Cry Engine 3 possible on DCS?

[sinelnic]

Yeah... Considering ED hasn't yet implemented multicore support, I believe your DX11 card money can be better invested in multiple monitors, which can give you a much richer and realistic experience as soon as today.

[SUBS17]

Not that people with multicontrollers had problems getting their setup to work proply.

 

Arma did not have multicontroller support.

[EtherealN]

The thing with using the graphics card for physics computation is that it is not always all that good at it - at least not realtime physics of this detail. The raw processing power is fantastic, but the latency for each calculation is much greater than in a "normal" general purpose CPU.

 

Now, it's hard to say if that will actually be an issue in this case. If the latency is inconsequential it could indeed be pretty nice, but it's not necesarily simpler to implement than just getting multithreading. Without access to the source code and documentation from nVidia it would be very difficult to say.

[GGTharos]

Yoda is currently working with OpenCL. You might want to solicit his opinion, but from what I've heard, using it isn't going to be quite as easy as most people think (porting is a serious effort, and not ALL problems are parallelizeable)

[T.Goat]

I would think that a combat flight sim would be a good candidate for multithreading, especially one that simulates as much as DCS does. Between AI behavior, missile and radar logic, rocket/bomb/shell ballistics, all the systems in the aircraft(s) and airframe stress and damage, the flight model, world and collision. Keep in mind though that I am by no means a programmer :music_whistling:

[EtherealN]

One of the issues, T-Goat, is that the massively parallellised architectures used in GPU's do have a phenomenal computation power - but there is considerable latency to each computation. That means that while you may be able to do X amount of calculations in Y amount of time, it may be that you don't get the results from those calculations when you need them.

 

This is usually not an issue in regular graphics processing since you only need to fill the screen some 60 times a second and since graphics processing can be made to run in a large part independent of the rest of the system (no need to pass information back to the CPU), which decreases risks of choking up the databuses on the motherboard. Simulating aircraft behaviour might (I don't know for sure) require faster updating than the latency in cGPU design permits, and it might require too much communication to and from the CPU.

 

"Normal" multithreading, however, is indeed a good thing and there are less architectural problems as far as I can see. But you should recall that it is estimated that a multithreaded application can cost several times more to develop due to the inherent difficulties in keeping everything synchronized. Rise of Flight found what I believe to be a good middle-ground where the process will simply split into giving a core all the physics and another everything else. However, they had the benefit of writing a product from scratch with this in mind - I don't envy the people that end up tasked with rewriting the LO/DCS engine for multithreading. (Though it'll still be sweet when they make it. :) )

 

EDIT:

A small addition that might help illustrate possible difficulties - an inherent advantage of in-line computing (that is, only one thing at a time) is that you will never end up without data for the next calculation. However, when you parallellise an application you need to make sure that you never end up with a calculation needing data from another calculation that's being run at the same time on another core. Therefore the programmers and design people need to look very very carefully at the process to eliminate such things. If Core 1 is about to collect the data on how the helicopter moves in the next instant but only 20 out of the other 200 cores on the chip have completed their calculations for their individual airframe elements things become "funny".

 

This is a problem that I would expect to get worse when the physics become more complex. Moving a ragdoll body around isn't all that complex and can be done in good pace, but when you are working on airflow, inertias, humidity, air density and all those fun things to figure out how a complex aircraft is supposed to behave you end up with a lot of calculations needing information from a lot of sources - and so far I'm not sure I've seen a cGPU architecture that can do that reliably (though there's always geniuses out there that find a way, hell, I remember the tricks people did to make the C64 (or was it the Spectrum?) play 16-channel audio on a 4-channel sound cirquit). Each processing unit must be able to all the data in, and the whole thing must be planned such that they all get their information at the same time so that the next step will have all it's information.

 

Not impossible, but oh my I'm happy I don't have a pointy-haired-boss asking me to do that. :P

Edited October 22, 2009 by EtherealN

[GGTharos]

I would think that people would have already understood, even though they are not programmers, that while in general everyone agrees on this point, it isn't some darned cakewalk that can be performed on an existing engine overnight - nor can development suddenly stop just for the engine to be made capable of this because everyone and their grandmother thinks it would be oh-such-a-spiffy idea.

 

 

I would think that a combat flight sim would be a good candidate for multithreading, especially one that simulates as much as DCS does. Between AI behavior, missile and radar logic, rocket/bomb/shell ballistics, all the systems in the aircraft(s) and airframe stress and damage, the flight model, world and collision. Keep in mind though that I am by no means a programmer :music_whistling:

[T.Goat]

EtherealN, I was actually thinking more of CPU multithreading, not GPU.

 

GGTharos, I wasn't saying that it would as easy as "numbproc=4", or copy/paste DCS.exe into Cryengine or whatever else. I understand that a small game/sim studio doesn't have the resources to take advantage of every new technology, and that they probably have a very small $ return for each hour used because of the very small market. But 75% of computers now have 2 or more cores, and it would be nice to know that ED is harnessing that potential, especially with Radar/ECM-carrying fighters possible on the distant horizon.

[EtherealN]

Judging by Valve's data on Steam users, you are overestimating the average home computer.

[Vekkinho]

Ok, just to demistify, here's what Cryengine 2 can do for a flight sim. Beware! this can be shocking!

 

http://www.crymod.com/uploads/mediapool/20090901FA/3.jpg

 

 

:huh:

 

I played couple of flight sims 10-15 years ago that had better eye candy!

 

Seriousely, the only shock related to that pic is that we're closing to 2010!

 

Check this for instance:

http://www.migman.com/ref/1990_combat/F16MiG29/MiG29.htm

Edited October 22, 2009 by Vekkinho

[Logan9773]

Yeah, I would call the graphics from high above "average". Decent enough to get away with, but not the greatest. However, you have to remember that what you are looking down on is all 3D. Every tree, every blade of grass, every bush, tanks, buildings - all exist in 3D. I would call that quite an achievement. Its what a helicopter sim really needs, as they are ambush predators. Plus, in MP, human intelligence is behind almost everything, which makes it a much more rewarding experience. The one thing Crysis lacks, just like Arma, is realistic flight and vehicle simulation.

 

Its needs to bring in Steel Beasts and Black Shark realism.

[EtherealN]

If you are referring to the picture posted, that's not looking down on 3D. That's a cleverly bumpmapped 2D texture. Heard of the thing called LOD?

 

There is no reason to use fully 3D objects at long distance, because they add nothing to immersion nor realism at those distances. What you do is that you keep track of distances to objects and switched them from 2D then to different LOD 3D objects as you get closer.

 

Hint: Cryengine 2 and 3 both do this. Not all of those awesome 3D thingies you're looking at there are 3D. Some are 2D objects with some well applied bumpmaps and parrallax mapping.

 

The only thing you'd achieve through rendering everything as 3D at the distances we're talking about is to make it completely impossible to run on any home PC.

[GGTharos]

Actually this is pretty much what you're saying - that it is this easy, and because <see your explanation below> ED should be doing it - if it isn't what you meant, it certainly came off this way.

 

So let's take it from the top:

 

ED agrees that multi-threading DCS would be awesome.

ED disagrees that multi-threading should be top priority.

ED does wish to implement multi-threading at some point.

ED currently places a higher priority on -other- features.

 

Re-architecting an entire game like this is not easy, nor fast, and I'll say again, they aren't going to stop doing what they're doing just to do so - my guess? The new code that is replacing old code within the engine is probably being written to be more easily transformed to MT, and once the amount of this code hits a certain percentage they will deal with ther est of the code that needs revamping.

 

In other words they'll do so when they are ready, and no matter how much we or they think 'it would be nice', that won't make it happen any sooner.

 

GGTharos, I wasn't saying that it would as easy as "numbproc=4", or copy/paste DCS.exe into Cryengine or whatever else. I understand that a small game/sim studio doesn't have the resources to take advantage of every new technology, and that they probably have a very small $ return for each hour used because of the very small market. But 75% of computers now have 2 or more cores, and it would be nice to know that ED is harnessing that potential, especially with Radar/ECM-carrying fighters possible on the distant horizon.

[Logan9773]

If you are referring to the picture posted, that's not looking down on 3D. That's a cleverly bumpmapped 2D texture. Heard of the thing called LOD?

 

There is no reason to use fully 3D objects at long distance, because they add nothing to immersion nor realism at those distances. What you do is that you keep track of distances to objects and switched them from 2D then to different LOD 3D objects as you get closer.

 

Hint: Cryengine 2 and 3 both do this. Not all of those awesome 3D thingies you're looking at there are 3D. Some are 2D objects with some well applied bumpmaps and parrallax mapping.

 

The only thing you'd achieve through rendering everything as 3D at the distances we're talking about is to make it completely impossible to run on any home PC.

 

I'm saying its a living world (there's even fish and sharks swimming around in the ocean ). There are human controlled players and vehicles moving on the ground. Its just one mass battlefield with almost everything of consequence controlled by human intelligence. The AI is decent, but I hate AI. I would just like to bring realistic flight sims into a high level FPS like Crysis. I can envision linking a bunch of servers together to run one map with multiple battles going on simultaneously in different hot spots and CAS, CAP, Strike, and BDA going on from above. All in a nice jungle setting. :D

[EtherealN]

And I can envision the amount of hardware you'd need for that, and the phenomenal bankruptcy suffered by the developer that spends money to create that and finds less than a dollar in revenue for each hundred dollars they've spent making it.

 

Pipe dream, I'm afraid. You'll have to wait a long time until something like that is feasible technically, and possibly even longer until it becomes feasible as a marketable product. If ever. That's the harsh reality.

[sweinhart3]

Sometimes we really do not look out of the window. It is called "release rush" ;)

But seriously, my view on the situation with CryEngine in DCS:

1) from informed sources, crysis source code is a horrible mess (like almost any complex, break-through game).

2) there're no shipped titles, based on cryengine 3

3) we are definitely capable of releasing new products on our own technology.

 

So, there's no guarantee that it is possible to create a DCS-level simulator on a CryEngine3.

What if we start to create next DCS sim on a CE3, and get stuck into some engine deficiency in the middle of the process? And note, this is a huge, complex engine written by people who did only FPS all their life. It can easily get to the point that this engine need a complete rewrite. As a result you will never ever get any DCS: xxx at all.

 

So, we will just humbly, painfully continue to develop our own, may be not spectacular, but definitely good working technology.

 

And you really don't have an option to have CE3 in DCS ;)

 

 

I would never expect CryEngine graphics in a simulator, however, and Im not an expert programmer, but why dont you just create different kinds of trees, grass, and other foilage and then have it auto generate? I realise that buildings and vehicles couldnt be done this way because they have to be set in place or moving on a specified path no matter what. How bout infantry? so far Ive only come across stick figures. Some animation would be nice but less necessary than having a full figured person over a stick model. While not sharing code with outsiders, Im sure it wouldnt be hard to see other games such as fps (Far Cry 2) and work to generate realism with your own code. Im sure you guys are doing something along those lines but Ive really got the impression that graphics are mostly on the side burner. I enjoy the realism of the flight engine but I dont think it would be difficult to stress a video more simply by increasing the amount autogen to fill the ground with seasonally textured 3d models (be it clouds, trees, grass, etc.) the more there is the more the video card will have to render.

[EtherealN]

Swein, I don't recall which thread it was in, but there was some screenies released with grass. So they are indeed working on that stuff. ;)

[ZaltysZ]

Model is partial representation of reality. It means that when modelling you pay attention to most essential things and give low priorities to unimportant details. If you have enough resources, you can extend the model by small features and details, however once again the first are implemented the most important details and so on. Now, what do you think is important for flight sim? You can say this and that, but please try at first to group the features and assign priority number to them. You will see how quickly grass, butterflies, birds, nice water, "living" environment go to the bottom of the list and features like flight and damage models come up.

 

By the way. Lets take ArmA as example. It looks nice when viewed from infantry perspective, but take a jet and you will notice how small ArmA's world is and how bad eyes of pilot are modeled. :) The same is true for any FPS engine, because they are created not for flight. I don't care if I can not see any grass, but I certainly care if I can only see a few km ahead on clear day. :)

[159th_Viper]

Swein, I don't recall which thread it was in, but there was some screenies released with grass. So they are indeed working on that stuff. ;)

 

 

[sweinhart3]

 

 

That looks pretty neat but do you think they will actually have that when A-10 rolls out? From some of the stuff I read and videos dont remember where ED showed some kind of joint venture with another company that would basically coordinate 3d ground troops with aircraft (i.e. calling in airstrikes). Not trying to be negative what they showed would be a complete graphics / engine overhaul. I just dont see that coming out in a years time. BTW what the heck is that ugly animal doing on the runway?

[GGTharos]

You're probably talking about the military branch of simulations coming from ED, which did include some sort of FAC.

 

What you are seeing in the screenshot there is already in the engine for the A-10C ... so be negative all you like, but that's like not believing a hot babe is asking you out, while she's right there doing so ;)

 

 

That looks pretty neat but do you think they will actually have that when A-10 rolls out? From some of the stuff I read and videos dont remember where ED showed some kind of joint venture with another company that would basically coordinate 3d ground troops with aircraft (i.e. calling in airstrikes). Not trying to be negative what they showed would be a complete graphics / engine overhaul. I just dont see that coming out in a years time. BTW what the heck is that ugly animal doing on the runway?

[159th_Viper]

.....BTW what the heck is that ugly animal doing on the runway?

 

Ugly?

 

There's a Face that Strikes Fear into even a Leopard!

 

 

Who you calling Ugly? Fookin' Gorgeous I would say :D

[EtherealN]

I'm not sure I should comment... :P

 

Let's just say I think the A10C is prettier than the animal it's named after. ;)

[sobek]

Who you calling Ugly? Fookin' Gorgeous I would say :D

 

Has the EMO hairdo encroached over to animals now? :)

[joey45]

Has the EMO hairdo encroached over to animals now? :)

 

nice one.