Digital Foundry Article Technical Discussion Archive [2015]

[Arwin]

This sounds like semantics... Irrespective of all this is that inconsistent behavior leads to different times between controller input and visual feedback. And unless you have any kind of G-sync capability, which the consoles don't currently, physical refresh rates are going to have to be multiples of 60 (or an integer divide). Draw 70 frames or 50 or 25, and in all these cases you end up with either judder or tearing or maybe even both in extreme cases.

 

[Deleted member 11852]

Digital Foundry: Performance Analysis: Dying Light

 

[Arwin]

Thanks Cjail for updating on www.techingames.net so fast I've added a few links and verified so you get the rep ...

 

[psorcerer]

Draw 70 frames or 50 or 25, and in all these cases you end up with either judder or tearing or maybe even both in extreme cases.

That's just makes judder worse, but the same calculations apply.
I would say that if we get a game where 50% of frames are 16ms and 50% are 33ms
The one that goes like this: 16,33,16,33,16,33,16,33,....
Will be much worse than the one that goes like this: 16,16,16,33,33,33,16,16,....

 

[London Geezer]

So what's the deal with AF on PS4? This is 2015 for chrissake.

 

[shredenvain]

So what's the deal with AF on PS4? This is 2015 for chrissake.

It is probably a bandwidth issue at least for Dying Light.

 

[Scott_Arm]

It is probably a bandwidth issue at least for Dying Light.

Does that make sense though? Sounds like the Xbox One version has anisotropic filtering.

 

[London Geezer]

The XO version has AF but runs at a lower res. This, according to DF, results in some scenes where the XO actually shows more detail.

What's the point of sticking to a full HD buffer if all that texture detail is lost at steep angles, which is most of the time?

 

[shredenvain]

Does that make sense though? Sounds like the Xbox One version has anisotropic filtering.

I'm by no means an expert, but it is the only thing that makes since to me. AF requires bandwidth to sample texture points. With The Xbox One version having AF as well as a slightly lower resolution maybe the devs found they had spare bandwidth from the main memory to implement AF. It can't be a TMU issue because the ps4 GPU has a considerable advantage in that department.

 

[London Geezer]

I'm by no means an expert, but it is the only thing that makes since to me. AF requires bandwidth to sample texture points. With The Xbox One version having AF as well as a slightly lower resolution maybe the devs found they had spare bandwidth from the main memory to implement AF. It can't be a TMU issue because the ps4 GPU has a considerable advantage in that department.

The question here is that if both machines only have resources for AF when the resolution is 20% less than 1080p (or whatever number it is), how does it happen that one will go for lower res with AF, while the other goes for 1080p without AF. It's not the first time this happens.

 

[shredenvain]

The question here is that if both machines only have resources for AF when the resolution is 20% less than 1080p (or whatever number it is), how does it happen that one will go for lower res with AF, while the other goes for 1080p without AF. It's not the first time this happens.

I know it isn't the first time it has happened. If the devs shrunk the res down to fit the entire frame buffer into the esram then it could lead to a slight bandwidth advantage compared to the other system having one pool of memory that is shared between everything. This could explain why some titles have AF on the Xbox one and don't on the Ps4. Both systems GPUs are based on the same architecture with the Ps4 having an ALU advantage as well as a TMU advantage. Excess ALU doesn't usually help you with implementing AF in most cases but extra bandwidth will.

 

[Rangers]

I'm by no means an expert, but it is the only thing that makes since to me. AF requires bandwidth to sample texture points. With The Xbox One version having AF as well as a slightly lower resolution maybe the devs found they had spare bandwidth from the main memory to implement AF. It can't be a TMU issue because the ps4 GPU has a considerable advantage in that department.

Perhaps it's some advantage in the dedicated ESRAM causing less CPU bandwidth contention on X1? The lower resolution is not enough to account for the X1's much lower system bandwidth by itself, as the X1's main bandwidth is much more than 20% lower.

What other examples are there of games with AF on X1 and less on PS4? Out of curiosity.

 

[Laa-Yosh]

Isn't disabling AF still a super simple optimization technique that can boost your framerate by 2-10% at the last minute?

 

[Globalisateur]

Maybe 90% of past games that lacked AF on PS4 were 100% confirmed to be using Unreal Engine 3. It's maybe also the case in this game. What engine did Techland use to make this game?

I don't think the lack of bandwidth (or GPU ressources) would explain it because XB1 seems to suffer by more technical problems running the game compared to the solid 30fps seen on the PS4, even at a reduced resolution (note that it's a not a short drop but sustained one and the PS4 is locked at 30fps during the whole scene).

AF in this game is apparently only used on ground textures, vertical walls don't use AF too on XB1. So the general use of AF is really limited on XB1. Shouldn't make a big difference to activate AF on PS4 too particularly if the game already runs better than on XB1.

Also we can notice here that the PS4 texture is even slightly sharper than the XB1 texture and it's an oblique texture! (better seen on a native screen). Apparently DF didn't notice it. Maybe they are not aware of the phenomenon but it's normal and usual, because of the native resolution everything will naturally look a bit sharper on PS4, particularly on a native screen, well except some ground textures in the background.

 

[Rangers]

It definitely smells like an Unreal engine game to me, visually, but I have no idea.

Googling Unreal Engine Dying Light brought me only this link http://victordima.net/2014/10/03/new-gameplay-dying-light-released-techland/ saying it's UE4. But that site seems very small and I dont see literally any other mentions so it seems dubious for now. In fact most of the hits on that search come back for Daylight not Dying Light, further casting aspersions.

 

[shredenvain]

I never said a lack of GPU resources. To me the engine wouldn't be the problem.
It has to either boil down to memory bandwidth or something in the Ps4's SDK or graphics driver causing some sort of issue with certain engines. Sure the Ps4 more often than not has less problem with bandwidth because of it's fast memory, but it is plausible that the way the Xbox One's memory is set up can provide advantages from time to time in certain situations. After all they have very comparable theoretic memory bandwidth available and shrinking the frame buffer could allow them room in main memory.

 

[shredenvain]

It isn't the unreal engine. It is there own in house tech

 

[DrJay24]

Isn't disabling AF still a super simple optimization technique that can boost your framerate by 2-10% at the last minute?

Then they should do it on the XB1.

 

[London Geezer]

From the DF ANALysis...

Running on Chrome Engine 6, Dying Light represents Techland's first efforts on current-gen hardware

 

[Globalisateur]

From the DF ANALysis...

The only logical explanation is that they reproduced the same 'AF bug' than Unreal Engine 3 in their own Chrome engine 6. In my knowledge every Unreal Engine 3 games lack AF on PS4 (even the most graphically simple game), without exception.

If games like TLOU can run at 1080p solid 60fps with AF 16x enabled on many textures or Far Cry 4 with its massive open world at 1080p and solid 30fps, great temporal AA, Parallax mapping (a first in console?), god rays and decent AF, there are no excuses in any others games to not using some basic level of AF as seen in the XB1 version, maybe 2x or 4x (max) here.

Anyway we'll see what they say about this, maybe they'll patch it later.