I loved the facemapping in that old Rainbow 6 game, so it'd be fun to get more of this kinda stuff in games:

"A Neural Algorithm of Artistic Style": http://arxiv.org/abs/1508.06576

Hahaha, "rendering"

LazyGunn wrote:

Nope, slowly introducing previously unseen ways in which to endeavour

... for the vastly superior robot overlords.

I am, mostly, just joshing.

make that vr and take my money

Nice video this one.



Not quite as good as the time Carmack previewed the Rift, but still very special...

Microsoft delivering on their Azure promise here. 

Like Tittyfall, a remote server maintains the empirical state of the multiplayer session, but unlike Tittyfall, the server is used to store and describe more interesting game critical details (in this case environmental physics and the state of the game map in realtime).

This is the type of shit I've been waiting for, though the potential of PS Now does nullify the brilliance of this in some way (though practically, will Sony every consider creating the kind of one-off cloud based game I imagine they could do? I fucking doubt it).

Wonder how they'll deal with net splits and realtime session state?
And just how dynamic will this server cluster be? 

I can imagine Microsoft writing off some of the costs to use this as an Azure showcase...

In fact, it would be interesting to do the maths on the projected cost of such a server... based on say, a sample of 20,000 concurrent users worldwide, playing (I guess) in groups of 32 players to each server cluster every night.

But to do that, we'd need to know the spec for each node/server, and take a rough guess at the work each would be performing...

One guess I'm willing to make, is that in the long run, it will cost the £30million quid to produce the game that goes on the shelf at Game, plus at least double that initial cost to run the server (based on that optimistic 20,000 users per night figure) for a year...  

I also have other questions, like what languages/tools are they using? How much data and what kind of data might they store? What if they decided to settle NPC AI on the game server?... 

...

Now...

If only Microsoft kept their colours nailed to the Kinect 2...

Skerret wrote:

Welcome to about a month ego eh eh Mbut srs that has extra bits and will be boning up for it.

Ok.

So you know all that backface culling, fogging, and viewpoint clipping type stuff that Chump knows about?
You'd probably have to do a similar thing here, but rather than just ensure that the computer optimizes what's drawn on screen, you'd also probably want optimize which server you make your UDP game state request too.

So if your character's facing north at any given time, give them the data for that particular section of the town. and don't bother with the stuff out of view to the east, south and west.

A production system would be much more complicated than that crude guess though.

Obviously, gameplay features like high vantage points and binoculars will help to either constrain or dilate the server workload.
They'd also want to make the cinematic 8 building takedowns an infrequent occurrence.

Also wonder if there's a CGI style LoD equivalent for physics?...

Yes. I would really enjoy a discussion about it. I fort dats' wat dis fred' dun du'.

For a start, for the demo at least, Sumo Digital are NOT using Azure...

This can be ported to PS4...

There's always time for humorous antagonism.  Anyway, I've had a shit week and why not take it out on dyno.

dynamiteReady wrote:

Skerret wrote:

Welcome to about a month ego eh eh Mbut srs that has extra bits and will be boning up for it.

Ok. So you know all that backface culling, fogging, and viewpoint clipping type stuff that Chump knows about? You'd probably have to do a similar thing here, but rather than just ensure that the computer optimizes what's drawn on screen, you'd also probably want optimize which server you make your UDP game state request too. So if your character's facing north at any given time, give them the data for that particular section of the town. and don't bother with the stuff out of view to the east, south and west. A production system would be much more complicated than that crude guess though. Obviously, gameplay features like high vantage points and binoculars will help to either constrain or dilate the server workload. They'd also want to make the cinematic 8 building takedowns an infrequent occurrence. Also wonder if there's a CGI style LoD equivalent for physics?...

Will there be multiple servers involved per game?  I'd imagine that the complication of multiple servers interacting / sharing state / routing packets to the right place would outweigh any savings?

Well... That's where it gets nuts, and that why I wanted to talk about it here, because you'd obviously want to have an empirical summary of the game world that would give you the exact position of each player, a simplified capture of what the player might see (the bounds of the view frustrum, as Skerret gave the proper name for it), and the general gameworld state that might say building '1', in section 'A' has been demolished, but this serve would have no idea how fucked building 'A1' is...

That said, as you hammer away at building 'A1', the data associated with your interaction with building 'A1' goes directly to the server that maintains state for building 'A1' in a separate request to the one you make to tell the main server where you are, and what you're looking at.

That said, because the main server knows your view frustrum is currently pointing at building 'A1', it will send back to you, along with the position of the other player, the flag the client needs to poll the server for section 'A'...

Maybe.

If I had a squillion quid, a clue, and some loyal spods with me, I would test this...

dynamiteReady wrote:

Yes. I would really enjoy a discussion about it. I fort dats' wat dis fred' dun du'.

Sorry dyno, I was meaning to reply earlier but was busy and didn't want to just crosspost what I'd written from the Crackdown 3 thread. 

dynamiteReady wrote:

For a start, for the demo at least, Sumo Digital are NOT using Azure... This can be ported to PS4...

That's interesting - I was under the impression that it was Azure through and through? As in, maybe they aren't literally running the demos on Azure cloud servers (as they'd want it to be running on the local network and known hardware setup for reliability when demoing to the world's press), but it's being done on something analogous that's running either a local Azure server hardware, or with code that's using the Azure API - as you wouldn't want to rewrite the whole thing when it comes to actually testing and launching on Azure.

As for porting to PS4, I don't think that's in any way likely as it's both Microsoft Studios' IP and Dave Jones/Cloudgine/Reagent Games are developing this on Microsoft's dime, so there's no chance whatsoever it'll come to Sony.

Aside from that, feel free to ask away and I'll put down my thoughts, for whatever they're worth as I don't know much about networking, server or cloud gubbins. 
My primary "concerns" with the whole idea currently are: 
1) How well will it fare on our shitty ADSL in this country?
2) Does the online-only tradeoff result in a significantly better gameplay experience than if you aimed for a similar level of destruction/physics/processing power but kept it local only? (e.g. Will Crackdown 3 in MP be noticeably/significantly better than something like Just Cause 3?)
3) How long will those servers run and what happens with load-balancing at peak times and if other Azure clients (e.g. corporate clients paying top dollar) want/need the resources? (and is that even something that's an issue? For all I know MS's Azure clusters have so much power to spare that we'll never touch the sides - but if its being run as a business, I wouldn't have thought they'd have significant excess sloshing around as that costs money to both setup and run).
4) Hopefully this'll get folded back in to the SDK and both MS and Cloudgine will make it easy to use this kind of tech - but with EA always running their own servers and and cross-platform games having to not rely on it for the non-MS versions of the game... how much is it going to get used? I mean, if MS Studios decides to do a big push on it and all 1st party studios are pushed to do online games then TBH that will probably work out quite well (much better than forcing them to do Kinect games, that's for sure).

I mean, I say "concerns" but I'm not really concerned, because it's Crackdown and I love those games so I'm sure it'll be great fun in both single player and that the clod tech will mean you can really let rip with magnitude greater levels of destruction in the MP

dynamiteReady wrote:

So they'll rely on Microsoft's deep pockets and mahoosive data centers for the final product. With that in mind, hardware and money will be no object for this, as Jones pretty much has the same sway Miyamoto would if he decided to pitch an idea at Microsoft.

I'll be interested to see how it shakes out because while I'm sure the Azure/cloud division will be providing services to xbox/entertainment division at cost, I wouldn't have thought they'd want any loss-leaders on their budgets, so (I'm presuming) they'll be getting paid by xbox division for the cloud usage. So it then boils down to "is the cost of this providing worthwhile return for xbox?" - which, while I know fuck all about how much this kinda stuff costs so this is all pulled completely out of my arse, it seems to be doing so far if only for the good PR for xbone and finally starting to deliver in some small way on that "infinite powah of teh clod" promise.

dynamiteReady wrote:

C++ for netcode this intricate though?  For the physics solver, yes, but I once had the good fortune to attend a talk in which a highly ranked Facebook employee openly discussed the problems one can run into when using multithreaded C++ code to write a highly distributed web server. The system the dude described was said to parse close to 40GB of data per second, which to my mind, is mental. But perhaps less so now, as Netflix, Twitch and Youtube servers will routinely chuck about far more data than that. This will be a very different application though... There was talk of the Halo 5 team using F# for their netcode

I dunno anything about Halo 5 and hadn't heard of F# before - but from a cursory google I couldn't find any mention of it in relation to Halo 5, only that they'd used F# for the Trueskill stuff in Halo 3 - which I would presume is an offline data/analytic processing task, certainly not realtime.

From cursory wiki read on F#, seems like it's a runtime interpreted/JIT language (like Java, .NET ... is C# JIT?), so in that respect it's highly unlikely to be used for their game server code simply for performance concerns. Typically you want to be as close to the ASM as possible, avoid any language with garbage collectors, use only a subset of C++ (e.g. no RTTI, pre-allocate as much as possible to avoid dynamic memory alloc/dealloc etc). Which is why I was pretty horrified when we got hired to port/tidy up Ace of Spades because it was mostly written in Python (fuck me). It's a miracle Minecraft works as well as it does in Java, but I'm pretty sure that when MS write Minecraft 2 it won't be in Java.

dynamiteReady wrote:

but then if that was Cloudengine's intention, that would surely feature in that Ad... So perhaps they are going to write the entire backend, from scratch, in C/C++... That would be very interesting...

I dunno, I don't think the language they choose to use isn't all that worthy of note TBH - the interesting part is the challenges they face and the techniques and designs of the solutions they come up with. Doesn't matter all that much what language the software uses or if it's from scratch or using 1st or 3rd party libraries, what's most interesting is the way they solve the thorny problems.

dynamiteReady wrote:

As for how it would fare over ASDL, I suspect it wouldn't be all that bad if the servers are in Ireland (speaking from a UK perspective), the host is centralised, and it's only object transforms you're requesting... It should actually be smoother than requesting a HD Twitch stream from the same location, because you'd be requesting less data if it's done correctly...

Well, as I've mentioned before, after reading about the effort Valve/Turtle rock went through to minimise the amount of data being sent for the horde of Zombies in Left 4 Dead, and we're dealing with several orders of magnitude *more* object transforms because of all the debris getting chucked about... the download bandwidth and lag is going to be pretty crucial for delivering on the promise of "infinite clod powah", not just for the quality on the player's receiving end but also for the cost of running the MP servers and the bandwidth out of the server center. 
But yeah, some quick post it note maths figures (which I always get wrong so pinch of salt here) reckons a twitch HD stream at 3.5Mbps = 437,500 bytes per second. Presume an uncompressed transform size of 37 bytes (3 floats for pos, vel, 4 for orientation quat) = 11824 transform updates per second = 394 transform updates each frame at 30 fps ...? Is that enough, having roughly 400 objects in flight? And we haven't included rotational velocity there for tumble/spin.
Obviously, lots you can do to compress transform for transmit, only send when an update needed (e.g. when there's actually a server-side physics update for that object because it has collided with another - otherwise you can let the local physics handle it... for instance if 100 bits of debris explode outwards from the top of a tower, you only need those 100 initial transforms and motion data then hand off to local physics until any of the pieces then hit the floor or something else) - but whenever there's client prediction it seems inevitable that lag will rear it's head and things will shift and move about... so, yeah, for folk with swanky 120Mbps fibre optic it'll probably be groovy, but it'll be interesting to hear how it plays for average punters on average ADSL.

(NB - this is the point where someone with much better maths than me points out how bad my maths is - Muzzy where you at? )