• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

PS4's AF issue we need answers!

nib95

Banned
Apologies. Here are some still shots.
screenshot-original8y8a83.png

screenshot-original15xhy1r.png


It has some form of AF, not an excellent one but it's totally workable for a game with typically high speed, high motion blur segments.

Honestly, the AF is sufficient. You don't notice the lack of it as much in racing games, especially when you start moving and all the motion blur etc kicks in. The AF in FH2 appears to be dependant on view and scene or texture, even in the shots you posted it looks slightly different in each shots, unless I'm gauging it wrong. In this example, it looks lower.

screenshot-originalj8rrn.png
 

jackdoe

Member
I can give you a reason for Theif at least. Reasons for individual games may be different. It ran at a higher res with better performance on the PS4 but the main reason is probably because unlike the XB1 version of theif which was missing it entirely it had Parallax Occlusion Mapping too. AF gets expensive with it enabled. Contrary to popular opinion, it's not free at all:

Review of Displacement Mapping Techniques and Optimization

btu.se/fou/cuppsats.nsf/all/9c1560496a915078c1257a58005115a0/%24file/BTH2012Hrkalovic.pdf
Honestly, for the cases of Thief, Evolve, and Dying Light, all titles that did more on the PS4 (higher resolution, additional effects, etc), I could definitely believe that the reason for reduced or lack of AF was due to performance.

It's the titles that look and perform exactly the same across both consoles that are the true head scratchers.
 

thelastword

Banned
I was watching DSP play GTA Online last night (PS4 version) and the AF was loading all over the place, even on buildings and stuff, not just the ground.
Does this happen on the X1 version as well?, it was really distracting.
By the way, GTA on next gen looks really good besides the AF problems, i was surprised how good it looked at times.
You're probably referring to lod (level of detail), af is either on or off, either you have it or you don't. The only other metric involved is the level of AF (4x,8x,16x) etc..

At least this tells us what we can think of nx gamer now.
That was not what NXgamer said, he never said he was right, he was only theorizing just like us. At least he gave a somewhat rational explanation as opposed to some of the more loony ones I've been reading.

PCs have dedicated vram. The actual bandwidth isn't hurt by system memory tasks.

The ps4 everything , cpu gpu, recording video in the background, os running in the back ground all go through same memory bus. No matter how small.

Think of it like a hdd. 1000 1k files will take longer to transfer that one 1000k file because of read write constraints.

Like a hdd gddr can not read and write at the same time.
What does that have to do with AF lacking on PS4 version of games?

Has anyone asked Insomniac, Turn-10, Playground Games, 343i, Saber, Ruffian, Certain Affinity or Capcom Vancouver (lol) why their games are lacking anisotropic filtering?
Ha, good point. There's no doubt that we have persons who are concerned (who really aren't). Still, I would like to know what the issue is here, especially on games which should have 16xAF standard like strider and this game.

Something must be done to prevent it's reoccurence because those titles would look much better for it. We can't be going to year 2 and devs are still halfassing what is a simple implementation, if Sony have to be more pro-active, then so be it. It looks silly when a weaker gpu has a simple feature like this implemented and it's not in the PS4 version.

The most logical answer would be bandwidth.
Xbone uses less bandwidth due to lower resolution in most games and has higher theoretical bandwidth if ESRAM and DDR3 is used right.
Do you even think of what you say before you type? This game is 1080p on both platforms, how is it that the lower resolution affords them ohhhhhh! (all that bandwidth necessary to implement AF). AF doesn't not require much bandwidth at all, like at all and the PS4 has more bandwidth than the XBONE.

It's possible, but as you point out it doesn't explain every situation we've seen. And there are several games that would seem to defy that idea. In a case like DMC it especially doesn't make sense, seeing as it's essentially using last gen assets at a higher resolution and framerate. The Last of Us had no problem hitting 60 at 1080p with something like 16xAF with better character models. Tomb Raider:DE likewise had decent AF with improved assets and effects in addition to resolution and framerate. We haven't gotten a concrete reason/statement from any of the devs of the specific games with the problem, so it's really hard to do anything but speculate and scratch your head.
Money post aka great post.

Anybody who believes that this is a hardware issue are the same persons buying into the "power of the cloud spiel" and "Dx12 will give a 20-50% boost pipedream". If anything, due to the stronger gpu, the PS4 should have a better AF implementation everytime against the xbone. It's clear that something may be up, but it's either the api or the devs.

I think you are taking this abit to extreme, from watching the video he does not say any of the above?!?
Dictator will always try to discredit Nxgamer at all costs. Out of all the theories I've heard, his is one of the most sound, still he never claimed that he was factually correct. I think some people just have an axe to grind. Nxgamer's theory is second on my list of believability. The first is that the dev simply did not implement AF, Sony should make it mandatory. Even bad ports like Re-R2 and Xenoverse have AF.


EDIT: That last Forza Horizon 2 pic has no AF whatsoever, seems like it toggles on and off.
 
Dictator will always try to discredit Nxgamer at all costs. Out of all the theories I've heard, his is one of the most sound, still he never claimed that he was factually correct. I think some people just have an axe to grind. Nxgamer's theory is second on my list of believability. The first is that the dev simply did not implement AF, Sony should make it mandatory. Even bad ports like Re-R2 and Xenoverse have AF.

...

Really?
 

thelastword

Banned
...

Really?
Yes, I think you always take a stab at him. He really has given very good information and in many cases discovered things DF never did. I think that puts him in great light. Even when he speculates, he explains his viewpoint, unlike DF editors who's reasoning goes like (it's most likely cpu bound on the PS4) and end it there. Nx gave a solid argument, yet he was adamant that he was only giving his take based on what is now a talking point and an issue across many games.
 
http://www.tweakguides.com/Graphics_1.html

GGDSG_22.jpg



The performance impact will differ, however more recent graphics cards now use highly optimized methods of selective AF, and can do even 8x or 16x AF with minimal impact on FPS, even at high resolutions

Article was written in 2012.

I've read a lot about the whole process, and really can't get a clear idea of why AF wouldn't be kicked up to the highest degree from the outset, and then the finalized engine put into place around it. Considering it is not a huge hit on the performance and it makes such a difference, you'd think that would be one of the main points of emphasis. But I mean ... it almost feels like they are implementing the texture filtering at the very end of the whole process, and if they run into performance issues they just lower the sampling until they hold to the more steady framerate.
 

DSN2K

Member
it really is bizarre considering the impact AF has on Modern GPU's is pretty much nil. Definitely not a hardware issue, I put it down to developers choice. AF is not something that requires optimizing either, its a case of level choice and type and that is a GPU thing completely.
 

Pimpbaa

Member
Honestly, for the cases of Thief, Evolve, and Dying Light, all titles that did more on the PS4 (higher resolution, additional effects, etc), I could definitely believe that the reason for reduced or lack of AF was due to performance.

There is no excuse for no AF in Dying Light, it is doing nothing particularly fancy in regards to graphics. The game doesnt even do trilinear filtering on the ps4 for fuck sakes.
 

RedAssedApe

Banned
...not sure if serious....

I'm going with not, but it's really hard to tell sometimes.

haha definitely not serious. tbh if i didn't read that it was missing i'd probably not even notice unless i was looking at stills of the game. especially when the games are in motion and/or there is a lot of action going on the screen like in something like DMC. still thought Dying Light looked great despite all the DF comparisons to PC and apparent lack of AF on PS4.

still curious why the omission though. seems like it should have the horse power to do it?
 

Shin-Ra

Junior Member
Honestly, the AF is sufficient. You don't notice the lack of it as much in racing games, especially when you start moving and all the motion blur etc kicks in. The AF in FH2 appears to be dependant on view and scene or texture, even in the shots you posted it looks slightly different in each shots, unless I'm gauging it wrong. In this example, it looks lower.

screenshot-originalj8rrn.png
If it's anything like Burnout Paradise you spend plenty of time crawling about at low speeds, starting events or just messing about.
 

c0de

Member
http://www.tweakguides.com/Graphics_1.html

GGDSG_22.jpg






Article was written in 2012.

I've read a lot about the whole process, and really can't get a clear idea of why AF wouldn't be kicked up to the highest degree from the outset, and then the finalized engine put into place around it. Considering it is not a huge hit on the performance and it makes such a difference, you'd think that would be one of the main points of emphasis. But I mean ... it almost feels like they are implementing the texture filtering at the very end of the whole process, and if they run into performance issues they just lower the sampling until they hold to the more steady framerate.

So far we only know that it can be heavy on bandwidth. But PC GPUs have exclusive "rights" to access the RAM on a card while PS4 has to share it with the CPU.
If we believe
or not but we can be sure that the theoretical maximum of the spec (176GB/s) is most likely not achievable if both CPU and GPU access the main RAM (which will always be the case, of course).
 

KKRT00

Member
Do you even think of what you say before you type? This game is 1080p on both platforms, how is it that the lower resolution affords them ohhhhhh! (all that bandwidth necessary to implement AF). AF doesn't not require much bandwidth at all, like at all and the PS4 has more bandwidth than the XBONE.
What game? I didnt even mention any game, i was talking generally.
AF requires bandwidth.
PS4 doesnt have more bandwidth, it has much easier usable bandwidth, but less of it.
 

RoboPlato

I'd be in the dick
What game? I didnt even mention any game, i was talking generally.
AF requires bandwidth.
PS4 doesnt have more bandwidth, it has much easier usable bandwidth, but less of it.
Where is the "XBO has higher bandwidth" coming from? The highest I've heard quoted is 133 gb/s while PS4 has 176.
 

CGwizz

Member
its because devs dont care about this shit, so far it hapened in bad looking games , i dont see a big problem hhere.
 

c0de

Member
Where is the "XBO has higher bandwidth" coming from? The highest I've heard quoted is 133 gb/s while PS4 has 176.

Again, this is the theoretical maximum which you won't see "in real life" as it is shared with the CPU which will always steal bandwidth from the GPU. Xbone has exclusive access to the esram, not to mention that access to esram should have way less wait-states.
 

iceatcs

Junior Member
What I understand that AA and AF are coming from outside (hardware) API, not part of game engines.
That's why we have all same number of type filters in every 3D games.

It mean some general engine or coding might not full work or ready with PS4 API.

It doesn't mean the hardware/bandwidth lacking, just API. Cos it is confirmed PS4 is enough for 1080p AF on any games, but not there likewise API problem (aka non-directX and OpenGL).
 

leeh

Member
Again, this is the theoretical maximum which you won't see as it is shared with the CPU which will always steal bandwidth. Xbone has exclusive access to the esram, not to mention that access to esram should have way less wait-states.
This. People keep on throwing the 176GBs number around but don't realise that the bandwidth drops the more the CPU needs it. With all the background tasks on the next-gen consoles, its no surprise that we don't have enough bandwidth at 1080p for 16AF.
 

Jux

Member
Again, this is the theoretical maximum which you won't see as it is shared with the CPU which will always steal bandwidth. Xbone has exclusive access to the esram, not to mention that access to esram should have way less wait-states.

ESRAM won't have any impact on AF. ESRAM is only 32MB so most (if not all, we are talking of hundreds of MB of texture memory here) material textures will sit in DD3.
 

dr_rus

Member
This. People keep on throwing the 176GBs number around but don't realise that the bandwidth drops the more the CPU needs it. With all the background tasks on the next-gen consoles, its no surprise that we don't have enough bandwidth at 1080p for 16AF.

No amount of "background tasks" will ever push PS4's GPU bandwidth below the numbers available on PC low end GCN chips - which are perfectly fine at providing nearly free 16x AF in all titles. AF bandwidth hit is very low on modern GPUs. AF is an operation which is almost completely cached on chip today. And please remember that XBO has less bandwidth than PS4.

Can we please stop with this bandwidth argument? Whatever the reason for PS4's AF issues is it is certainly not memory bandwidth.
 
No amount of "background tasks" will ever push PS4's GPU bandwidth below the numbers available on PC low end GCN chips - which are perfectly fine at providing nearly free 16x AF in all titles. AF bandwidth hit is very low on modern GPUs. AF is an operation which is almost completely cached on chip today. And please remember that XBO has less bandwidth than PS4.

I guess that is the part of all this that is the core. Even chips with far less bandwidth in 2010 or 2011 were capable of nearly free AF... so why are people skimping on it given its incredible impact on texture clarity?
 

Jux

Member
No amount of "background tasks" will ever push PS4's GPU bandwidth below the numbers available on PC low end GCN chips - which are perfectly fine at providing nearly free 16x AF in all titles. AF bandwidth hit is very low on modern GPUs. AF is an operation which is almost completely cached on chip today. And please remember that XBO has less bandwidth than PS4.

Can we please stop with this bandwidth argument? Whatever the reason for PS4's AF issues is it is certainly not memory bandwidth.

There are still latency issues in the shaders though. The smaller the latency, the easier it will be hidden by other ALU operations while waiting for the result to be available.
 

RoboPlato

I'd be in the dick
Again, this is the theoretical maximum which you won't see "in real life" as it is shared with the CPU which will always steal bandwidth from the GPU. Xbone has exclusive access to the esram, not to mention that access to esram should have way less wait-states.
Right but the CPU can only take up to 20. That still leaves 156 for the GPU, still higher than the XBO eSRAM.
 
There are still latency issues in the shaders though. The smaller the latency, the easier it will be hidden by other ALU operations while waiting for the result to be available.
Are these latency issues you speak of enough of a hinderance / deficit in performance that makes having 1x or at most 4x AF justifiable?
Right but the CPU can only take up to 20. That still leaves 156 for the GPU, still higher than the XBO eSRAM.

Which even then is also quite a bit more than a lot older GPUs who dont care at all about AF performance costs. (although there is that cahrt floating around about 20 gb/s throuput on the CPU limiting the GPU to about 120 gb/s for some reason).
 

KKRT00

Member
Where is the "XBO has higher bandwidth" coming from? The highest I've heard quoted is 133 gb/s while PS4 has 176.

PS4 has 176gb/s for everything where Xbone has DDR3 has 68gb/s and ESRAM 204gb/s theoretically. Its impossible in normal workloads to use 204gb/s from esram, but You can easily exceed 176gb/s on both DDR3 and ESRAM with proper bandwidth management.
 

TAJ

Darkness cannot drive out darkness; only light can do that. Hate cannot drive out hate; only love can do that.
There is no excuse for no AF in Dying Light, it is doing nothing particularly fancy in regards to graphics. The game doesnt even do trilinear filtering on the ps4 for fuck sakes.

Heh. I've only seen Youtube video of it, but it was obvious even with all the compression. I mentioned it in the OP.
 

R_Deckard

Member
ESRAM won't have any impact on AF. ESRAM is only 32MB so most (if not all, we are talking of hundreds of MB of texture memory here) material textures will sit in DD3.

Just going to say this, Esram is only for Render targets and at 1080 even these can be split into DDR3 which will take more bandwidth, real-world PS4 still has more bandwidth than X1 just not as much as many think.

The issue is certainly a software one and if the X1 version of DmC has full 8-16x AF then this proves the fact that be something consistent with the PS4 API/Engine connect.
 

c0de

Member
Right but the CPU can only take up to 20. That still leaves 156 for the GPU, still higher than the XBO eSRAM.

Did you look at the picture I posted above? You can't just subtract what the CPU can get and call it a day. There are constantly reads and writes happening, caches involved, waiting for data, huge chunks getting written while you wait for smaller data and so on. Again, it is a theoretical maximum and nothing what you can expect in real life. It's like the 1.8 TFlops. Theoretically the hardware has the capability but you won't see a game managing to execute so many TFlops ever in a game.
 

dr_rus

Member
There are still latency issues in the shaders though. The smaller the latency, the easier it will be hidden by other ALU operations while waiting for the result to be available.

What's that has to do with AF? GPUs have dedicated texture caches specifically to handle stuff like AF in a way which won't interfere with shading. Shader fetches happen via their own L1 caches. And let me repeat this again - low end GCN GPUs are providing nearly free 16x AF while running shaders which are kinda close to what PS4/XBO are running.

If you're saying that AF isn't free and will impact memory access latency - sure, it'll cost you some 10% or something. But the issue here is that PS3 and XBO have AF while PS4 doesn't. And that's not something which can be explained by the lack of power on PS4 as both PS3 and XBO are clearly less powerful than PS4.
 

c0de

Member
What's that has to do with AF? GPUs have dedicated texture caches specifically to handle stuff like AF in a way which won't interfere with shading. Shader fetches happen via their own L1 caches. And let me repeat this again - low end GCN GPUs are providing nearly free 16x AF while running shaders which are kinda close to what PS4/XBO are running.

If you're saying that AF isn't free and will impact memory access latency - sure, it'll cost you some 10% or something. But the issue here is that PS3 and XBO have AF while PS4 doesn't. And that's not something which can be explained by the lack of power on PS4 as both PS3 and XBO are clearly less powerful than PS4.

But aren't we talking about bandwidth here? Nobody is denying PS4 is the most powerful hardware in the console business but memory and bandwidth don't provide any power.
 

dr_rus

Member
But aren't we talking about bandwidth here? Nobody is denying PS4 is the most powerful hardware in the console business but memory and bandwidth don't provide any power.

Re-read what I've posted on this page please. PS4 is more powerful than XBO and especially PS3 in every way with the exception of some CPU stuff - which has nothing to do with AF anyway.
 

RoboPlato

I'd be in the dick
PS4 has 176gb/s for everything where Xbone has DDR3 has 68gb/s and ESRAM 204gb/s theoretically. Its impossible in normal workloads to use 204gb/s from esram, but You can easily exceed 176gb/s on both DDR3 and ESRAM with proper bandwidth management.
I thought the 204 number was from adding DDR3 and eSRAM. You sure that's for eSRAM alone?

Did you look at the picture I posted above? You can't just subtract what the CPU can get and call it a day. There are constantly reads and writes happening, caches involved, waiting for data, huge chunks getting written while you wait for smaller data and so on. Again, it is a theoretical maximum and nothing what you can expect in real life. It's like the 1.8 TFlops. Theoretically the hardware has the capability but you won't see a game managing to execute so many TFlops ever in a game.
Looked back a few pages in the thread and didn't see any picture. I remember that the Oddworld New and Tasty devs said they were getting 172 pretty early on so most teams will probably be relatively close to that. Doesn't the cache coherency of the RAM pool also help with preventing some of the wait time and competition for bandwidth?
 

Jux

Member
What's that has to do with AF? GPUs have dedicated texture caches specifically to handle stuff like AF in a way which won't interfere with shading. Shader fetches happen via their own L1 caches. And let me repeat this again - low end GCN GPUs are providing nearly free 16x AF while running shaders which are kinda close to what PS4/XBO are running.

If you're saying that AF isn't free and will impact memory access latency - sure, it'll cost you some 10% or something. But the issue here is that PS3 and XBO have AF while PS4 doesn't. And that's not something which can be explained by the lack of power on PS4 as both PS3 and XBO are clearly less powerful than PS4.

AF is not free, there is actual latency issues inside the shader. When a texture fetch is done, even when coming from the cache, the shader still has to wait before getting the result. That wait can be hidden by other independ ALU operation in some cases but not always (especially when the shader is texture bound and not ALU bound). The GCN architecture also does a lot to hide those latencies by parallelising computation between wavefronts (a bit like pipelining) but in case of big shaders with a lot of register pressure that parallelisation cannot be done (not enough ressource to actually execute more wavefronts in parallele).
PS4 GPU profling tools easily show that kind of behavior.
Of course this does not explain those specifics cases (only the related devs could explain that), but I'm just saying that it's a lot more complicated than people think.
 
AF is not free, there is actual latency issues inside the shader. When a texture fetch is done, even when coming from the cache, the shader still has to wait before getting the result. That wait can be hidden by other independ ALU operation in some cases but not always (especially when the shader is texture bound and not ALU bound). The GCN architecture also does a lot to hide those latencies by parallelising computation between wavefronts (a bit like pipelining) but in case of big shaders with a lot of register pressure that parallelisation cannot be done (not enough ressource to actually execute more wavefronts in parallele).
PS4 GPU profling tools easily show that kind of behavior.
Of course this does not explain those specifics cases (only the related devs could explain that), but I'm just saying that it's a lot more complicated than people think.

The question of course is, does this extra layer of "complication" justify the use of low AF? Or is still pittance in terms of performance?
 
There's an in depth discussion of Bandwidth from ESRAM on Beyond3d at the moment. Since there's some calculations sprouting up here it seems appropriate to link it.

https://forum.beyond3d.com/threads/...ming-space-specifically-the-xb1.55487/page-57

Too much detail to easily quote, but just for example, comparing PS4 and Xbox One below. If you're interested in the detail I strongly suggest reading the thread linked above.

The ESRAM can dual-issue reads and writes for 7 out of 8 cycles. The 8th cycle cannot issue a write, but a read is possible.
The 192 would be the number offered before the upclock to 853 MHz, 204 is the current peak.
This is subject to patterns matching some unknown set of banking limitations. The real-world utilization figures show this is not readily acheived.

GDDR5 cannot simultaneously read and write, and any mix that isn't very long stretches of pure reads followed by pure writes is begging to have that 176 GB/s figure chopped by in bad cases 10x or worse.
DRAM is also subject to a number of banking restrictions, and the penalties are likely far worse for it that the ESRAM.
The sustained bandwidth figures are somewhere around 140, per some of Sony's own slides, and the figure drops further under CPU contention.

Comparing the two systems is a very complex proposition, particularly since both are not what would be considered mature.
 

Jux

Member
The question of course is, does this extra layer of "complication" justify the use of low AF? Or is still pittance in terms of performance?

I'm not saying it's complicated to implement (it's pretty trivial actually). I'm just saying the hardware implications are.
 

Jux

Member
Aye. If you had to guesstimate a range of performance hit what would you speculate?

Unfortunately there is no guesses to be made, it's too highly dependant on a lot of factors like the kind of shaders, how they are optimized, the amount of texture fetches needed for materials, etc. Every engine is different in that regard.
I think it's pretty safe to say that it's not a major bottleneck in most games though (other things can be cut down for better performance gain). It can be more touchy when targeting with 60fps though where every millisecond is important.
 
Unfortunately there is no guesses to be made, it's too highly dependant on a lot of factors like the kind of shaders, how they are optimized, the amount of texture fetches needed for materials, etc. Every engine is different in that regard.
I think it's pretty safe to say that it's not a major bottleneck in most games though (other things can be cut down for better performance gain). It can be more touchy when targeting with 60fps though where every millisecond is important.

That I assumed, especially concerning 60fps. Do you find it odd as a PS4 dev that people are turning it off / limiting it to 4x in games targetting 30fps with high quality textures and complex shading? It seems like the opposite of what I would do if I authored some great 2048X2048 PBR texture.
 
Top Bottom