• Hey Guest. Check out your NeoGAF Wrapped 2025 results here!

"The Power of the Cloud" - what happened?

This was done on remote servers and it was a early build of Crackdown and not a proof of concept.

Alright then. I believe in the concept of remote server computation. I'm just not sure if I believe in the internet infrastructure that is currently in place being able to support that consistently for enough of the audience to enjoy it.

Call me callous, but until there is an out in the open non-to-lightly-controlled demonstration, things like this are still a proof of concept in my limited opinion.
 
While multiplied by the number of people playing the game at once, especially on launch.

Exactly. It's not like Live is bullet proof.

But you were in the thread talking about the tech wasn't needed and Crackdown shouldn't be used to demo a new tech. I don't recall you having this attitude in that thread, but who knows with you

I've had the same attitude towards a cloud powered CD from the beginning. I've expressed a number of times how worried I am about the game using cloud powered tech when they could just go with an offline solution or just not have destruction as it's not something CD needs.

I haven't been as openly critical because I honestly did start to believe that Kampf was legit (claims of verified helped with this) and wanted to see if his claims had any basis before ripping into it completely. I did want to believe, but after he was exposed as a hype man, the realisation that CD was completely fucked quickly sank in and has left me rather bitter...

I won't deny the tech they demoed looked really good and I hope it works out as they hope, but I don't have any faith it will. I've seen nothing to suggest that tech will work on a much larger scale with lots of player influenced events happening and still be able to work as intended.
 
They take statistics on how your friends drive and simulate it in the bots by attacking the turns in the similar ways. The consequence was they were all poorish drivers which smashed into other cars in the corners as most average players do. Thus the 'drivertard'. They have since adjusted it to do that less but that was the 'cloud' feature.

Have you even played against them on unbeatable?

Why are you trying to convince someone with heavy experience driving against the AI that they are different from what I've observed myself? If you set the difficulty lower you're going to get worse drivers and inherently nastier drivers. However the higher the difficulty you will get better drivers they are less likely to hit each other.

Drivatars are the best AI I've ever seen in a racing game and driven against.

They weren't anything revolutionary, Q3 had some very good pack in bots. UT as well.

Who cares if they weren't revolutionary, they were still an example of the servers at work...
 
I assume bandwidth can be a bottle neck? Are they using Azure compute or something?
Bandwidth shouldn't be the main problem (physics or game object data isn't exactly large), unless you plan to calculate some massive lightmaps or other with it (but then again, 3x XBO wouldn't be nearly enough nor would you want load times in minutes).
 
Have you even played against them on unbeatable?

Why are you trying to convince someone with heavy experience driving against the AI that they are different from what I've observed myself. If you set the difficulty lower you're going to get worse drivers and inherently nastier drivers. However the higher the difficulty you will get better drivers they are less likely to hit each other.

Drivatars are the best AI I've ever seen in a racing game and driven against.

That is how they work now; at launch they tended to be closer to how your friends actually driver. They patched that in because they didn't forsee how bad many people play.

Who cares if they weren't revolutionary, they were still an example of the servers at work...

The point is the 'cloud' portion of those features are over sold. They aren't super 'special' features.

The drivertars were mainly done locally, the cloud was shared storage; while the bots in TF were for multi-player synchronization. Lots of games already did that.

They were unremarkable. It was being sold as 'special' when many non XB1 games do those anyways.
 
That is how they work now; at launch they tended to be closer to how your friends actually driver. They patched that in because they didn't forsee how bad many people play.

They had restrictions from the very beginning on behavior learning and they've made improvements on this over time. The fact that it turned out how it did just shows how well the technology worked.

Drivatars are fantastic.

That is how they work now; at launch they tended to be closer to how your friends actually driver. They patched that in because they didn't forsee how bad many people play.

The point is the 'cloud' portion of those features are over sold. They aren't super 'special' features.

The drivertars were mainly done locally, the cloud was shared storage; while the bots in TF were for multi-player synchronization. Lots of games already did that.

They were unremarkable. It was being sold as 'special' when many non XB1 games do those anyways.

How many other games on console this generation are currently running server side bots?

Drivatars were not mainly done locally, they learned habits locally and sent that data to the server. The server would then have to review this data and build a driving profile for how you handle each type of car on each track. They would then have to match your profile with likely competitors and send that data to them on command.
 
They had restrictions from the very beginning on behavior learning and they've made improvements on this over time. The fact that it turned out how it did just shows how well the technology worked.

Drivatars are fantastic.
It's a good use of the cloud, but that said it's not something that other developers couldn't do by themselves using Amazon or whatever (whilst being able to configure it themselves and have cross-platform functionality) and it's not real-time. It's simply a calculation of an AI racing line with some driver attributes added to the AI. It's definitely the sort of thing cloud is best at; managing massive amounts of data and doing some predetermined calculations that isn't accessed in real-time simulation.

Because they obviously work the same way.

Just stop, dude. Your arguments have very little merit.
Not different from Quake 3 bots, just that Microsoft is sponsoring them for the backend.
 
It's a bunch of BS. Maybe it'll eventually see a real use with some complex server side AI in the future but I distinctly remember them touting it as a real GPU aid. They demoed some particle effects that "couldn't be done" with high end PCs. I dug up this article that shows the demo, and another demo from Nvidia from 2010 that simply blows their demo out of the water.

http://www.cinemablend.com/games/Microsoft-Demonstrates-Power-Cloud-Xbox-One-63207.html
 
In the early day MS emphasized the cloud as something that could improve the gaming experience. It seems like this has yet to manifest itself.

Has MS given up on this approach or is it being used but not being talked about?

Ever since Phil Spencer took over, they don't hype it up anymore. It is absolutely being used though. For things like matchmaking and dedicated multiplayer servers. Pretty awesome stuff really, much better than EA and activision servers.
 
It's a good use of the cloud, but that said it's not something that other developers couldn't do by themselves using Amazon or whatever (whilst being able to configure it themselves and have cross-platform functionality) and it's not real-time. It's simply a calculation of an AI racing line with some driver attributes added to the AI. It's definitely the sort of thing cloud is best at; managing massive amounts of data and doing some predetermined calculations that isn't accessed in real-time simulation.

You couldn't be any more wrong about Drivatars.

Who cares if it isn't real time? It is still calculating your profile in a server without your intervention.

If it isn't anything special where are all the other games doing this?

Drivatars are not anywhere as simple as just driving lines and driver attributes. It specifically learns where you break, the maneuvers you perform to pass, how you enter and exit corners, how you react to other drivers when they are near to you and more.

I've had people tell me that my driver will perform the same passes that I do against AI and beat them on the last corner of a race.
 
The drivertars were mainly done locally, the cloud was shared storage; while the bots in TF were for multi-player synchronization. Lots of games already did that.
Got a source for this?

I thought that player driver statistics were captured locally, uploaded to cloud servers and then processed en masse to derive Drivatar profiles which could then be downloaded to the game client. This mass processing was also supposed to lead to Drivatars learning new behaviors.

Drivatars are fantastic.
I'm a firm believer in the concept, and man do racing games need improvements in AI, but Drivatars still need a lot of work. Selecting the very highest difficulty Drivatars does minimize the demolition derby stuff, but also makes it damn near impossible to get a podium finish unless you're pretty damn skilled at the game already. If you're not? Your car will look like a wreck by the end.
 
It's a good use of the cloud, but that said it's not something that other developers couldn't do by themselves using Amazon or whatever (whilst being able to configure it themselves and have cross-platform functionality) and it's not real-time. It's simply a calculation of an AI racing line with some driver attributes added to the AI. It's definitely the sort of thing cloud is best at; managing massive amounts of data and doing some predetermined calculations that isn't accessed in real-time simulation.

So was the conversation whether they were cloud computing, or whether another company could maybe have built that system and paid another company to handle infrastructure to use in a non-specific game?
 
It's not even worth continuing down this line of conversation.

It is. Please tell me how the Titanfall bots are revolutionary. Because I don't get it. The only thing that is different is that Azure will scale automatically so the computing is more flexible. Which is great for MS because they don't need as much raw hardware but makes no difference on the user side.
 
Oh, please tell me more. That is not a valid argument.

Please go back to my post, I did.

I'm a firm belkiever in the concept but they still need a lot of work. Selecting the very highest difficulty Drivatars does minimize the demolition derby stuff, but also makes it damn near impossible to get a podium finish unless you're pretty damn skilled at the game already. If you're not? Your car will look like a wreck by the end.

You only need to achieve top 3 to podium gold in Forza 5. Pro, expert and unbeatable will all lessen the amount of wrecking that goes on.

Crackdown is not gonna be cloud-based, yall...

Based on what?

It's funny how Microsoft has a cloud calculation company developing the game then.

http://www.cloudgine.com/
 
Got a source for this?

I thought that player driver statistics were captured locally, uploaded to cloud servers and then processed en masse to derive Drivatar profiles which could then be downloaded to the game client. This mass processing was also supposed to lead to Drivatars learning new behaviors.

One of the devs was talking about the feature. Profiles are compiled locally, shared to the cloud then downloaded to other peoples machines and the local AI then uses it as a 'personality' for it's driving.

They also describe it here:
http://research.microsoft.com/en-us/projects/drivatar/forza.aspx

The interesting part is the local part. compiling of tendencies and then simulating it. The uninteresting part is the 'cloud' which is for sharing the profile.
 
You're saying to wait until they show the game, then when we see the game you're going to say that we should wait to play it ourselves. I'm saying that it's clearly working in some form and they've shown us the backbone of the game running which is what matters for this discussion.

Good thing you know how I will post in the future, I will give you my login and password and you can just take care of all my posting henceforth. Working "in some form" is not actually showing the product at hand....it is a proof of concept, hence why I say wait until we see them show the actual game. I couldn't give two shits that we haven't played it at that point.
 
You couldn't be any more wrong about Drivatars.

Who cares if it isn't real time? It is still calculating your profile in a server without your intervention.

If it isn't anything special where are all the other games doing this?

That's a bizarre argument. If it was that awesome, you'd think many devs would use it.
Drivatars were in the original Forza btw.
 
One of the devs was talking about the feature. Profiles are compiled locally, shared to the cloud then downloaded to other peoples machines and the local AI then uses it as a 'personality' for it's driving.

They also describe it here:
http://research.microsoft.com/en-us/projects/drivatar/forza.aspx

The interesting part is the local part. compiling of tendencies and then simulating it. The uninteresting part is the 'cloud' which is for sharing the profile.

That article isn't even about the Drivatar in Forza Motorsport 5 or how it works in Forza Motorsport 5.

For another, you can employ Drivatars to race for you in any Career Races that you are having trouble with or do not want to bother with (did I hear someone say "endurance"?). You can even pursue an entire career via your Drivatar.

You can't let Drivatars race for you in Forza 5, the article also makes no mention of cloud or server.

That's a bizarre argument. If it was that awesome, you'd think many devs would use it.
Drivatars were in the original Forza btw.

I'm not the one claiming it isn't "special" or unique, I'm saying that if it wasn't unique or special where are the other examples of it?

Drivatars in the original Forza are not anywhere near what we have now.
 
So tell us, how is it different?

The reason online co-op in most engines is tricky to do when there's the chance of divergent AI behaviors is that you need a way to keep all game-side behaviors in sync. Latency/bandwidth quickly falls apart if you treat every entity in a game as its own element that's constantly giving new information as though it were (in effect) another player, and there's problems reconciling those simulations if they're returning different results. That's why lockstep is used in the first place; it makes one player's instance the "home" machine and all other players telephone into that. But lock-step becomes harder and harder to manage when you have more and more (a) AI units to process on any one instance and (b) more players to share information with. It creates a bottleneck after a certain point.

That's presumably why the number of grunt units in Titanfall benefited from being able to handle all the AI "outside" of a player instance. Their role isn't to challenge the players outright, and the way they're implemented is a solution to a challenge in game design.
 
So was the conversation whether they were cloud computing, or whether another company could maybe have built that system and paid another company to handle infrastructure to use in a non-specific game?
I thought it was about the features cloud computing and specifically what the XBO component could do for games. I do think that Drivatars is a good feature and a good use of the XBO Cloud, but at the same time it's not real-time and has very little to do with the "3 times the computing power of Xbox One" stuff. I mentioned the other developers because this thread was what the Xbox One in specific could do, and Drivatars isn't a great example of that (unlike the Crackdown for example) since the bulk of feature isn't computationally very complex. Innovative and great feature yes, but I think it shows more about the strengths of T10 rather than the XBO Cloud.

Please go back to my post, I did.
I explained it simplified, but I don't see how it conflicts with your previous post (before the edit)
You couldn't be any more wrong about Drivatars.

Who cares if it isn't real time? It is still calculating your profile in a server without your intervention.

If it isn't anything special where are all the other games doing this?

Drivatars are not anywhere as simple as just driving lines and driver attributes. It specifically learns where you break, the maneuvers you perform to pass, how you enter and exit corners, how you react to other drivers when they are near to you and more.

I've had people tell me that my driver will perform the same passes that I do against AI and beat them on the last corner of a race.
The problem with it not being real-time is that the original power of the cloud was implied to be able to provide more interactivity, better graphics and all that sort of stuff and Drivatar is none of that. I won't contest it's value, but it's relevance to the "Power of the Cloud" talk. Nothing you say contradicts the "It's simply a calculation of an AI racing line with some driver attributes added to the AI." (I probably shouldn't have the "simply" there though, since I didn't mean that their algorithm is simple in any way) but all the overtaking manuevers are just attributes given to the AI and it's base driving line. It's not an simple system, but like I said it has little to do with the XBO and more with the T10's prowess.
 
It was a load of absolute bollocks. Truly this generation's blast processing (almost; it does at least exist and is used in games.)
 
So tell us, how is it different?

It is. Please tell me how the Titanfall bots are revolutionary. Because I don't get it. The only thing that is different is that Azure will scale automatically so the computing is more flexible. Which is great for MS because they don't need as much raw hardware but makes no difference on the user side.

It's obvious to anyone who has played games with standard server-side bots like Quake 3, and also spent a decent amount of time with Titanfall, to realize what the difference is. Yes, at their core they are still server-side bots. However, the difference in artificial intelligence of the bots is where it matters.

To say they're just server-side bots are doing them a disservice. There have been many articles written about how Titanfall is using Microsoft's Cloud to provide a more immersive experience.
 
Only game data, but if you want it to be a signifiant, noticeable improvement for the game graphics, you'll have to use a lot of bandwidth anyway. Or else we're speaking about the kind of things the console would do well enough on it's own.

This isn't true. You could play against the smartest AI a game has ever seen, that requires the processing power of a machine far more capable than the Xbox One, and all you need to transmit would be the choices it makes. It would take no more bandwidth than you playing against me, and would even be less latency sensitive as it would essentially be pinging 0 to the servers.

This has only direct advantages for the provider and renter though.
And for sure this isn't what they hyped up/what people fell for.

This has the knock-on effect of benefitting end-users. Much like how the cost of hardware affects the specs of our consoles, inefficient and costly server implementations affect what we can have in our online games. Take FH2 for example, you can create a private match, just for you and one other friend, and it'll host you a server immediately, with the world populated with traffic that is synced for both of you. Can you provide a prior example of something like that? If not, why do you think that is?

The dynamic scaling may also help with the longevity of games, so rather than flagship titles getting their server cut after two years Sony style, the only persistent cost would be the matchmaking server, with all others ceasing to exist unless people are actively playing on them. Only got 2 people in the world playing on one server? Then only one server has to exist for now.

Just look at Kinect, that turned out to be everything they said it would be.

Disregarding whether you like the Kinect or not. How many things in MS' original proposal for the Kinect, can it not actually do?
 
I thought it was about the features cloud computing and specifically what the XBO component could do for games. I do think that Drivatars is a good feature and a good use of the XBO Cloud, but at the same time it's not real-time and has very little to do with the "3 times the computing power of Xbox One" stuff. I mentioned the other developers because this thread was what the Xbox One in specific could do, and Drivatars isn't a great example of that (unlike the Crackdown for example) since the bulk of feature isn't computationally very complex. Innovative and great feature yes, but I think it shows more about the strengths of T10 rather than the XBO Cloud.


I explained it simplified, but I don't see how it conflicts with your previous post (before the edit)

The problem with it not being real-time is that the original power of the cloud was implied to be able to provide more interactivity, better graphics and all that sort of stuff and Drivatar is none of that. I won't contest it's value, but it's relevance to the "Power of the Cloud" talk. Nothing you say contradicts the "It's simply a calculation of an AI racing line with some driver attributes added to the AI." (I probably shouldn't have the "simply" there though, since I didn't mean that their algorithm is simple in any way) but all the overtaking manuevers are just attributes given to the AI and it's base driving line. It's not an simple system, but like I said it has little to do with the XBO and more with the T10's prowess.

Not being real-time is a incredible silly argument for trying to disprove the effect of server-side calculations.

The way you're describing what Drivatars do is extremely simple and portrays the technology in a way that makes it look trivial and pointless.

It does a hell of a lot more than just record your racing line and attitude.
 
I haven't been as openly critical because I honestly did start to believe that Kampf was legit (claims of verified helped with this) and wanted to see if his claims had any basis before ripping into it completely.I did want to believe, but after he was exposed as a hype man, the realisation that CD was completely fucked quickly sank in and has left me rather bitter...

Whoh, I missed this somehow. Can you provide a link to the thread/post? Thanks!
 
The reason online co-op in most engines is tricky to do when there's the chance of divergent AI behaviors is that you need a way to keep all game-side behaviors in sync. Latency/bandwidth quickly falls apart if you treat every entity in a game as its own element that's constantly giving new information as though it were (in effect) another player, and there's problems reconciling those simulations if they're returning different results. That's why lockstep is used in the first place; it makes one player's instance the "home" machine and all other players telephone into that. But lock-step becomes harder and harder to manage when you have more and more (a) AI units to process on any one instance and (b) more players to share information with. It creates a bottleneck after a certain point.

That's presumably why the number of grunt units in Titanfall benefited from being able to handle all the AI "outside" of a player instance. Their role isn't to challenge the players outright, and the way they're implemented is a solution to a challenge in game design.

Yes that's true. In multiplayer enviroments it's better to make the trade offs in AI latency and have a server do it rather than a local on all machines; for purposes of synchronization. So in SC/SC2 games on a multiplayer comp stomp one of the players is 'host' and the AI comes from them. UT and Q3 had server bot packs for the same reason or the host would simulate them. Depending on how you set up your game; one of the other would occur. Even for really simple AI games like Dota or LoL the AI script is run ont he server for synchronization purposes.
 
Since the very beginning there has been quite a lot of people explaining why the whole real time cloud computing was BS. Let's not be surprised now.
 
It's obvious to anyone who has played games with standard server-side bots like Quake 3, and also spent a decent amount of time with Titanfall, to realize what the difference is. Yes, at their core they are still server-side bots. However, the difference in artificial intelligence of the bots is where it matters.

To say they're just server-side bots are doing them a disservice.

I put many, MANY hours into Quake and UT across multiple builds. I also loved different bot types you could get. I use to play with aggressive bots, defensive ones, etc.. I had a blast against their bots, but they're absolutely NOTHING like the AI in Titanfall.

In the hoard mode they added it can be crazy fighting two Titans on your own, let alone 8 Titans and 30 different kinds of ground troops.

Watching the AI Titans drop smoke, put up shields to catch your projectiles, sending your projectiles back at you while simultaneously hitting you with a missile barrage and then hitting you with heavy machine gun fire after dropping your shields....that's nothing like Q3 or UT...lol.

I'd say the complexity of both defending with countermeasures, attachments, and movement along with attacking is at least a slight step up.
 
It's obvious to anyone who has played games with standard server-side bots like Quake 3, and also spent a decent amount of time with Titanfall, to realize what the difference is.
I've played against some incredible Quake 3 bots. There were a lot of varieties available for download in that game's heyday. Can't even imagine what might be available today.

And I have no idea whats supposed to make Titanfall bots impressive, but I always thought they were there as simple cannon fodder by design, and not as a showcase of complex AI. The player's titan when in AI mode is more interesting imho.
 
I've played against some incredible Quake 3 bots. There were a lot of varieties available for download at the time.

I have no idea what makes Titanfall bots impressive, but I always thought they were there as simple cannon fodder by design, and not as a showcase of complex AI. The player's titan when in AI mode is more interesting imho.

From my experience, outside of "important" units like AI pilots and Titans, it was the sheer number of grunts that was their merit. They aren't built to outsmart the player, but they sure do swarm.
 
I am really interested if and if yes, to which extent they are using cloud computing in Crackdown.

I think the main problem is, that things also have to work if someone wants to play offline, or even has a shitty connection.

But I could see it happening in Crackdown if you destroy a building you either get a scripted basic sequence or you see many, many bricks falling down physically correct.

Could be cool :D
 
It's obvious to anyone who has played games with standard server-side bots like Quake 3, and also spent a decent amount of time with Titanfall, to realize what the difference is. Yes, at their core they are still server-side bots. However, the difference in artificial intelligence of the bots is where it matters.

To say they're just server-side bots are doing them a disservice. There have been many articles written about how Titanfall is using Microsoft's Cloud to provide a more immersive experience.

They were cannon fodder.

Ergo bots.
 
I am really interested if and if yes, to which extent they are using cloud computing in Crackdown.

I think the main problem is, that things also have to work if someone wants to play offline, or even has a shitty connection.

But I could see it happening in Crackdown if you destroy a building you either get a scripted basic sequence or you see many, many bricks falling down physically correct.

Could be cool :D

Could powered suggests online only. Doubt there will be any offline option...

I'd love to be proven wrong though...I'd love for them to ditch destruction altogether and give us a traditional CD in an all new city with a focus on expanding the lieutenants and kingpins roles to where they pose a real threat and where battles can end up being city wide and where reinforcements from other lieutenants come to back others and where gangs regroup and fight back even after you've defeated them, but I guess it's all about pushing cloud powered destruction...
 
Top Bottom