*merged/spinoff* for the Neverending Killzone Discussion on Graphics

[patsu]

But what does "muddy" mean? In all honesty, when it's about Killzone 3 it doesn't have to mean a thing.

...

I judge by screenshots or videos and the muddyness isn't as present as it was before, but they certainly use a lot of filters for this game, it's very difficult to tell what it is exactly. I never thought about the smoke/dust thing, to be honest.

…

Another game I found muddy at first was Mass Effect 2, because of the use of the fish eye camera, or whatever it's called, and the film grain filter which I turned off. But I must admit it was a feeling I had at first, then I realized it looked really fine.

If it's a fuzzy, changing feeling, then it may be your personal preferences at work. KZ2 has a soft look with heavy effects. I have read comments from "muddy" to "like a movie" since it launched. but I doubt GG is trying to recreate ME2:

Regarding KZ, maybe they try to recreate something similar to ME2, a heavy atmosphere in a different planet. ME2 is crisp, while KZ2 is blurry I could have a hard time playing it, and KZ3 has improved this, it's closer to ME2 than KZ2.

They are completely different in art and settings. KZ2 literally has a heavy atmosphere with dust, swirl and subtle flare. ME2 has very stylized graphics with clean lines. ME2 has people who speak/yell politely. e.g, Marinda, Jacob, Grunt, .... KZ2 has

Spoiler

Rico and other foul mouthed soldiers.

 

[Toaddio]

No one said it was an ugly game....

Actually, ugly was the exact word that was used on these forums.

 

[Shifty Geezer]

Interesting, so LBP2 and Gow3 aren't deferred. That LBP2 article that Shifty linked says they dropped in mlaa support in a day, so I can't image that is has to be tightly tied to geometry to work. Maybe both of those games aren't deferred but they already had normal buffer representations of the screen due to how they handled their lighting. Barring that how else could it be so quick to integrate mlaa into the games? The geometry side on ps3 games can get fairly complicated and customized, it just doesn't seem like a one day job to fit mlaa in there.

I think there has to be a scale on the MLAA system. To integrate as a drop in feature, it'll be a pure post effect running on the final backbuffer, but with more work you can integrate more evaluation option to fine-tune it. That's my guess. KZ3's MLAA comes across as less noisy than LBP2's. There must be mnore to it than the drop-in effect though. I mean, TB has been at ATG fora while now, and if since GOW3 released SCEE ATG said he did such a good job he can kick back and relax for a few months...well, that'd explain some of the slothful software progress from Sony!

 

[Laa-Yosh]

I think your guess is pretty reasonable
Knowing that both ATI and Nvidia are hard at work on their own implementation, I'd expect there to be a lot of more or less different post-processing AA solutions soon, with somewhat different capabilities.

What I'm not sure about is if this makes me happy or not. Image processing is still no replacement for proper sampling in my opinion, as much as I respect the work of all those engineers involved in the development of these technologies. As the asset detail keeps increasing it'll probably get a little more obvious in the next generation of hardware, so I'd still like to see some research on that other end.
It's the same I feel about the various SSAO implementations, they all produce noisy inaccurate stuff and then use various blur and dilate/erode filters and stuff to cover it up.Of course it's because reason is my background in CG where we don't try to get around these obstacles, but then we're not forced to do it in real time either... Still, I hope there's going to be enough performance in the future to allow for small gradual sacrifices to increase image quality.

 

[Shifty Geezer]

I think overall it bodes very well. Seems to me for the past decade or so, the main improvements in graphics came from graphics hardware power - more vertices and textures and samples. Things like MLAA and the various AO solutions show cleverer workarounds, with programmability leading to novel solutions. What we're seeing now are early-days solutions. Over the coming years with improved performance and programmability, people will find even better ways of doing things, building on these first steps. MLAA becomes SRAA become QZAA and BBQAA, pulling together the best of subsampling and approximation techniques to get improved IQ within the technical limits of realtime processing. I don't think 'proper' methods will be abandoned altogether because develoeprs are acutely aware of the shortcomings of these 'cheats' and in the pursuit of best results will mix and match to find the best compromise.

 

[patsu]

Looking at AA in isolation is important, but what may be more interesting is how the devs can rearrange the pipeline to do things differently on a new h/w architecture.

 

[BoardBonobo]

... and BBQAA, ...

Ummm, BBQ-AA. Is that image quality so good you could eat it?

 

[Laa-Yosh]

http://www.pixologic.com/interview/killzone3/1/

Reading it now; thought it'd be interesting for most people here

 

[Laa-Yosh]

Well, not much in terms of technical info, lots of very basic and generic ideas (as if they thought this article was for people outside the industry...), but at least a lot of pretty pictures too. Only thing worth mentioning is the triangle count, 10k for most characters, extra 2000 for those with special equipment and 4500 for human heads. Also, the two bad guys seem to appear in cinematics only, as their wireframes suggest higher poly counts.

Oh and there's quite some self-polish there, too

 

[Arwin]

Well, not much in terms of technical info, lots of very basic and generic ideas (as if they thought this article was for people outside the industry...), but at least a lot of pretty pictures too. Only thing worth mentioning is the triangle count, 10k for most characters, extra 2000 for those with special equipment and 4500 for human heads. Also, the two bad guys seem to appear in cinematics only, as their wireframes suggest higher poly counts.

Oh and there's quite some self-polish there, too

You can't be very hardcore at the same time as being proud of an 8.5/10 from IGN ... (though obviously he was more interested in the 9.5 for graphics)

 

[Laa-Yosh]

Their source artwork really is top notch, there can be no argument about that. I maintain my opinion that the engine imposes a bit too many limits on how it's translated to the game, but at least they'll get an immediate quality jump whenever the PS4 comes out.

Interesting to see Francisco Cortina's work there, he's been one of the guys behind the Final Fantasy movie's characters (and Animatrix, too), also did a lot of work on FF games.

 

[Lucid_Dreamer]

Well, not much in terms of technical info, lots of very basic and generic ideas (as if they thought this article was for people outside the industry...), but at least a lot of pretty pictures too. Only thing worth mentioning is the triangle count, 10k for most characters, extra 2000 for those with special equipment and 4500 for human heads. Also, the two bad guys seem to appear in cinematics only, as their wireframes suggest higher poly counts.

Oh and there's quite some self-polish there, too

I thought the following was interesting as well.

My first assignment for Killzone 2 was modeling the first-person arms and gloves of Sev, our main character. We used a different sculpt solution, which I?ve since converted into a ZTool. When I look at this sculpt now, I would say it conforms to about 60% of what we expect a Killzone 3 model to look like. In that sense, you might say we have improved the quality quite a bit. We also looked at how to improve the texture workflow. Balance is the key; our goal was to use more sculpted detail and less dirt and splotches. We also added a few nodes to the shading network to add micro-detail, which was especially useful for fabric.

 

[Lucid_Dreamer]

Killzone vs World meltdown garbage thread number 102412415918348976

C2 has tons of stuff done on the SPUs, atleast 5 SPUs are tasked with work. There are many workarounds/hacks to achieve more. It's in black and white in the shaders and configs. I dont even have to name RSX, if anyone can tap that to it's limit it's Crytek with their previous extensive experience with G7x architecture.

Tons of stuff on the SPUs? Like? And, how long did these tasks take? Optimized or just ported code?

Oh there's definitely interest beyond the usual conference proceeding or behind-the-scenes sharing (repi has certainly been quite enthusiastic as of late regarding FB2).

The problem, for starters I think, is gauging interest. At GDC or Siggraph etc, folks are expecting the cutting edge algorithms, not necessarily post-mortems (unless there were extraordinary circumstances the devs felt sharing). Afterall, there are only so many days and slots for these conferences.

So then there are sites like DF (and possibly B3D *ahem*) going after these sorts of interviews, and it's a problem of getting ahold of the right folks and finding out if there's interest within the dev team; a lot of questions span more than the scope of an individual's work on the dev team, so things can take time. Keep in mind we are also putting them on the spot, so they do have to be quite careful for PR approval.

Regarding "processes <-> end results" I think the best we could hope for are these debug captures beyond the usual descriptions (PR or not), but they are still only individual frames, enlightening as they are, compared to the scope of an entire game.


-----

That makes sense. Thanks for that bit of "behind the process" info.

Quite frankly, the revealed info should take precedence over PR-speak. Arguing and trying to match the two are just wastes of time when we can simply observe the data being shown.

Agreed, when they are different things. I think people maybe taking the wording wrong with the whole 3x polys. Then, again, maybe I am taking it wrong. To me, 3x polys is not per frame. I think GG was talking about the level itself. I think the whole 4+ million polys per frame thing was born from one member.

Debuginfo says otherwise. 1m rendered polys/frame for that scene in KZ3 and a quiet good gauge considering scene size. Thats some hard facts. Also in one of GG rendering presentations they said KZ2 peaked at 1.2m polys/frame. Interesting as ND said roughly the same in a presentation about U2 peaking at 1.2m rendered polygons per frame. All this can be found either in their released tech pdf/ppt files or interviews/presentation videos.

That's 1.3 million for KZ3, I believe. Didn't you say C2 was around 0.7 million? I remember seeing the debug screen go between .5 million and .73 million, during the video.

 

[Ruskie]

You always seem to think of worst...What ported code when they have been developing CE3 since 2007?I would assume they could know a thing or two about SPUs and GF7XXX hardware.
If you run SSAO and real time GI on PS3 at 1ms I would guess they take care of things rather carefully.

Poly count on Crysis 2 seems to go from 0.3 to 1 million for geometry and similar numbers for shadows.

 

[Lucid_Dreamer]

I honestly don't think I've ever seen a stranger, more far from the truth, assertion ever here at B3D.

Maybe you reside in a different Internet, but the one I visit praises the technical aspects of the 2 PS3 Killzone games far, far more than any other from this generation.

Maybe he means from the same few people that have a hard time putting their hands together for titles like KZ2 (well maybe not years after the fact), KZ3, etc.

 

[Laa-Yosh]

Crysis 2 poly counts might be subject to change if they can tune the hyper-aggressive LOD... It certainly is quite visible and annoying.

 

[Lucid_Dreamer]

You always seem to think of worst...What ported code when they have been developing CE3 since 2007?I would assume they could know a thing or two about SPUs and GF7XXX hardware.
If you run SSAO and real time GI on PS3 at 1ms I would guess they take care of things rather carefully.

Poly count on Crysis 2 seems to go from 0.3 to 1 million for geometry and similar numbers for shadows.

Ported like MLAA, in it's original form, was ported to the SPUs. It was 120ms. The was unoptimized code. It, eventually, got down to 20ms with improvement over the original ported code. Are you saying you have evidence of Crytek doing this sort of process to their SPU code? That's what I'm asking. It's a completely valid question. Why do you seem offended by it? Please bring forth your findings.

Edit: Just watched the debug video, again. The highest point was 730K and 365K for shadows. That's a total of almost 1.1 million.

 

[Scott_Arm]

Ported like MLAA, in it's original form, was ported to the SPUs. It was 120ms. The was unoptimized code. It, eventually, got down to 20ms with improvement over the original ported code. Are you saying you have evidence of Crytek doing this sort of process to their SPU code? That's what I'm asking. It's a completely valid question. Why do you seem offended by it? Please bring forth your findings.

I'm going to go out on a limb here and make an assumption that Crytek does their best to optimize all of the code that they run on the SPUs. Why would you assume they don't? That's what developers do.

 

[Cyan]

I started playing Killzone 3 it recently and I am extremely impressed!
I am blown away by the amount of extra performance they were able to squeeze out of the hardware.
The texture and environment variety is also second to none.
Each area is distinctively different than the other. From the buildings and destroyed structures, to the objects and the textures on each place.
the structure of the environments in K2 were often like rectangular blogs on parallel with corridors and routes connecting them.
But this one's environments is like Uncharted 2's with Killzone's art and additional variety. Also its crazy how many stuff are going on screen with all those light sources, bullet storming and effects without the slightest effect on framerate.
It also deserves lots of credit about its scale. The environments are often HUGE! And the textures on some of the characters appear to be high res and more detailed than what we got from some characters on the demo stage.
The AI continues to be amazing and the animations smooth and life like. There were a few instances where I shot a few Helghast standing on a structure who realistically fell and ended up hanging from the balcony below.
Truly amazing stuff.
The only lowlight so far were a few sound bugs here and there.
Why arent some of its aspects not praised as much from a technical point of view as other games?

I also visit another sites and the game is praised pretty much everywhere. I read, I don't participate, it's enough for me to write here but this has been my experience with reading other people's opinions on the game so far.

KZ3 improved A LOT of things over KZ2. For me the most reprehensible aspect of the game, judging by the screenshots, is that I don't know why, the game still looks muddy to me. it's odd that so many people either don't know that muddiness exist or ignore it for whatever reason.

Very very far from KZ2, not half as really annoying, but not as crisp as God of War or Halo Reach either.

The environments are larger than KZ2, the textures look better, the MLAA gives the environment a CG look sometimes and the smoke effects along with the sick lighting adds a feeling of large scale fun, which is awesome, however, I am a bit biased because to me KZ2-3 are quite possibly one of the most pretentious games out there and I take almost anything about them with a massive grain of salt.

It was meant to be the supposedly Halo killer since Killzone 1 days -HUH??-, it was supposed to be the 2005 CGI -which, I agree with others, the total level of fail in the comments of the DF article was just beyond words, it was that awful-, etc etc.

If DF journalists lie one more time like that, I will not be gracing that part of the Eurogamer's page with my heartless and frigid presence ever again.

 

[Ruskie]

Ported like MLAA, in it's original form, was ported to the SPUs. It was 120ms. The was unoptimized code. It, eventually, got down to 20ms with improvement over the original ported code. Are you saying you have evidence of Crytek doing this sort of process to their SPU code? That's what I'm asking. It's a completely valid question. Why do you seem offended by it? Please bring forth your findings.

Don't worry,I'm not offended,there is nothing to be offended by.
I don't have the evidence but I'm just looking at what they are running on consoles and I kinda assumed that they had to be pretty good to achieve that on these consoles.Beside iffy AA and lower res there is absolutely nothing that is missing.

Thats why I gave you their numbers on SSAO and SSGI which are pretty damn fast.This numbers are quite good,in line with NDs.Plus you have first real time global illumination running on any hardware,also pretty fast on consoles.Both 1 ms.

Another thing is that they have engine to sell.Their engine is suppose to work very best on both platforms with all fancy effects turned on so I guess It wouldn't make sense for them not to do their best on PS3.