, but he would have a problem with the system as a whole being focussed on those things at the expense of graphics, if system designers started thinking that graphics was "done".

The funny thing is though, Cell does do better with graphics than the typical CPU.

So I don't think his point there was about the focus of CPUs specifically. Rather what more power is needed for generally (across the system, CPU, GPUs everything). And he was making the point that graphics still have a long way to go, and need a lot more power. But I'm sure he does not expect CPUs to bring him that power, I'm sure he's relying on GPUs for that.

 

[RaolinDarksbane]

It seems the guys at Bethesda aren't too worried about it:



http://www.telefragged.com/interviews/oblivion/

You sure? Is this Todd Howard guy a programmer or is he speaking for the programmers? Their games weren't programmed that well too... remember Morrowind/PotC on the PC and XBOX? Oblivion looks good now, but most of us said the same thing with Morrowind before it was released.

 

[mckmas8808]

But I realise things like the basic boxes falling down, knocking things off, bouncing around the world, ragdolls, that's all good stuff for the games, but I do think it's a mistake for people to try and go overboard and try and do a real simulation of the world, because it's a really hard problem, and you're not going to give really that much real benefit to the actual gameplay on there.

I couldn't disagree with him more. The guys from Epic stated that with a CPU like CELL like 15,000 pieces could be calculated at once compared to around 500 today. How can physics the way he is describing not help gameplay?

If I was a number one kind of programmer I would never want to work for someone like Carmack. He doesn't have the vision that I like in a leader. So much complaining, but no good things to deal with these problems.

He just complains and says don't worry about super duper physics, don't worry about super duper AI because the gamer doesn't see it anyway so why work your butt off. What was the universal score of Doom 3 like a 7.0. Dude you make games for PCs, talk to devs from Blizzard, Ninja Theory, Kojima Productions, Polyphony, etc.

 

[Guden Oden]

I'm not stretching at all. Its you thats stretching Unless you believe that developers just use the force to know what code will excell on what chip and how to push them to thier upmost potential and ot optimize the code to go over 7 cores

Damn man, you are looking for things to nag about aren't you?

We've been over this many times before. You don't program specifically for the particular SPEs, the model is you write your code, and the compiler and OS schedules it for you. Naturally some code will run better than others like with any CPU architecture but it's not neccessary to manually find work for each and every chip in the system.

As for very telling . Its telling that you jump to the ps3s defense yet say nothing about the xbox 360 .

I'm not jumping to its defense, stop fucking telling me what I mean with my own posts dammit. That you think I'm jumping to its defense shows YOU are the silly fanperson here.

Why would I need to say anything about the x360 when even you yourself said the xcpu is an uniform design? This issue doesn't even apply to that system, I don't understand why you think I need to say anything in regards to it, or what that would be!

If i was wrong you should be correcting me about both not about your fav system . Which is obvious by your postings

Don't pretend to know better than me what my "fav system" is, little man.

 

[Farid]

Warning before "the lock"

This is exactly why people should not do personal remarks about other people's supposed bias...
Because it does nothing but turn any discussion into a personal attacks fest.

I'm not locking this thread now, because the thread had a good debate going on, so I'll just ask jvd and Guden to leave it at.

 

[MfA]

Developers seem not so much worried about being denied performance, but moreso scared at having the responsibility on their shoulders of makeing use of many more oppurtunities.

JC is actually honest about that though.

 

[Titanio]

So ignore half or a third of the processor ... if it's really all about the GPU does it really matter that the processor is a little slower than it could have been for single threaded applications?

I think it relates back to his "raising the bar" concern.

He's a graphics guy. I think he'd prefer if games were always graphics focussed. But as he says himself, now there are "physics guys" coming along, raising the bar for everyone, forcing him/devs to spend more time with physics software. I get the feeling that's not what he wants to be doing. These console CPUs reflect a future where devs will contine to raise the bar in areas other than graphics further and further, but for Carmack personally, I don't think he digs that. So he could just work on one core and ignore the rest of the chip, but he knows other people won't, and they'll be "raising the bar" - forcing him to keep pace also. So he can't really ignore it.

He actually went into a little detail about how this will look for his engine. How they're approaching it is: 2 threads on one core, one for the game, one for the renderer. And then they'll use the other cores with "targets of opportunity" - physics etc, the things that can be parallelised easily I assume. So worker threads on the other cores will kick in when appropriate to handle things that can be more readily made parallel and so forth.

 

[MfA]

Yeah I didnt really read that bit correctly, as usual I edit my posts after writing them :/

 

[mckmas8808]

Did JC really say this Titiano?

The next-gen consoles are about as powerful as current high-end pc's, but their cpu power is current PC processor

If so I in no way can believe this.

 

[Titanio]

Did JC really say this Titiano?



If so I in no way can believe this.

He did say they were about as powerful as current high end PCs, yes. He later said they were all in the same "ballpark-ish". I'm not sure what perspective he was coming from with the former at least, but yeah, it certainly is one of the odder comments (and frankly isn't true - there's no GPU on the PC market as powerful as RSX or Xenos, and the CPUs, while simpler in design and more difficult to harness, offer MUCH more performance in certain areas. Maybe he just disregards the CPU because they're not running his X86 code as well - he later referred to the CPUs again being "twice as slow", as if that was the general case or something, and maybe he just sees the gap on the GPU side as being smallish). I think he must simply mean just in terms Quake4 performance, where he's making no effort to design for them, rather porting from PC.

 

[Gabrobot]

I couldn't disagree with him more. The guys from Epic stated that with a CPU like CELL like 15,000 pieces could be calculated at once compared to around 500 today. How can physics the way he is describing not help gameplay?

You can't simulate a world with that. And it doesn't help make designing such a world easier. Great you can do a shitload more physics objects...what exactly does that do for gameplay? I don't think you really understand these design issues at all...

If I was a number one kind of programmer...

...

 

[Titanio]

You can't simulate a world with that. And it doesn't help make designing such a world easier. Great you can do a shitload more physics objects...what exactly does that do for gameplay? I don't think you really understand these design issues at all...

Yeah, but it can make things look a lot better, something Carmack is otherwise preoccupied with. Physics doesn't have to impact the core gameplay (although certain games have proved that it can do so positively, but obviously that's going to vary game-to-game).