How does Uncharted 4 look and run so well

[Witchking]

Hey guys,

 

I came across this feed of uncharted 4 running at 1080p 60 fps (thanks to a neat slow motion mode unlocked) to show off how amazing this game actually looks. This is truly the pinnacle of current gen gaming. The details are just absolutely astounding. Everything reacts to the player, the crowds/AI are believable (enough), the environments are destructible. Everything reacts as expected.. wood breaks, sacs of rice spill over/leak when shot.. smoke has a realistic dispersion pattern.

 

 

Really some fantastic details and shows the attention to detail put in by those artists and animators. To imagine a console with limited computational power compared to modern PCs run so efficiently is truly a work of art. It shows how dedicated the developers were at providing the absolute best that they can.

 

It lead me to this: an interview of the tools used to create some of those FX: https://www.sidefx.com/stories/fx-adventures-in-uncharted-4-a-thiefs-end/

 

That lead me to this:

 

Take a look from 3 min. Interesting to see how you can develop some amazing FX. I know ED develops their own tools due to their own engine parameters and constraints. But I was lately thinking about the new water renderer in DCS 2.5. The screenshots may just be a glimpse...but I can't help but imagine if the new water renderer in 2.5 may actually have some really cool things. (PS: You can see some ways by which Uncharted also has a great water simulation in the same video above ) An actual true fluid simulation where ships and as you can expect, carriers, carrier landings get affected by the fluid motion. On top of this, I am excited to see where DCS will go with the FX and clouds that are still in development. Hopefully we will get to see some of the amazing things they are cooking up in Moscow. :)

 

Note: This is not meant to annoy / nag ED about visual FX. Its just what I came across while I was wondering how other games (non-simulations) can produce some truly fantastic and realistic looking visual FX. And yes, I do know they don't have to worry about simulating an A-10's / DCS aircraft's cockpit systems/avionics/physics/etc etc. ;)

 

cheers

 

V

[MikeMikeJuliet]

A few points that the consol-developement can use to get away with (apart from not needing to actually simulate details) are that a console generation has a constant hardware setup, meanin the developers can optimise the games to the extreme. This can't be done on pc for the reason alone that as many pc:s as possible ahould be able to run a game, so long they are at or above minimum specs. Then again at the same time pc games usually don't require such optimization due to the excess power available.

 

Secondly, many console games (especially FPS:s) are very limited in field of view, which reduces the amount of items needed to be rendered at any given instant. Many times further optimisation is achieved by enlargement of gun models. The advantage of running on a TV at a distance, is that the field of view doesn't need to be large for the game to feel good. Pc has a monitor quite close to the eyes, asking for a wider field of view to feel natural to the player.

 

Both systems have advantages, and such optimisation is indeed marvellous to see. Take any console's lifecycle and compare the earliest games to the latest ones, and you will see a huge improvement.

 

Sadly for flight sims, it is a nieche for us consumers, and probably even more so for developers. I highly doubt all the super talented individuals end up here, and as there are few of us in this playspace to begin with the talent required to create the most complex entertainment softwares (and professional training tools!) in the market simply falls to the hands of those willing, not necessarily those best qualified.

 

Do NOT get me wrong here! Each developer is a professional and very skillful, but I feel such complexity would require ALL the best developer minds for us to actually get "The Perfect Simulator".

 

 

These were my thoughts and a little analysis os why we see what you showed us.

 

Humble regards,

MikeMikeJuliet

[EagleEye]

Thanks for sharing Witchking! It`s always interesting to read about the background of such projects.

I really doubt we will ever see great visuals/FX and deep (flight physics) simulation at the same time (although one of recent WWII flight sims do have great FX at some point, i would say). FX is not that high in priority and it´s always a matter of budget.

[shagrat]

One of the most important things to consider are two simple facts.

 

1) procedural generation - e.g. Star Citizen or No Man's sky.

What is a real game changer for sci-fi or fantasy worlds, even normal maps of "jungle", "desert", "woodland", etc. Isn't easily applicable to a realistic sim world simulating real world places.

If you model a map of Nevada, or Abu Dhabi, or Normandy you expect the area's prominent features to be as in real life.

Now that doesn't mean there is no use to procedural generation in DCS, but it is more limited to things like trees/woods, bushes/gras.

With large realistic maps we still need a lot of prepared/preloaded regions and objects defined.

If you fly at 25k feet and look around you expect to see what you would see in real life.

 

2) optimizing field of view/object count/map layout for performance.

Most games today are designed, tested and optimized for performance from the start. A luxury we cannot achieve easily in DCS. Even in open world titles there is usually a tiny map area available compared to DCS. I remember this nice picture overlaying the ArmA Altis map onto DCS NTTR Las Vegas... so Vegas alone is bigger(!) than the complete ArmA map!

Titles like GTA, Assassin's Creed or Uncharted take a lot of efforts, to produce maps, where you can only reach places that allow for viewpoints the engine can handle.

City scapes are planned with large objects obstructing long range views, to optimize performance. Worst case you place a Skyscraper, hill or mountain in the medium range, to prevent a rendering area rendering 10 or more miles into the distance.

On foot, in a very detailed and angled environment, you don't notice that you actually walking in a very small physical space.

 

Now with DCS I can be anywhere on the map, on the ground or even at 40k feet in the sky above it... Still I expect a "believable" field of view with Las Vegas or Batumi visible in the distance where it is supposed to be.

 

This means the currently most effective optimization techniques available are if limited use and difficult to implement for DCS.

[Witchking]

Thanks for your input everyone.

 

I understand that FOV is a major thing that limits the amount of detail/performance in simulations such as DCS. Going up to 25k and being able to see the map as how it is supposed to be ...is something which console games don't have to worry about. But.. a lot of these games use sofware such as Umbra to cull objects out of view. I guess unlike DCS, the game is unfolding in front of the player (scripts etc)... On the other hand, a battle across the map in DCS is still happening inspite of the player not directly looking at those objects.

 

You are absolutely right.. the market share of sims such as DCS is too small of an incentive for a lot of new talent. It is interesting to see how games have pretty much become recipes made with various tools/ingredients. They have shareware for physics/destruction/fx/AI/trees/optimization etc. I know it is very easy to assume that ED can just plug those components into the sim, but the restrictions made by these middleware may not be worth the hassle to tailor a professional simulator that may require a more "controlled" output.

 

But hey.. it is always worth a look. Perhaps "playing" with a few of these tools might give ED devs some new ideas of how to cheaply implement "baked" physics/lighting etc to push their creativity further.

 

In the end, its just magnificent to see how far we have come: I started playing computer games / sims since Flanker 2 in 2001. We have now come to games which are better than some of the old pixar/pre-rendered visuals.

 

Heck... DCS 2.0 / 2.5 looks better than the pre-rendered flanker 2.5 intro. haha. I can't wait to see where this goes in the future. :)

 

NjpPMPe5AiI

[cichlidfan]

But hey.. it is always worth a look. Perhaps "playing" with a few of these tools might give ED devs some new ideas of how to cheaply implement "baked" physics/lighting etc to push their creativity further.

 

No offense intended, especially since I have no idea what you do for a living.

 

The reason that threads like this serve little value, to the developers involved in this sim, is the same reason that you don't need to show an astrophysicist the latest article by Stephen Hawking. Just about anyone working in a specialized field is very aware of the current state of their 'art'.

 

If you were a woodcarver, by trade or even hobby, and someone came out with some sort of revolutionary chisel design, you would be one of the first to know due to your reading habits and other connections related to your area of interest.

 

Just sayin'

 

EDIT: I also, won't even get into the fact that anything developed for a single platform (every system is identical) has inherent advantages over a PC based game.

Edited September 14, 2016 by cichlidfan

[MikeMikeJuliet]

No offense intended, especially since I have no idea what you do for a living.

 

The reason that threads like this serve little value, to the developers involved in this sim, is the same reason that you don't need to show an astrophysicist the latest article by Stephen Hawking. Just about anyone working in a specialized field is very aware of the current state of their 'art'.

 

If you were a woodcarver, by trade or even hobby, and someone came out with some sort of revolutionary chisel design, you would be one of the first to know due to your reading habits and other connections related to your area of interest.

 

Just sayin'

 

EDIT: I also, won't even get into the fact that anything developed for a single platform (every system is identical) has inherent advantages over a PC based game.

 

I guess the point of this whole thread is to peak the interest of forum users shed some light to the general issues with software developement, since so many people here are talking about "this and that" feature being easy to implement when they actually have no idea. And because this thread probably has no value to the developers it is here, in the chit-chat section.

 

People can discuss and wonder about things, right?

[Drag0nWIng]

I hope ps4 pro could have the size of ps4 slim and run win10 as the SysOS.

The sony-console-only games? I totally don't care about them.

Glory for AMD!

[Bushmanni]

You can also use partial pre-rendering in real time rendering to speed up things. The more dynamic your scenario, the less pre-rendering tricks you can use. For example if the lights you have in the scenario are static, you can pre-render the shadows and lightfields and add them as texture layers (used already in Quake). Same applies to the particle effects, ie. static lighting allows preshading of the particles so the effect looks more realistic with essentially zero additional computation (just compare Call of Duty explosions vs. DCS explosions). DCS uses pre-rendered ambient occlusion which means the shadows generated by blue sky light are pre-rendered and baked into the textures. As the sky lights up objects pretty much evenly from every direction, it looks pretty realistic even if the object rotates to different orientation. This looks the best in overcast weather as the lighting is most even in this weather.

 

Although DCS uses whatever tricks there are to improve graphics, the dynamic nature of the environment in DCS prevents the use of pretty much all the optimization tricks you see used in most other games.

[sobek]

1) procedural generation - e.g. Star Citizen or No Man's sky.

What is a real game changer for sci-fi or fantasy worlds, even normal maps of "jungle", "desert", "woodland", etc. Isn't easily applicable to a realistic sim world simulating real world places.

If you model a map of Nevada, or Abu Dhabi, or Normandy you expect the area's prominent features to be as in real life.

Now that doesn't mean there is no use to procedural generation in DCS, but it is more limited to things like trees/woods, bushes/gras.

With large realistic maps we still need a lot of prepared/preloaded regions and objects defined.

If you fly at 25k feet and look around you expect to see what you would see in real life.

 

Procedural generation can be made to set in at any LOD you want it to. It doesn't need to be an 'either/or' type of technology and at least according to Matts statements, they are working to integrate procedural techniques further down the line.

[shagrat]

Procedural generation can be made to set in at any LOD you want it to. It doesn't need to be an 'either/or' type of technology and at least according to Matts statements, they are working to integrate procedural techniques further down the line.

That was exactly what I meant. There is stuff that can be procedurally generated, still look awesome and realistic.

Yet, the full extent of that technology (e.g. rendering complete planetscapes) isn't easy to put.

If you generate a forest area, or Cityscape from procedural algorithms, they may look utterly realistic, but I can't see an easy way to put each specific Building in the correct space and every hill and wood line in the place it would be in the real world...

Though it is possible to generate "generic" trees and forests, bush and grass, built up areas etc.

The overall "layout" of the world, height map, etc. will still require resources, other games can utilize for other stuff.

 

 

Now that doesn't mean there is no use to procedural generation in DCS, but it is more limited to things like trees/woods, bushes/gras.

[sobek]

I'm not sure procedural generation can do much for you in part of rendering performance at all. It can help reduce the physical memory footprint of theaters though.

[shagrat]

I'm not sure procedural generation can do much for you in part of rendering performance at all. It can help reduce the physical memory footprint of theaters though.

It can't, even if the complete scene is build from a number crunching algorithm, it still needs to be rendered and post processed in 60fps plus.

The GPU still needs to do its job.

 

For the CPU, RAM and data shuffling (memory organisation) it should have an impact. Rather than shuffling objects, textures and depth information in a complex way to have everything for a scene available, you can just drop every part not in the sightlines and free up the memory... Without any CPU cycles used for managing storage/RAM optimizations.

 

If the procedurally generated grass for example doesn't need a texture file to be loaded, it shouldn't matter how detailed the generated texture/object is...

 

Just my two cents, I'm no math genius, but from what I've read, this can free up CPU and a lot of RAM resource, which is often a bottleneck.

[sobek]

I don't know if this is feasible, because these assets are generated from more or less complicated algorithms. I don't think that you can empty them from the cache, as regenerating them would most likely exceed frame time by a large margin.

 

I'm no specialist on this either, but the largest advantage i see is that you can get the footprint of theaters down on the non volatile memory (HDD/SDD, not RAM).

Edited September 16, 2016 by sobek

[MikeMikeJuliet]

Procedural generation can be made to set in at any LOD you want it to. It doesn't need to be an 'either/or' type of technology and at least according to Matts statements, they are working to integrate procedural techniques further down the line.

 

On details like bushes, grass and ground elevation noise perhaps, but not on things that are always expected to be there... so practically buildings and everything larger.

[sobek]

On details like bushes, grass and ground elevation noise perhaps, but not on things that are always expected to be there... so practically buildings and everything larger.

 

Yes, this was sort of my point.

 

Procedural does not imply statistical distribution. In fact, for things like multiplayer to work, you need procedural generation to be absolutely deterministic at any LOD, that is, to generate the same result for the same parameters each time the algorithm is called by any client. So even though the elevation noise or texture noise might be generated by an algorithm, the same noise speck will always end up landing at the same terrain spot.

 

The problem is that while you can impose 'rules' onto procedural generators on where to put water or what kinds of terrain features will likely be populated with trees, if it comes to landmark features like buildings, you'd have to impose so complicated restrictions on the generator that you just might as well skip the procedural aspect in these parts and generate them the old fashioned way, by hand.

[shagrat]

Yes, this was sort of my point.

 

Procedural does not imply statistical distribution. In fact, for things like multiplayer to work, you need procedural generation to be absolutely deterministic at any LOD, that is, to generate the same result for the same parameters each time the algorithm is called by any client. So even though the elevation noise or texture noise might be generated by an algorithm, the same noise speck will always end up landing at the same terrain spot.

 

The problem is that while you can impose 'rules' onto procedural generators on where to put water or what kinds of terrain features will likely be populated with trees, if it comes to landmark features like buildings, you'd have to impose so complicated restrictions on the generator that you just might as well skip the procedural aspect in these parts and generate them the old fashioned way, by hand.

 

Yep, exactly this. As shown in the 2.5 updates they work on this already, at least they had a nice grassland shot, obviously with the new procedural grass.

 

I would love this for trees and ground crop, bushes especially, but the graphics engine needs to be capable of integrating it. Guess that is consuming some work...