Server based game augmentations. The transition to cloud. Really possible?

Maybe a planetside like mmo where npc fight with you and against others ai while two army fights nearby
but...
 
Read the original interview and you get a clearer picture. I don't think Ryse is using the cloud at all.

I'll just go ahead and quote him:
I joke, dude, if I can do 10,000 AI, and I use the cloud to compute the processing power, then you bet your ass you're going to have a battle with 10,000 AI in it, because that's what we want to do.

He even says, "I joke". Then the interviewer says is it is at all possible, he says "technically it is", instead of the more realistic "I was just making up numbers to prove my stance on how we develop our games."

So the link title is quite misleading indeed.
 
If you have 10k+ AI...you then have to figure out how to render them on screen...so it sounds a bit hypothetical to me...

It doesn't make much sense for a game like Ryse, but would be cool for a games like GTA, Skyrim, Fallout, etc, where you aren't visually rendering all 10k+ plus ai's on screen at once but instead they all live, think and hence populate the world in a more realistic manner, and make the game feel more real and unpredictable when you arrive at different locations and see what's going on. So all 10k ai's live and do their thing but you only visually see the ones in your immediate vicinity. This type of thing was totally possible via cloud over high latency.
 
It doesn't make much sense for a game like Ryse, but would be cool for a games like GTA, Skyrim, Fallout, etc, where you aren't visually rendering all 10k+ plus ai's on screen at once but instead they all live, think and hence populate the world in a more realistic manner, and make the game feel more real and unpredictable when you arrive at different locations and see what's going on. So all 10k ai's live and do their thing but you only visually see the ones in your immediate vicinity. This type of thing was totally possible via cloud over high latency.

That is a good point. How many AIs are already in Skyrim or GTAV? Any idea?
 
That is a good point. How many AIs are already in Skyrim or GTAV? Any idea?

Hmm not sure. I haven't played GTAV but in Skyrim it doesn't seem like it would be in the 10k range, but then again towns in Skyrim for example do feel kinda empty. I don't think that's necessarily due to render limits but more like with ai being as primitive as it currently is, I think the game would look really goofy if you saw 1000 people in a town doing mostly the same things. Basically there isn't enough ai variety right now to support high density of people without it looking weird. Hitman Absolution packed tons of people into one small space on one of their missions (looked like ~800 people) but they were all just standing around talking or watching fireworks so that worked ok. But more often than not it wouldn't look too good. Look at Assasins Creed for example, they pack more humans into their cities compared to other games and as such human ai movements look oh so predictable and repetitive, it's kinda weak and breaks the illusion of it being a living city.

It would be cool if future games farmed non destructive human player movement via cloud in games like Skyrim or GTAV, and added them as ai's in our game working alongside traditional cpu ai. That would really liven up the cities and make them seem more human. If you didn't have an internet connection then you'd see people, cars, etc all moving along in their predictable ways as they have been for a decade or so now, but if you had internet access then the game could throw human behavior into the mix which in turn would also serve to liven up the existing cpu based ai as it tried to work around the more unpredictable human cloud farmed ai. Could be really neat.
 
It is true, a Forza game with cloud AI could not exist without said cloud.

"So, would [cloud AI] have been possible last generation? Sure. And last generation we could have done 1080p, 60fps. But we wouldn't have been able to do that and the physics we did in Forza 4. So Forza 4 was 720p and 60fps, with the physics we had. Now we're doing things in the physics that requires so much more computational power, we would have struggled to be 720p, 60fps."

Asked to clarify that Turn 10's ready access to the cloud has allowed it to spend more time improving Forza 5's resolution and physics, Greenawalt responded: "Right. And that's what's allowed us to, in three years, have a game that has innovation like Drivatar, looks beautiful as well as having the new things we're doing in physics, like the tyres and suspension."

Sounds like PR BS. I'd love to know what physics they are doing on the cloud :)

Their physics engine was something like 360 calculations per second in the past, now apparently they can wait 30-200ms for those calculations right?
 
Well if they really wanted to go to town with cloud physics they could model the entire car's physics in the cloud....so let's say you tap another car which might not seem like a big deal but then they could model the small physical damage and how it affects the physics of the handling etc. Not only that..so lets say a piece of the carbon body panel becomes dislodged and start dangling around while driving, the compromised aerodynamics could be modeled in the cloud and even material fatigue so eventually the carbon body panel completely falls off which changes the aero yet again.

Not saying that's what they're doing, just saying there's tons of things you could simulate using the cloud. If this is their ultimate goal then I could certainly see how a ton of fine grained physics calculations for a bunch of different things in the game could take up a bunch of resources that the console itself cannot do and still run at 60fps at 1080p.
 
Last edited by a moderator:
They're basically saying that their cloud AI implementation could not be done locally without impacting the rest of the game, either by robbing resources or requiring more dev effort because Microsoft's cloud infrastructure takes care of much of the foundation work.

All things not Drivatar have benefited by having more local system resources and more development time allocated to them.
 
They're also saying that without offloading compute to the cloud they couldn't have hit 1080@60!? So it would have been another 720p game or maybe 900p?

Also what happens when the cloud goes down or your internet connection does? Does the game stop working or do you just race alone on empty tracks?
 
I wish people would just read more before making frankly dumb comments, but then again this is the 'net.

If you don't connect online, then there are default/existing AI profiles that'll race you.
If you connect online, you download other real player's driving profiles as your AI opponents, and you upload your driving style too. I don't know where this AI profiling happens, one can assuming the analysis is done in the cloud after the racing lines are uploaded.

Depends on the types of simulations, one can either try to model the system, or your model the envelope. It's been a while and I don't know where the state of things are right now, assuming that you are trying to model the envelope, then the user can upload the car's config and have the cloud run a physics simulator and download the model.
 
Last edited by a moderator:
They're basically saying that their cloud AI implementation could not be done locally without impacting the rest of the game, either by robbing resources or requiring more dev effort because Microsoft's cloud infrastructure takes care of much of the foundation work.
I would very much like to see a breakdown of how CPU time is broken down and where they are spending it.
 
That breakdown wouldn't include work that is done in the cloud, for cloud enabled features.

In their desperation to tear MS a new one, people are acting like "server side" processing doesn't require server side processing.

Thank god we're moving away from user hosted multiplayer sessions, with all the crippling limitations that user hosting has forced on games.
 
"Physics lag" is an unavoidable consequence of playing games over the internet. Whether its client/server, or user hosted, or peer to peer with some kind of arbitration .... this isn't a problem that has just appeared with the advent of "the cloud".

User hosting and peer to peer are the absolutely fucking worst setups for handling lag (maybe writing updates on a postcard and sending them through the physical post would be worse - but not much). And the line between "single player" and "multiplayer" is for some game types blurring quickly.

Thinking of "the cloud" as giving a resolution boost to consoles is the wrong way of looking at it. And MS were silly to try and present it as such. The benefit is not pissing local processing resources away in an attempt to do something in a laggy and unreliable fashion, where you'd need cloud storage for persistence anyway.
 
I wish people would just read more before making frankly dumb comments, but then again this is the 'net.

If you don't connect online, then there are default/existing AI profiles that'll race you.
If you connect online, you download other real player's driving profiles as your AI opponents, and you upload your driving style too. I don't know where this AI profiling happens, one can assuming the analysis is done in the cloud after the racing lines are uploaded.

Depends on the types of simulations, one can either try to model the system, or your model the envelope. It's been a while and I don't know where the state of things are right now, assuming that you are trying to model the envelope, then the user can upload the car's config and have the cloud run a physics simulator and download the model.

I didn't see any explanation in the article.

So the game crowdsources user game data like Demon's Souls. Then process the data on the server farm. Analysis or preprocessing of individual data can be done before uploading if they want to (User is out of the race anyway and MS can use the GPGPU like folding@home). But the servers may perform additional processing by comparing and aggregating other user data. It would be similar to GT6 users uploading GPS track data and ripping a new "user generated" track (collectively) based on the users' telemetry and GPS data.

It also supports dedicated server for gaming.

The server power doesn't always have to contribute to 1080p/60fps (or above) though. The concepts can be separated if designed properly because the processing and aggregation is "asynchronous". The results can also be refined with more and more user submissions along the way.
 
Back
Top