Technical Comparison - Killzone 2/Killzone 3 vs Crysis 2 (console version)

[kagemaru]

Problem is in C2, world shadows are all there are. Local lights generate no shadows. The main thing you'll notice is that in C2, firing your weapon in a dark area doesn't light it up much, almost like a last-gen game. In KZ2, muzzle flashes result in some pretty impressive effects. KZ2/3 have far more local lights, which result in far more shadows reacting to far more of the action. As a matter of technical comparison, shadows that respond to all lights > shadows that respond only to area lights. Deal with it.

Deferred shading also allowed them to use MSAA in KZ2. As a matter of technical comparsion, MSAA > jaggies. Deal with it.

http://www.guerrilla-games.com/publications/dr_kz2_rsx_dev07.pdf

You will notice from the presentation that all those complex dynamic lights they're able to do (and Crysis can't) due to deferred shading preclude the use of HDR. That's the tradeoff.

I don't have C2 any more, but it had a whole lot more fidelity issues than KZ2/3 had. Have you played any of the games?

CE3 uses deferred lighting (light pre-pass) and can also display many lights at a time. Also nothing stops developers from using MSAA with deferred rendering, it's just the bloated memory consumption that is an issue (no different for KZ). An additional buffer is also necessary for transparencies I believe, consuming more memory. There's also an issue with material shaders in a fully deferred set up IIRC.

KZ's use of light and shadow maps can produce some cool effects, but the use is limited IIRC (as in muzzle flashes not casting shadows with a certain number of characters on screen). A developer from 4A games commented on the limitation of their lighting in a DF interview.

I could be off on all this though.

This. I agree with this. Now if you could L. Scofield and Laa-Yosh to agree instead of pretending Crysis makes zero meaningful trade-offs in order to have nice scene lighting. And you don't see me or anyone else claiming there are no technical sacrifices in KZ. There are technical sacrifices in every game ever made for a finite-state machine.

Well while I'm not as bothered by the lack of precision as Laa-Yosh, I don't really disagree with his points. As for Scofield, I don't agree with his approach, but that's between you two.

Also my comments aren't really directed towards you specifically. Just tying to view things from both sides.

 

[L. Scofield]

Problem is in C2, world shadows are all there are. Local lights generate no shadows. The main thing you'll notice is that in C2, firing your weapon in a dark area doesn't light it up much, almost like a last-gen game. In KZ2, muzzle flashes result in some pretty impressive effects. KZ2/3 have far more local lights, which result in far more shadows reacting to far more of the action. As a matter of technical comparison, shadows that respond to all lights > shadows that respond only to area lights. Deal with it.

At this point I know I'm just feeding a troll, but what the hell

Deferred shading also allowed them to use MSAA in KZ2. As a matter of technical comparsion, MSAA > jaggies. Deal with it.

Since when can MSAA only be used in a deferred renderer? Oh right, since never. KZ2 uses quincunx MSAA which is not that great since it blurs the image too much. C2 TAA is not particularly better though.

http://www.guerrilla-games.com/publications/dr_kz2_rsx_dev07.pdf

You will notice from the presentation that all those complex dynamic lights they're able to do (and Crysis can't) due to deferred shading preclude the use of HDR. That's the tradeoff.

That document contains no references of any kind to HDR lighting, so how did you infer that DR precludes the use of HDR? It's nonsense of course since Little Big Planet 1 and 2 (among others) use a deferred renderer with HDR support.

Crysis 2 uses a hybrid renderer. It allows it to render dozens of lights on screen while keeping the abilty to use complex shaders. All while supporting a pretty good linear-space HDR implementation, SSAO and IBL (including reflections, KZ2/3's is limited to the diffuse term). Too bad the PS3 can't handle it

Plenty of info here:

http://www.crytek.com/cryengine/presentations/

This. I agree with this. Now if you could L. Scofield and Laa-Yosh to agree instead of pretending Crysis makes zero meaningful trade-offs in order to have nice scene lighting. And you don't see me or anyone else claiming there are no technical sacrifices in KZ. There are technical sacrifices in every game ever made for a finite-state machine.

When did Laa-Yosh or myself said there are no trade-offs? We simply spoke about why HDR is important and then you just made excuses about why it's missing in KZ2/3 and acted all defensive. Chill.

 

[kagemaru]

At this point I know I'm just feeding a troll, but what the hell

Since when can MSAA only be used in a deferred renderer? Oh right, since never. KZ2 uses quincunx MSAA which is not that great since it blurs the image too much. C2 TAA is not particularly better though.


That document contains no references of any kind to HDR lighting, so how did you infer that DR precludes the use of HDR? It's nonsense of course since Little Big Planet 1 and 2 (among others) use a deferred renderer with HDR support.

Crysis 2 uses a hybrid renderer. It allows it to render dozens of lights on screen while keeping the abilty to use complex shaders. All while supporting a pretty good linear-space HDR implementation, SSAO and IBL (including reflections, KZ2/3's is limited to the diffuse term). Too bad the PS3 can't handle it

Plenty of info here:

http://www.crytek.com/cryengine/presentations/


When did Laa-Yosh or myself said there are no trade-offs? We simply spoke about why HDR is important and then you just made excuses about why it's missing in KZ2/3 and acted all defensive. Chill.

-I really don't think he's trolling, just mistaken.

-QAA is essentially 2xMSAA with a blur filter, so unless I'm mistaken, it should have the same memory cost.

-I don't think he's the only one that needs to chill.

 

[L. Scofield]

-I don't think he's the only one that needs to chill.

When you're battling misinformation it's though not to get a little worked up

 

[Billy Idol]

Properly calculated lighting is a must.

As I understand it, real time lighting is always a rather inaccurate approximation of reality. Can you define 'proper'? Proper does not sound like a technical term to me.

 

[Shifty Geezer]

While stylish, the KZ games use reality as the foundation for their world. Properly calculated lighting is a must.

Surely KZ3 is using proper lighting calculations, with an angle of incidence affcting the illumination level from a given lightsource. Not that many games do that these days, because instead they just paste a lightmap on.

HDR isn't essential though, depending on the environment conditions. It's a desired bonus, just as is smooth shadows, dynamic lights, etc. as everyone else is saying. Laa-Yosh may have a point regards how the game looks in full, but his criticism of that particular reference image doesn't point to technical issues with the engine, or that Guerilla failed to calculate decent lighting approximations.

If we go back to Laa-Yosh's original comment:

That is of course an opinion. I can not tolerate the color issues in the KZ engines; my preference would be the Uncharted games on PS3 and probably the new Quantic Dream game, Beyond.

...it's these colour issues that are the crux of the matter. Laa-Yosh thinks some people are being deliberately blinkered to them, but he hasn't gone on to describe what exactly is wrong with them other than one example of a post-effect clamping the highlights. There's no pointing to colour banding, wrong colour, colour-bleed, colours being wrong on reflection, um...whatever other colour issues there might be.

I for one am actually interested in the colour problems of the KZ engine that I'm not seeing, and would like them spelled out for me because I don't know what to look for. If these were described, then screenshots of different games could be compared for examples of where one game is making compromises the other isn't. Not all of us are CGI professionals who know what to look for though, so aren't readily equipped to notice these compromises ourselves unless they are obvious.

 

[kagemaru]

When you're battling misinformation it's though not to get a little worked up

That should only be the case if people aren't open enough to what you're saying.

I for one am actually interested in the colour problems of the KZ engine that I'm not seeing, and would like them spelled out for me because I don't know what to look for. If these were described, then screenshots of different games could be compared for examples of where one game is making compromises the other isn't. Not all of us are CGI professionals who know what to look for though, so aren't readily equipped to notice these compromises ourselves unless they are obvious.

If this request was made earlier in the thread, instead of others arguing with him, there's a chance he could have done this before leaving the thread.

Not speaking to you, but some other posters here should be more considerate towards those who obviously know more than them and try to gain what they can from such discussions instead of holding firm to their limited opinion.

 

[Billy Idol]

That should only be the case if people aren't open enough to what you're saying.

If this request was made earlier in the thread, instead of others arguing with him, there's a chance he could have done this before leaving the thread.

Not speaking to you, but some other posters here should be more considerate towards those who obviously know more than them and try to gain what they can from such discussions instead of holding firm to their limited opinion.

But it is exactly people like you that only want to derail threads without participating a discussion, but instead accusing everyone of being a PS3 fanboy. Sorry pal for being this direct this time, but half of your posts are about screaming PS3 fanboys, the other half finish with a razz lolcat.

With respect to Laa-Yosh: I guess he is old enough to speak for himself. I like his posts, often I gain some information reading them. I am with Shifty, it would be great if he could explain his posts a little bit more as I am not professional CGI artist and it is often difficult to follow his argumentation. I look at the KZ3 screenshot and think that it looks good, even very good. Laa-Yosh says it is terrible looking. I just want to know why, I am honestly interested in the reasons, in his opinion...

 

[kagemaru]

But it is exactly people like you that only want to derail threads without participating a discussion, but instead accusing everyone of being a PS3 fanboy. Sorry pal for being this direct this time, but half of your posts are about screaming PS3 fanboys, the other half finish with a razz lolcat.

So who did I scream at for being a ps3 fanboy? All I've done is point out how some people are dismissive to certain points because it goes against their platform of choice. Which is the core issue with threads like these. Go ahead and create a thread for people to circle jerk on the 360, and I'll be apposed to it just the same. Edit: This is a poor thread that was created to play up a PS3-exclusive, there was no way this was going to go how some of us had hoped (more technical, less opinionated), don't go blaming me for that.

 

[Arwin]

So who did I scream at for being a ps3 fanboy? All I've done is point out how some people are dismissive to certain points because it goes against their platform of choice. Which is the core issue with threads like these. Go ahead and create a thread for people to circle jerk on the 360, and I'll be apposed to it just the same. Edit: This is a poor thread that was created to play up a PS3-exclusive, there was no way this was going to go how some of us had hoped (more technical, less opinionated), don't go blaming me for that.

You just made his point very eloquently though.

 

[fearsomepirate]

CE3 uses deferred lighting (light pre-pass)

Killzone 2/3 uses a fully deferred renderer. Crysis 2 uses a partially deferred renderer. There's an argument (a subjective one) to be made for the hybrid approach being preferable, but the fact is that KZ's approach precludes HDR. See every technical presentation GG's made for details.

Off the top of my head, here is what I remember in the PS3 version of C2:
Good:
-Best outdoor lighting I've seen in a console game (HDR + SSAO + single global light bounce and all that jazz).
-Plenty of geometry.
-Post-patch, the frame rate was actually pretty stable
-Large areas, many games that look half this good don't have such open areas.

Bad:
-When the frame rate did drop, it was pretty bad.
-No AA most of the time
-Enemy models were pretty simple, and there was a pretty obvious tradeoff between number of mobs and scene complexity.
-Enemies often didn't cast shadows. This seemed to be a bigger problem indoors.
-Local lights rarely (never?) cast shadows. Was there a flashlight? Seems like every game has one, and if it did, I'm sure it cast shadows.
-Texture quality was extremely variable. Underneath the downed saucer (before you see your first live alien), it was so bad I couldn't tell what I was supposed to be looking at.
-Particles looked awful at times. The "nanotech" particles swarming downed aliens looked especially bad, very pixelated. Dust/haze/fire/smoke pretty primitive most of the time. Some effects were very pixelated.
-Animation wasn't very good. Lots of dropped frames, especially at a distance
-Lots of LOD pop-in
-Shadows were poorly rendered more than a few yards away from the player, and there was no smooth interpolation between the "good" area of the shadow and the "crappy" area. This could be very distracting in areas where it was an issue.
-Not much of the way in effects other than lighting.
-Reflective/metallic surfaces look awful (i.e. the alien ships and alien armor).
-Indoor areas didn't look very good, since a realistic light bounce has less dramatic results there, and the lack of everything else we've seen this gen is more apparent.

I'm not in any way denying or discounting how well it's revolutionary lighting is implemented. But it does so many things poorly that I don't think it's far and away obvious that it's the best technical accomplishment on the PS3, especially not if the only significant criticism of KZ3 is that it sacrifices the ability to use a floating-point RGB buffer to do the other things it does.

That document contains no references of any kind to HDR lighting, so how did you infer that DR precludes the use of HDR?

Lots more info here:
http://www.slideshare.net/guerrillagames

 

[kagemaru]

You just made his point very eloquently though.

Of course I did.

Forget this thread, it's a joke.

BTW regarding your earlier answer about KZ3's geometry with the picture I linked to, you're wrong and it's been discussed here before:

When people praise the amount of geometry detail in pictures like these, I almost wish I didn't do 3D for a living. I would love to believe that these scenes are highly detailed. Unfortunately, flat surfaces with flat texture maps jump out at me like Linda Blair in those scary maze pranks.

I'm not singling out this game at all, I just comment on what I see in those pictures. The buildings and the ground use so few polygons that you can pretty much count them (I'm not kidding). I could draw a wireframe on top by hand. The large objects scattered on the floor use very few triangles as well. In fact, the rifle appears to be the densest object in the shot.

I'd estimate the amount of visible triangles in a shot like this at no more than 100,000.

http://forum.beyond3d.com/showpost.php?p=1532632&postcount=240

Some assessment, whether you choose to believe it is up to you. Like I said, I would love to be in your shoes.
Funny enough, the low wall around the fountain is reminiscent of the Killzone buildings in terms of detail.

http://forum.beyond3d.com/showpost.php?p=1532638&postcount=244

But hey apparently it's ok to have these same threads, and assumptions, pop up over and over again.

 

[fearsomepirate]

I thought I was getting deja vu.

http://forum.beyond3d.com/showthread.php?t=57523&page=80
http://forum.beyond3d.com/showthread.php?p=1532638#post1532638

I'll just repeat what I said before. At the end of the day, the only "objective" comparison is whether (a) all the available computational resources were used, and (b) whether the methods used were efficient. You can argue until the cows come home about which use of resources is best (e.g., should I use the available fill for more particles or higher resolution?) and never come to a conclusion everyone will agree on, because that's ultimately a subjective, aesthetic judgment.

 

[Scott_Arm]

This has been a remarkably non-technical comparison from the start.

 

[Arwin]

Of course I did.

Forget this thread, it's a joke.

BTW regarding your earlier answer about KZ3's geometry with the picture I linked to, you're wrong and it's been discussed here before:

I have a PSP game with several of those horses in it - I have a feeling you don't need as many polys as you may think. The Killzone 3 shot is also zoomed in more (even if the pixels were the same size it would be, because its running 20% more pixels).

In general, even if there a lot of small relatively square objects littering the place in Killzone, this is nothing compared to the huge square buildings all over the place in Crysis 2.

But hey apparently it's ok to have these same threads, and assumptions, pop up over and over again.

You don't have to be here, and if all you care about is accusing people of partisanship, then you're worse than the disease you're trying to cure, if that's what you're here for. Take your own advice, and forget this thread and leave it for people who are still interested.

 

[sebbbi]

No, the previous one. Although this one also has the same problems - high end of the histogram flattened out, highlights clamped at 220-200 intensities (particularly disturbing with the sun).

Heh... this comment brought back old memories.

We discussed about the exactly same issues with our artists several times (we had only 2x dynamic range as 60 fps pretty much prevented float formats). At least we had linear space lighting in Trials Evolution. Trials HD did lighting in gamma space, and thus all it took was two light sources to burn the surface details away (pretty bad since the bike had head light = at least two light sources in the areas where the player was focusing his/her attention). We had to tune lighting/ambient down, and the game received a murky dark outlook (we added dusty fog to enhance it further). This was ok for a warehouse based indoor game, but for the sequel (bright outdoor world) the first thing we did was to implement proper linear space lighting.

I am surprised that they managed to make the game look this good with simple gamma space lighting. Must have been a pain for the artists to tune all the environments to look as they wanted (balance between burning and flat look is very hard to control). I am eagerly waiting to see that they achieve with next generation consoles (and physically correct lighting).

 

[Shifty Geezer]

Heh... this comment brought back old memories.

We discussed about the exactly same issues with our artists several times...

Can you please explain the results of this problem regards that screenshot? AFAICS the clipping Laa-Yosh is talking about is happening after the full screen-space dynamic range of 0-255 has been mapped. It would be quite possible, and I presume was an option for the artists, to render the sun and sky and highlights all suitably blown out instead of capped to a brightness of 200-220. Where is HDR letting down this particular scene, and how was that a concern for the game in general?

Some years ago there was talk on this board, including myself, of how a screenshot cannot tell you if HDR or LDR is being used, in the same way a photograph can't tell you whether it was taken with HDR or LDR - it all gets truncated down to screen colourspace, so if it was created at that colourspace or not you can't tell. Now it sounds like people are looking at screenshots and saying the lack of HDR is hurting them. I can understand HDR in motion, but I'm failing to understand the issues being suggested here.

 

[Graham]

but the fact is that KZ's approach precludes HDR.

Not strictly true.
It restricts the precision of baked/static/emissive lighting to what can be fit in the g-buffer (which will typically mean 8bit RGB or RGBM), but not dynamic lighting from point lights, etc - which can be processed in FP16, etc, in a second pass.

 

[Scott_Arm]

Trying to understand Sebbbi's post and found the following links talking about gamma space vs linear space:

http://filmicgames.com/archives/299
http://molecularmusings.wordpress.com/2011/11/21/gamma-correct-rendering/
http://http.developer.nvidia.com/GPUGems3/gpugems3_ch24.html

Edit:
And another http://renderwonk.com/blog/index.php/archive/adventures-with-gamma-correct-rendering/

So basically, if you do LDR lighting in gamma space, any lighting math, in terms of adding or subtracting intensities, is going to be totally wrong (1 + 1 = 3), because gamma space is non-linear. If you work in HDR, in linear-space, adding lights of two intensities will equal a linear a sum of those intensities (1 + 1 = 2). The former might "look good", but it will be physically incorrect and difficult for artists to work with as the results of lighting will be inconsistent, less predictable.

Am I roughly right?

 

[L. Scofield]

You know, for a supposedly technical oriented community, not knowing what HDR is useful for or why it's important, even though it's been a key feature of the renderers of every major game this gen (except KZ's) says how much this community has decayed, as pointed by Laa-Yosh.

Let's just take a look at fearsomepirate's post: "Crysis 2 only has couple of good effects, everything else sucks. KZ2/3 are gen above the competition, except when it comes to HDR but that's not a shortcoming, it's a feature"