I mean why have they gone for a lower resolution than what would be rendered for checkerboard?
4147200 total pixels vs 3686400
-
Hey Guest. Check out your NeoGAF Wrapped 2025 results here!
DF: Mafia 3 PS4 (Pro) patch 1.05
- Thread starter Caayn
- Start date
I mean why have they gone for a lower resolution than what would be rendered for checkerboard?
4147200 total pixels vs 3686400
Liabe Brave
Member
You are wrong about what CBR does, maybe my thread about it will help. The actual comparison is 8294400 pixels (half with a quicker process) vs. 3686400.
But I don't see how the question is pertinent even if you weren't wrong. Why does any game fail to render at a higher resolution? Because the dev decided to put processing power to some other use.
PeterLegend
Member
Hmmm it definitely seems like resolutions above 1080p take quite a bit of resources. I think regular PS4 resolution is 1080p but PS4 Pro which has 2.3x the GPU power can only do 1440p while keeping the same performance.
Good patch either way, maybe a future one to optimize performance a bit more...
Good patch either way, maybe a future one to optimize performance a bit more...
That's the pixel count for native 4k rendering in a single frame, which CBR is not.
CBR renders half of that in a single frame and combines it with temporal data from a previous frame.
3686400px is 1440p
4147200px is CBR
8294400px is native 4K
adamsapple
Or is it just one of Asha's balls in my throat?
We're still getting a 77% increase in pixel count at 1440p native, over 1080p.
So, no. The Pro isn't struggling, the developer is just allocating resources differently. It's like asking why Uncharted 4 didn't user CB when other similar games (RoTR) do. If you're wondering why they chose what they did, you're better off asking the developers directly over their twitter than here.
Liabe Brave
Member
This is the wrong way to look at things. Every dev differs; every game differs; the effort given to each Pro patch differs. Not all of them can be expected to utilize the hardware to the very utmost...especially multiplatform games.
That said, the 1440p is ~1.8 times as many pixels, so that leaves an "excess" of 50% improvement over standard PS4. This extra GPU power is being used to improve motion blur, depth of field effects, some shadows, and possibly more.
You misunderstand the method. CBR 2160c does not render only 4.14mp, it renders 8.29mp. What you mean to say is that only half of these pixels are rendered in the standard manner. However, the other half are not simply repeated from a previous frame (no cost), nor are they upscaled (trivial cost). They are rendered in a novel manner, incorporating both prior frame knowledge and linking to the current frame's polygonal positions, as well as confidence data to improve accuracy, and sometimes further corrections such as sawtooth artifact culling. There is a cost involved with all this. You're right the total is not as high as standard rendering of 8.29mp, but neither is it the same as standard rendering of 4.14mp. The bare statement "only 4.14mp are rendered with 4K CBR" is wrong.
Liabe Brave
Member
No, it isn't. CBR renders the same number of pixels as a native buffer of the same size. The difference is that half the pixels are rendered in a different--quicker but less accurate--manner.
Mafia III uses TAA, which it seems can play badly with checkerboard, causing extra artifacts. That could be the reason for avoiding it.
Looking at how reconstruction works in Siege..techniques like checkerboard could be around 30% faster than rendering the same numbers of pixels with a native buffer.
Pretty sure I won't be wrong if I said 1800P CBR is still more expensive than 1440P.
vpance
Gold Member
I know, that's what I meant. 1800p CBR is effectively less costly to render than 1440p native.
If true then that's slightly disappointing.
I suppose there could also be certain engine and rendering pipelines thungmees that means it's not easily implemented in certain games.
I understand why this is, because it's effectively using a temporal technique on top of another temporal technique. But why does the TAA in Siege (which uses MSAA reconstruction with some temporal data, effectively making it similar to CBR) work so well ? Because the artifacts are barely visible in motion and reconstructed 1080P with TAA looks pretty much the same as native 1080P with TAA even during gameplay (more or less ofc). It's only when you disable TAA the difference become obvious with the typical artifacts associated with reconstruction. Although the TAA in Siege does cause a really heavy blur.
So does it look better or worse than 1440p?
FFXV is still unfixed...Pretty sure there are others too.
Stevey
Member
Game can look *very* nice on a Pro sometimes, despite looking a bit hazy.Here's a few taken from my own Pro with the latest patch. 4K down sampled to 1080p via PS4s capture feature.
That's 4K downsampled to 1080?
That looks terrible, the aliasing on the car looks awful.
adamsapple
Or is it just one of Asha's balls in my throat?
Yep, 4K down sampled to 1080p, it looks like bad banding from the rain effect, not aliasing. The game doesn't exhibit much AA.
Stevey
Member
Like the whole car looks a blurry mess :/
Is that not a reflection in the bumper? I understand what you're saying but that's what it looks like to me. It could be aliasing but it just doesn't seem to be constant across the entire image.
vpance
Gold Member
Better, certainly at least from living room TV distances. Sony themselves recommend devs to go for at least 1800p CBR instead of straight 1440p.
It's not 4K, it's 1440P downsampling to 1080P.
Also that aliasing in the rear bumper seems to be caused from either reflection, DoF or low Res alpha rather than jagged polygons. See how the portion of car directly above (the horizontal edges of the car boot) that isn't aliased at all.
Liabe Brave
Member
No, it probably isn't. As nOoblet16 says, extrapolating from the R6S implementation already gives us a render time greater than 1440p. Now, this is only a single game so can't be a perfect guide for every other title. But at the same time, CBR as implemented on Pro has further refinements not included in Ubisoft Montreal's version, which presumably increase cost (even if just a little).
But even if such estimates are wrong, that's beside the point. Your argument is based on the implicit assumption that one particular approach is best. The fact of the matter is that devs have chosen an array of different targets, checkerboard or not, for their games on Pro. I think the conclusion that many of them are incompetent, ineffectual, or idiotic is perverse. All the facts I see show them to be enthusiastic trained professionals.
I'm sure the devil is in the details, but I simply don't have the knowledge level to answer this.
adamsapple
Or is it just one of Asha's balls in my throat?
Probably this:
but as you can see, it's just mostly in that pic, the other's look fine. Game does have a very 'hazy' look overall though, definitely not the sharpest looking game out there.
Liabe Brave
Member
That's from the reflections, which are a very low resolution in Mafia III. I find the blockiness less noticeable in motion, but it's probably the most disappointing setting in the game.
But it's not from downscaling 1440p. Notice that there's no such aliasing on the license plate right above the bumper.
That's a depth of field effect; look at the car partially shown at the left edge of the screen. So these objects aren't supposed to be in sharp focus. Notice that background objects which are further away are sharper--that'd be impossible if the softness was a product of overall resolution.
It's 1400p downsampled to 1080p
Liabe Brave
Member
Several of the members of DF do have a very strong personal preference for performance over graphical improvements, generally greater than that felt by most players. This may lead to language that others may feel sounds like hyperbole, but it's their honest opinion and at least they flag it as such. My biggest complaint, instead, is how they keep using the term "adaptive v-sync". I think this is at best needlessly obscure, and at worst simply wrong.
The term itself came from Nvidia, and is an instance of the GPU manufacturer jargonizing an existing technique to make it seem more proprietary than it is. What both they and Digital Foundry are actually talking about is games that never go above their framerate target (30fps or 60fps). But they can fall below their framerate target, and when they do there's screen tearing. I think it's obvious that a better term for this was already around: "framerate cap". It's transparently clear what that means even to a layman, unlike the sciencesque "adaptive v-sync".
More importantly, what does v-sync mean? Well, it means there's no screen tearing. But "adaptive v-sync" is used solely about games that do have screen tearing! By capping framerate it does remove potential tearing for games that run faster than their target. But below the target it does absolutely nothing. It's "adaptive" in the sense that it switches itself off whenever there's slowdown. By my lights, that at best deserves the name "partial v-sync", or maybe "one-way v-sync".
But really, those are still obscurantist. What does "adaptive v-sync" do? Stops framerate from going above a certain value, and nothing else.
So call it a framerate cap.
I thought games with TAA tended to do the best with checkerboard rendering. CoD IW handles it beautifully and upcoming stuff like Horizon and Days Gone have impressive checkerboard implementations and TAA as well.
I would invite you to load up a game which, using whatever settings are appropriate to your hardware, hovers around 30 fps and then use RTSS to cap the framerate to 30 and turn off v-sync altogether.
Then, disable the cap in RTSS and force adaptive v-sync (half refresh).
I believe you will find the differences illuminating.
Liabe Brave
Member
I based that statement on stuff said by KOCMOHABT and Durante in my CBR thread. Perhaps the problem is with implementations? ("naive" TAA versus TXAA versus SMAA with temporal addition, etc.)
thelastword
Banned
Why would anyone one want standard PS4 mode on a PRO? You might as well stick with vanilla PS4 if that's what you want. So you're telling me you would rather go with lows of 27fps with tearing on base PS4, a lower resolution, worse MB, worse shadows because you have a problem with 26fp vsynced in one fight scene?
Technically the pro patch is better and the advancements given there is why they brought a pro in the first place. Hell, they can also downsample if they have 1080p TV's for better IQ and 4k owners get a sharper image over vanilla PS4 mode which is what they want...
In any case, there hardly remains a game which performs worse on PRO over OG PS4.....
The game had issues at launch, they attempted and fixed many of those issues, it doesn't a solid 30fp sis not possible on PRO, it just means the game needs a bit more optimization even now. Still, I'll take the tech in this, the current performance, the resolution upgrades over other openworld games like Witcher 3, Just Cause and Fallout 4 on consoles any day...
It's sad to imagine, that people would rather have the Pro version tear to maintain an FPS lead over the base PS4 in every scenario, as opposed to playing the game fully vsynced with graphical enhancements and better IQ. I mean that 26fps would just trouble their minds whilst playing......"It's worse than base PS4 man"... these guys are probably at 27fps in that scene ;}....I cannn't playyyyyyy thissssss.................SMH!
Perhaps as the dev works more on this and improves their engine they may bring even further performance upticks to the PRO in the future, especially during shotgun battles. I think they were strapped for time here, but the end product is no below-low trife, they have some nice tech and the perf is not bad at all on average, unlike some bigger budget open world games.....I like how quickly they've patched things and are still giving support. I expect more from them tbh...
Liabe Brave
Member
What do you believe is the consistent difference? Because ignoring what my experience might be--regardless of any single personal experience--all the discussion I've seen indicates there is no consistent difference. That is, whether RivaTuner or Nvidia's adaptive v-sync (or an in-game cap option) gives better results varies, in no predictable manner. It seems dependent on multiple variables, specific to certain games and the intricacies of their executables.
If results differ, then any contrast between the methods must rely solely on the micro-level details of how they actually perform their task. This would preserve the distinction of them as separate approaches...but it wouldn't affect my call for unified descriptive nomenclature. This is because Digital Foundry are working solely from the end result, not from reverse engineering of the individual game engines. Absent input from the developers, their analysis can only preserve the phenomena, leaving the path to that destination underdetermined.
In this context, "framerate cap" as a general term is much more fit for purpose.
SmoothRunningGun
Member
I'd prefer framerate improvements.
FarFromAnyGame
Banned
You can't get over that 1 frame?
adamsapple
Or is it just one of Asha's balls in my throat?
After more playing and as multiple users have said, it's definitely an issue with the reflections on the car. Look at this snapshot, everything else looks sharp but look at the reflections on the bumper and the sides, the reflection has a very low res jagged look.
Game seems to be using very low resolution for the car reflections, same for the rear view mirror.
Otherwise, the game has good AA coverage.
