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

[patsu]

How would an online Xbox One running BF4 stack up vs. the gaming PC that ran BF4 awhile back during DICE's SP reveal? If say 300,000 cloud servers were up and running, then what things could we see pop up in favor of the Xbox during a technical comparison? Or would only 1st party titles offload to the cloud? I'm confused on how useful cloud computing is for games like Crysis, God of War, Battlefield, etc.

In my view...

Probably a more “connected” experience because the server can provide you with real-time data and stats about other players in the same world. Perhaps changing the world balance and AI behavior as you play.

The FLOPs are far away, so you can’t really use them for say... rendering for a fast paced game. Something async is best.

People are saying it’s bullshit because for instance, your game AI can also change their behavior based on local, immediate data. So the game designer probably need to find a good way to use the servers. I think it works best with BigData.

e.g., You can pull map data from OpenStreetMaps or GoogleMap and claim cloud power.


Personally, as a user, I just see them as MMOs because structurally, the game client and servers already split their responsibility up nicely.

EDIT: Demon’s Souls is another example. In the SP game, you can see the white ghosts of other players on-the-fly, and also change world balance collectively via the servers. I read that PSN provides an API for you to get the real-time info of other players in the game.

 

[Scott_Arm]

In my view...

Probably a more “connected” experience because the server can provide you with real-time data and stats about other players in the same world. Perhaps changing the world balance and AI behavior as you play.

The FLOPs are far away, so you can’t really use them for say... rendering. Something async is best.

People are saying it’s bullshit because for instance, your game AI can also change their behavior based on local, immediate data. So the game designer probably need to find a good way to use the servers. I think it works best with BigData.

e.g., You can pull map data from OpenStreetMaps or GoogleMap and claim cloud power.

In the architecture panel discussion they specifically said it was more suitable for jobs that are less latency sensitive. Unfortunately they did not give any real examples, but they obviously were not implying graphics specifically. Maybe there is some "pre-computed" graphics data they could pull from the crowd, but that's more of a disc space issue than a computation issue.

 

[patsu]

I gave a few examples above. The game servers of today already perform in-game computation of sorts. If they want, they can also generate a new map or model on-the-fly (e.g., to persistent the destruction of a level after a battle in MMO).

It’s a case of Microsoft servers providing common in-game services to the developers. The games themselves of course can roll their own.

 

[ERP]

People are saying it’s bullshit because for instance, your game AI can also change their behavior based on local, immediate data. So the game designer probably need to find a good way to use the servers. I think it works best with BigData.

e.g., You can pull map data from OpenStreetMaps or GoogleMap and claim cloud power.

FWIW the costly parts of AI are enormously latency insensitive. Certainly a 100ms here or there is irrelevant to most implementations.
People seem to be forgetting that every time they see an NPC in a game when playing online there is a good chance his AI is running remotely.

Now I still question the value vs added complexity of doing it remotely unless there is some game design reason to simulate across a broad area.

I think the real value is in persist world state, and even potentially allowing interaction with that state from other devices. But the game would have to be designed to exploit it.

A couple of years ago I wanted to do a largely single player game where the world was live all the time, on the primary device you could micromanage the world immediately around you, but you could check state and give broad strategic instructions from a web client or a mobile app whether or not the game was running.

 

[astrograd]

Someone at TXB found this. It's an interesting read on the topic from MS's own research teams.

http://research.microsoft.com/pubs/72894/NOSSDAV2007.pdf

 

[patsu]

Yes, that’s why I keep mentioning MMOs. The new machines should be powerful enough to handle large scale AI. It’s more robust to keep things locally. Someone has to pay for the online infrastructure too.

If they want to push the AI update to the server, perhaps aggregating other players’ data makes it more worthwhile.

The MMOs already do these things. It’s a matter of using MS’s infrastructure or roll your own.

The service may be more valuable for phone and tablet games where the power is more limited. Games like DUST514 that interconnect phones, tablets, home consoles and PCs probably do something like this.

 

[astrograd]

Arstechnica article on cloud stuff directly from MS:

http://arstechnica.com/gaming/2013/...s-more-processing-power-from-cloud-computing/

Our first question had to do with the 30,000-server cloud architecture that Microsoft says the Xbox One will use to help support "latency-insensitive computation" in its games. What does that mean exactly, and can laggy cloud data really help in a video game where most things have to be able to respond locally and immediately?

"Things that I would call latency-sensitive would be reactions to animations in a shooter, reactions to hits and shots in a racing game, reactions to collisions," Booty told Ars. "Those things you need to have happen immediately and on frame and in sync with your controller. There are some things in a video game world, though, that don't necessarily need to be updated every frame or don't change that much in reaction to what's going on."

"One example of that might be lighting," he continued. "Let’s say you’re looking at a forest scene and you need to calculate the light coming through the trees, or you’re going through a battlefield and have very dense volumetric fog that’s hugging the terrain. Those things often involve some complicated up-front calculations when you enter that world, but they don’t necessarily have to be updated every frame. Those are perfect candidates for the console to offload that to the cloud—the cloud can do the heavy lifting, because you’ve got the ability to throw multiple devices at the problem in the cloud."

Booty added that things like physics modeling, fluid dynamics, and cloth motion were all prime examples of effects that require a lot of up-front computation that could be handled in the cloud without adding any lag to the actual gameplay. And the server resources Microsoft is putting toward these calculations will be much greater than a local Xbox One could handle on its own. "A rule of thumb we like to use is that [for] every Xbox One available in your living room we’ll have three of those devices in the cloud available," he said.

 

[Scott_Arm]

Grandiose claims. They're setting expectations very high. I'll take the route where I keep my expectations very low, though I do expect to see some use of the idea in first party games. Just not sure how significant it'll be.

Edit: The section on cloud computing is actually really interesting. There are some concrete examples in there, and it sounds like stuff they've probably toyed with. The idea that there will probably be a local placeholder for some effects when the cloud is not there is interesting. This does seem like a tough sell for multiplatform, unless this is also available for PC games. I would expect first parties will try this stuff.

 

[patsu]

As long as the data is used in an asynchronous fashion, you should be able to offload it elsewhere. It's up to the designers how to tap on the additional computational power. e.g., they can try to stream car simulation results from the server. The client just read the latest value on-demand.

BigData type results would be interesting too, but it would be more than just computational power. The aggregated data makes things more interesting.

 

[Brad Grenz]

Arstechnica article on cloud stuff directly from MS:

http://arstechnica.com/gaming/2013/...s-more-processing-power-from-cloud-computing/

Those are exactly the kinds of claims that come across as complete horseshit. You're going to calculate lighting on the cloud? Not in real-time, I trust. And if you're talking about generating "just-in-time" lightmaps, or somthing like that, I can't imagine it will be easy to deliver that much data fast enough to be practical. I would have been willing to believe you're sending AI updates every 10 frames and the game just interpolates their movement in between, but this guy is trying to say cloud computing will hipcheck PhysX out of the way with remote fluid simulations? I can't believe anyone is buying that garbage.

 

[patsu]

Yeah, not lighting within the frame time directly. He's referring to just some computations that can be pre-calculated before entering an area. I suppose you can also precompute it during compile/bake time ?

 

[Strange]

Only 40x? That seems really low.

Using the cloud they can get 4x the processing power of an Xbox One.

They're going to allocate 1.2TFlops *3 = 3.6 TFlops to every Xbox One online!

 

[Bagel seed]

Yeah, not lighting within the frame time directly. He's referring to just some computations that can be pre-calculated before entering an area. I suppose you can also precompute it during compile/bake time ?

Right, but if it were a dynamic player created world you'd need to generate new lightmaps after the changes.

A photorealistic SimCity with complete simulation of millions of citizens (not a fudged simulated amount) and realistic traffic AI is something I'd like to see come out of cloud based game. But I know it won't. This over any other singleplayer turned MMO+ based bs.

 

[Rangers]

Those are exactly the kinds of claims that come across as complete horseshit. You're going to calculate lighting on the cloud? Not in real-time, I trust. And if you're talking about generating "just-in-time" lightmaps, or somthing like that, I can't imagine it will be easy to deliver that much data fast enough to be practical. I would have been willing to believe you're sending AI updates every 10 frames and the game just interpolates their movement in between, but this guy is trying to say cloud computing will hipcheck PhysX out of the way with remote fluid simulations? I can't believe anyone is buying that garbage.

someone's having a meltdown...

i had no idea the cloud stuff was going to be this big or that it could directly affect graphics. this is getting pretty intriguing to say the least. i figured it was just mumbo jumbo bullshit.

I cant really overstate how much that ars technica article changes things, especially because it is Microsoft itself talking about their plans. If it comes to fruition. Lets see some concrete results MS! Is E3 too early?

They could be really ahead of the curve here and caught the competition flat footed, this could be really amazing stuff.

Maybe their plan was a lightweight box augmented by the cloud all along, you know? We know the leaked doc had some hints that direction which was 2009 or 10.

 

[bagofsuck]

I will give them the benefit of the doubt with this Cloud Compute..lets hope we can something at E3.

 

[Scott_Arm]

I imagine there is a realistic middle ground between "This is total BS!!!!!!!! The bees! The bees!" and "Three X boxes in the cloud! Oh yah! Ohhhh yeeeaaah .... uuuuuggghhhhh"

 

[Rangers]

What I hate is that example he gave of SSAO, being less for a couple seconds when you enter a room then loading in the good stuff from the cloud...

Whatever the solution is lets not do that. I think I'd rather wait out a 2 second load than that.

 

[astrograd]

Those are exactly the kinds of claims that come across as complete horseshit. You're going to calculate lighting on the cloud? Not in real-time, I trust. And if you're talking about generating "just-in-time" lightmaps, or somthing like that, I can't imagine it will be easy to deliver that much data fast enough to be practical. I would have been willing to believe you're sending AI updates every 10 frames and the game just interpolates their movement in between, but this guy is trying to say cloud computing will hipcheck PhysX out of the way with remote fluid simulations? I can't believe anyone is buying that garbage.

I don't see how this stuff is all that difficult to grasp for you ppl. World simulation, physics and animations, global AI, and certain lighting aren't sensitive to latency. They don't have to be updated or calculated every frame. We aren't talking about things that the player is interacting with.

Lemme try explaining it this way:

Imagine playing Skyrim. You've got a bounding volume surrounding you that extends some significant distance on all sides. Say it extends just to the point a bit beyond how far your spells can be cast or your arrows will fly when fired. Within that box, you can interact and influence the world around you directly, on the fly. The physics, AI, lighting, animations, etc within that box have to be assumed to totally dynamic and fully real time to ensure good gameplay experiences.

But anything beyond that box doesn't. In order to influence things outside the box surrounding you, you'd need to take the time to walk your character across a boundary of some transition region between your box and the external world beyond it. The time it takes to move your character around in this box and cross its boundaries allows the cloud stuff outside of it to gently transition to a state where it is locally rendered and fully dynamic and interactive on the fly.

Anything that is outside of the box can be baked. Not because it doesn't play animations or physics or simulations or AI out continuously, but rather because it can pre-bake that stuff, or pre-render lighting if you like, and just stream in the data for the game to render and display outside the box. Instead of using the GPU like a GPGPU and leveraging it for computing these things, you can let the cloud do it far more accurately and quickly for the stuff outside the bounding box, freeing up your GPU for only localized, latency sensitive tasks.

 

[Rangers]

I imagine there is a realistic middle ground between "This is total BS!!!!!!!! The bees! The bees!" and "Three X boxes in the cloud! Oh yah! Ohhhh yeeeaaah .... uuuuuggghhhhh"

yeah, it's some stuff that's hard to wrap your mind around. i would like to see some concrete examples soon.

I mean, it would sure be interesting if they actually tell devs "ok, program your game with exactly 4.26 teraflops of cloud processing in mind because that's how much we're allocating to every Xbone right now, go". I could see that being a reasonable start point.

 

[3dilettante]

Right, but if it were a dynamic player created world you'd need to generate new lightmaps after the changes.

A photorealistic SimCity with complete simulation of millions of citizens (not a fudged simulated amount) and realistic traffic AI is something I'd like to see come out of cloud based game. But I know it won't. This over any other singleplayer turned MMO+ based bs.

The cloud servers may be individually specced to be much more powerful than a console, but they are going to be vastly outnumbered.
A running simulation is a constant demand task that really cuts into their ability to take advantage of idle periods to sneak time slices to other instances.

Unless SimCity6 has everyone playing The One City, the idea of having hundreds of thousands of instances that require more resources than Durango is going to be a far-off idea.
If you don't want to spend enough money to outfit a Haswell EX server with hundreds of gigs of RAM for your one city, they sure as heck don't for everyone else.


That, and besides the photorealistic part, it's not like it's not possible to simulate millions of agents. It's just that SimCity can't.

If there's one thing that was a red flag to me, it was talking about cloud services with EA in the neighborhood.

SimCity is a warning to developers. Even if a cloud is good, and Amazon's EC2 is what SimCity uses, it doesn't matter if the application running in the cloud sucks.

I imagine there is a realistic middle ground between "This is total BS!!!!!!!! The bees! The bees!" and "Three X boxes in the cloud! Oh yah! Ohhhh yeeeaaah .... uuuuuggghhhhh"

Nope. It's a design feature of Azure Live. You either get enough computation to find the answer to life and everything, or it has negative ZetaFLOPS and you wake up Amish.