gtx970 vs. rx480 Doom.. peculiar image quality discrepency

[MrTeal]

All I've said is that I disagree with everyone discussing this using a Youtube encode.

As I've said, if you guys are all happy to discuss it with the evidence you have, that's fine, but you have to accept criticisms from people like me who think the Youtube encode is complete tosh and you are way underestimating the damage it will randomly cause to video quality.



I apoligies for making a generalisation on the scientific community; that was probably a bad example to use.



Yes: http://lmgtfy.com/?q=.png

I didn't ask for a lossless image format, I asked for lossless images for us to look at. Since there's easily available data that takes no extra effort to get, it would be great if you could provide that for us so we can have a better discussion using this best available data.

 

[realibrad]

All I've said is that I disagree with everyone discussing this using a Youtube encode.

As I've said, if you guys are all happy to discuss it with the evidence you have, that's fine, but you have to accept criticisms from people like me who think the Youtube encode is complete tosh and you are way underestimating the damage it will randomly cause to video quality.



I apoligies for making a generalisation on the scientific community; that was probably a bad example to use.



Yes: http://lmgtfy.com/?q=.png

Your premise is flawed. the encoding is not random. Encoding through Youtube will degrade video quality, but it will degrade it the same way if the videos were the same. The effects we see, if caused by encoding would happen to both Nvida and AMD, yet clearly we see differences. If your argument is that encoding is random, then we would not see the same effect on just Nvidia's side and not see it on AMD's side.

To say we dont have all the details so we should not make a judgement is flawed. We can say there is a problem on Nvidia's side. How big and what the cause is cannot be determined by Youtube videos, but we can say there is a problem.

You dont need perfect understanding of physics and the location of every atom to use shoes. You can use shoes quite well without any understanding of chemistry. We dont need to be pedantic.

 

[gamervivek]

Fair enough, let's work with available data.

Based on what we can gather from the sequence pictured bellow, would you say Fury X suffers from similar problems (barcode texture)? What does that tell us?

It's not similar to the pic that was posted earlier because you could clearly see the difference in texture between 1080 vs. Fury X and 980Ti in that, whereas this is more of a focus issue.

The timstamped video I posted shows this difference as well between all the cards with the clothes looking different on the agent, but that wasn't the issue in the video I posted.

Look at the rug in this scene and the green machine in the next, this is not a video encoding artifiact.

https://www.youtube.com/watch?v=i9ZG7pA5INc&t=2m15s

The pic Erenhardt posted shows similar lower texture setting on the 1080.

 

[antihelten]

Fair enough, let's work with available data.

Based on what we can gather from the sequence pictured bellow, would you say Fury X suffers from similar problems (barcode texture)? What does that tell us?

Yes I would absolutely say that AMD/Fury X suffers from the same/similar issues (I also said as much earlier), this is not the only example btw, there are several other places in the same video where AMD is also suffering from low res textures compared to the 1080, furthermore this link (which I also provided earlier) clearly shows an R9 Nano having texture issues (at least under DX12).

As for what it tells us, well it would certainly indicate that this is not an issue specific to Nvidia cards or Nvidia drivers, but more likely is an issue with the game itself.

 

[railven]

When people react to images rather than arguments, give them an image :

In the Fury X segment of the video, the barcode texture becomes sharper only after the character makes the first head movement. Before that it's a blur.

What does that tell us?

Barcode? Is no one looking at the drapes/curtains? WTF is going on in the Fury X and GTX 1080 that the tassels are basically blurred garbage? Did no one else notice that? The lower scoring 980 Ti is the only one actually rendering the tassels.

 

[antihelten]

Barcode? Is no one looking at the drapes/curtains? WTF is going on in the Fury X and GTX 1080 that the tassels are basically blurred garbage? Did no one else notice that? The lower scoring 980 Ti is the only one actually rendering the tassels.

PCGameshardware.de saw very similar issues when they tested Hitman.

Based on their tests the issue would appear to be partly due to a faulty DX12 port (the R9 nano looks perfectly fine under DX11, but starts having issues under DX12), and partly due to a memory protection feature (The 980 Ti shows clear improvements when this feature is turned off, although it is still not perfect).

 

[coercitiv]

Barcode? Is no one looking at the drapes/curtains? WTF is going on in the Fury X and GTX 1080 that the tassels are basically blurred garbage? Did no one else notice that? The lower scoring 980 Ti is the only one actually rendering the tassels.

I was just preparing that, all in good time! :biggrin:

Even 980Ti gives up on proper rendering by the end of the sequence cut. See frame 4.

 

[gamervivek]

Barcode? Is no one looking at the drapes/curtains? WTF is going on in the Fury X and GTX 1080 that the tassels are basically blurred garbage? Did no one else notice that? The lower scoring 980 Ti is the only one actually rendering the tassels.

And that's the problem with still images. If you look at the video, Fury X has that texture just for a moment while everything comes into focus but has the best defined texture for the rest of the time while 1080 is still that way for all the time.

Above sequence starts from 20 seconds into the video.

 

[TheRyuu]

The fact that Youtube compresses video is a constant

Except it isn't, at least in the example of a side by side comparison such as the videos posted. I'd agree with that statement if it was two separate videos uploaded with the same content but the only difference was the video card used where the user was left to tab between them but even then this could only be used to suggest the possibility of an issue (which I'm fine with). With a side by side by side comparison you really don't have any idea what the encoder or rate control is going to do. There are also psychovisual behaviors in the encoder which make it not as predictable as it otherwise would be with similar but not identical content.

Also 60fps videos on Youtube appear to take a quality hit due to them not raising the bitrate (or quality) level enough[1]. Although granted even if his Far Cry Primal video was 30fps I still think the encoder would shit itself with the detail in that case.

I'd like to reiterate I'm not saying that an issue doesn't exist here just that it's hard to use the youtube videos to draw any conclusion other than "perhaps an issue exist here in this side by side by side comparison". I may buy the evidence (a little) more if an upscaled 1440p quality was also included.

[1] https://www.youtube.com/watch?v=dJQX0tZsZo4

 

[3DVagabond]

Except it isn't, at least in the example of a side by side comparison such as the videos posted. I'd agree with that statement if it was two separate videos uploaded with the same content but the only difference was the video card used where the user was left to tab between them but even then this could only be used to suggest the possibility of an issue (which I'm fine with). With a side by side by side comparison you really don't have any idea what the encoder or rate control is going to do. There are also psychovisual behaviors in the encoder which make it not as predictable as it otherwise would be with similar but not identical content.

Also 60fps videos on Youtube appear to take a quality hit due to them not raising the bitrate (or quality) level enough[1]. Although granted even if his Far Cry Primal video was 30fps I still think the encoder would shit itself with the detail in that case.

I'd like to reiterate I'm not saying that an issue doesn't exist here just that it's hard to use the youtube videos to draw any conclusion other than "perhaps an issue exist here in this side by side by side comparison". I may buy the evidence (a little) more if an upscaled 1440p quality was also included.

[1] https://www.youtube.com/watch?v=dJQX0tZsZo4

So, if you aren't saying the issue doesn't exist then why are you arguing the point? Just trying to throw up a smoke screen?

 

[railven]

So what I'm seeing here, Hitman is a shoddy port or something? It isn't consistent on any platform?

 

[Raising]

Looks like it's one of these games that require an SSD because the redactedengine doesn't pre-cache nearby textures as it should.




No profanity in tech.


esquared
Anandtech Forum Director

 

[sweetusernames]

It's not hard to see there's a difference in Hitman textures. I'd even argue you'd be able to see the difference in compressed images.

There is literally no way that compression could cause such a large difference between the texture quality. You're not running the side by side through needsmorejpeg.com.

You guys are implying that video compression is random, or as random to the point where it is capable of creating the texture difference in the carpet still we see. There is no way as far as I can see. Explain to me how it could compress three near identical images differently? Other than "encoder voodoo magic" of course.

As far as I know, video compression is mostly DCT (basically rewriting the data as cosine functions to make it easier to compress), quantization and encoding (making your DCT'ed block nice and compressed), frame prediction, and psychovisual hacks that reduce image data even further, by reducing colors, deciding which parts of the image get more or less compressed, dealing with luminance, etc.

I mean, I can't see how all of that can manage to produce the difference with the carpets in this scene. In fact, I think these techniques should be more formulaic and make the images more similar, rather than make them look different. I don't think it would be desirable for codec designers to have small differences in the source video to manifest as larger differences in the final one.

I mean, I might (there's a good chance, since I don't know much about the nit and grit of encoding) be completely wrong, feel free to show me, but if you need more proof than a youtube video, I need more proof than saying "oh no you don't know what the encoder can do to the image".

 

[coercitiv]

I mean, I can't see how all of that can manage to produce the difference with the carpets in this scene.

Since this scene keeps being brought up, has there been any discussion on the image quality for the character's clothes in that very scene? The shirt is most detailed on 1080 footage, loses quality by comparison on the Fury X and is a blurry mess on 980 Ti. So it seems to me one gets to choose between a better carpet and a better shirt, but will not get both. (and in the next scene the same shirt is more detailed on the Fury X footage)

Image quality for all the cards in these recordings is strange to say the least, no card is consistent throughout. For the third time I find myself asking the same question: What does that tell us?

 

[Bacon1]

Since this scene keeps being brought up, has there been any discussion on the image quality for the character's clothes in that very scene? The shirt is most detailed on 1080 footage, loses quality by comparison on the Fury X and is a blurry mess on 980 Ti. So it seems to me one gets to choose between a better carpet and a better shirt, but will not get both. (and in the next scene the same shirt is more detailed on the Fury X footage)

Image quality for all the cards in these recordings is strange to say the least, no card is consistent throughout. For the third time I find myself asking the same question: What does that tell us?

Maybe its a DOF setting, so its focusing in different places? No idea but it looks terrible.

 

[SirCanealot]

So, if you aren't saying the issue doesn't exist then why are you arguing the point? Just trying to throw up a smoke screen?

See, that's the thing. People need to stop exploding and saying 'OMG, OUR OPINIONS DON'T MATCH... EXTERMINATE, EXTERMINATE!!!!'

I'm just trying to say that you don't know how the Youtube encoder works and I don't think it can be used as reliable evidence for issues like this. If you want to get to the bottom of it, use lossless pictures. Please read TheRyuu's post above.

But then, again, if you all want to discuss this based on Youtube footage, then I don't have a problem with this. I just agree with the criticisms that now an initial investigation has been done with Youtube footage, someone should get looking at lossless pictures.

Again: It's okay for you to have a different opinion to me and other people in this thread. It's okay.

 

[3DVagabond]

See, that's the thing. People need to stop exploding and saying 'OMG, OUR OPINIONS DON'T MATCH... EXTERMINATE, EXTERMINATE!!!!'

I'm just trying to say that you don't know how the Youtube encoder works and I don't think it can be used as reliable evidence for issues like this. If you want to get to the bottom of it, use lossless pictures. Please read TheRyuu's post above.

But then, again, if you all want to discuss this based on Youtube footage, then I don't have a problem with this. I just agree with the criticisms that now an initial investigation has been done with Youtube footage, someone should get looking at lossless pictures.

Again: It's okay for you to have a different opinion to me and other people in this thread. It's okay.

What I saw was obvious difference in LOD. It wasn't any type of video compression error or loss of info from you tube. Anyone who's made game models would recognize the effect.

And nobody's exploding. Maybe you need to relax a little.

 

[antihelten]

See, that's the thing. People need to stop exploding and saying 'OMG, OUR OPINIONS DON'T MATCH... EXTERMINATE, EXTERMINATE!!!!'

I'm just trying to say that you don't know how the Youtube encoder works and I don't think it can be used as reliable evidence for issues like this. If you want to get to the bottom of it, use lossless pictures. Please read TheRyuu's post above.

You still don't seem to understand that the encoder would not in a million years be capable of creating issues like the ones discussed here and as such the video is perfectly reliable.

You're basically arguing that the youtube encoder is somehow this magical beast that is capable of perfectly identifying the exact boundaries of textures/decals/character models in a frame and then even though the exact same object is present three times in the same frame selectively blur only 1 or 2 of them. Even more than that the encoder would also have to be capable of outright removing certain alpha textures (the drapes) whilst accurately drawing in the part of the scene that would normally be hidden behind said alpha texture.

If you think the Youtube encoder is actually capable of doing this then you might want to let Google know that they have just invented the most powerful image recognition and scene reconstruction software known to man.

Your claims are about as realistic as forensic cops going "enhance, enhance" on cop shows.

But then, again, if you all want to discuss this based on Youtube footage, then I don't have a problem with this. I just agree with the criticisms that now an initial investigation has been done with Youtube footage, someone should get looking at lossless pictures.

Again lossless images are not necessary, what we have is more than sufficient to unequivocally say that Hitman has a texture rendering issue.

The thing about lossy formats is that the errors they introduce are essentially random and as such they would never be able to introduce the kinds of selective errors we are seeing here.

These images below are jpg images, and thus in a lossy format, but the selective texture errors seen in the second picture would never result from lossy encoding:

(The first image is an R9 nano in DX11 and the second image the same card in DX12. Exact same settings except for the change in API)

Again: It's okay for you to have a different opinion to me and other people in this thread. It's okay.

And it's also okay for you to actually try and do the bare minimum to inform yourself about how image rendering and encoding actually works.

 

[realibrad]

Except it isn't, at least in the example of a side by side comparison such as the videos posted. I'd agree with that statement if it was two separate videos uploaded with the same content but the only difference was the video card used where the user was left to tab between them but even then this could only be used to suggest the possibility of an issue (which I'm fine with). With a side by side by side comparison you really don't have any idea what the encoder or rate control is going to do. There are also psychovisual behaviors in the encoder which make it not as predictable as it otherwise would be with similar but not identical content.

Also 60fps videos on Youtube appear to take a quality hit due to them not raising the bitrate (or quality) level enough[1]. Although granted even if his Far Cry Primal video was 30fps I still think the encoder would shit itself with the detail in that case.

I'd like to reiterate I'm not saying that an issue doesn't exist here just that it's hard to use the youtube videos to draw any conclusion other than "perhaps an issue exist here in this side by side by side comparison". I may buy the evidence (a little) more if an upscaled 1440p quality was also included.

[1] https://www.youtube.com/watch?v=dJQX0tZsZo4

If you were right, then why do we only see the issues on the Nvidia video? I addressed this in a different post after the one you responded to. If he encoding were changing the video then it should be affecting the AMD video randomly as well. Unless you believe that the Nvidia video got super unlucky and got all of the random issues that made it look worse, then you are left with the issue being on Nvidia's side.

Do you believe, or anyone else for that matter, that the encoding randomly affected one video, but not the other?

 

[SirCanealot]

You're basically arguing that the youtube encoder is somehow this magical beast that is capable of perfectly identifying the exact boundaries of textures/decals/character models in a frame and then even though the exact same object is present three times in the same frame selectively blur only 1 or 2 of them.

No, I'm saying the Youtube encoder and decoder applies vaseline all over the image and you have no way of telling how much vaseline it has applied to what parts of the image.

But let me quote myself:
"But then, again, if you all want to discuss this based on Youtube footage, then I don't have a problem with this. I just agree with the criticisms that now an initial investigation has been done with Youtube footage, someone should get looking at lossless pictures."

Let's also quote TheRyuu again:

"Except it isn't, at least in the example of a side by side comparison such as the videos posted. I'd agree with that statement if it was two separate videos uploaded with the same content but the only difference was the video card used where the user was left to tab between them but even then this could only be used to suggest the possibility of an issue (which I'm fine with). With a side by side by side comparison you really don't have any idea what the encoder or rate control is going to do. There are also psychovisual behaviors in the encoder which make it not as predictable as it otherwise would be with similar but not identical content."

Also the merits of comparing Youtube encodes and JPEG images are completely different too — are you going to tell me they're not?

 

[realibrad]

No, I'm saying the Youtube encoder and decoder applies vaseline all over the image and you have no way of telling how much vaseline it has applied to what parts of the image.

But let me quote myself:
"But then, again, if you all want to discuss this based on Youtube footage, then I don't have a problem with this. I just agree with the criticisms that now an initial investigation has been done with Youtube footage, someone should get looking at lossless pictures."

Let's also quote TheRyuu again:

"Except it isn't, at least in the example of a side by side comparison such as the videos posted. I'd agree with that statement if it was two separate videos uploaded with the same content but the only difference was the video card used where the user was left to tab between them but even then this could only be used to suggest the possibility of an issue (which I'm fine with). With a side by side by side comparison you really don't have any idea what the encoder or rate control is going to do. There are also psychovisual behaviors in the encoder which make it not as predictable as it otherwise would be with similar but not identical content."

Also the merits of comparing Youtube encodes and JPEG images are completely different too — are you going to tell me they're not?

What do you think there is to gain at this point looking at lossless images? Do you think that will somehow help show what is causing the issue?

 

[Nhirlathothep]

on my old 2 x 7970ghz

doom has terrible textures because they load too late.

 

[antihelten]

No, I'm saying the Youtube encoder and decoder applies vaseline all over the image and you have no way of telling how much vaseline it has applied to what parts of the image.

Except you absolutely do have a way of telling how much vaseline it applies to what parts of the image, namely due to the fact that no encoder in existence will selectively "vaseline" specific textures and nothing else on the image.

But let me quote myself:
"But then, again, if you all want to discuss this based on Youtube footage, then I don't have a problem with this. I just agree with the criticisms that now an initial investigation has been done with Youtube footage, someone should get looking at lossless pictures."

Repeating this doesn't make it any more relevant or correct. Lossless images are still not necessary, no matter how many times you keep claiming it.

Let's also quote TheRyuu again:

"Except it isn't, at least in the example of a side by side comparison such as the videos posted. I'd agree with that statement if it was two separate videos uploaded with the same content but the only difference was the video card used where the user was left to tab between them but even then this could only be used to suggest the possibility of an issue (which I'm fine with). With a side by side by side comparison you really don't have any idea what the encoder or rate control is going to do. There are also psychovisual behaviors in the encoder which make it not as predictable as it otherwise would be with similar but not identical content."

TheRyuu is just plain wrong if he honestly believes that the encoder would behave in a way that could create the errors we are seeing here.

Also encoders don't have psychovisual behaviors, they have various compression tricks to take advantage of the psychovisual behaviours found in the human visual system.

Also the merits of comparing Youtube encodes and JPEG images are completely different too — are you going to tell me they're not?

At the most fundamental level they are not different. Sure they may differ in implementation, but no encoder or compression format will ever behave in the manner observed here.

The only notable difference between video compression and image compression is the possibility of interframe compression with videos, but again that will never result in the kind of errors we are seeing here.

 

[SirCanealot]

"Except you absolutely do have a way of telling how much vaseline it applies to what parts of the image, namely due to the fact that no encoder in existence will selectively "vaseline" specific textures and nothing else on the image."

So how does the encoder decide how much detail it removes from a scene to compress to a certain bitrate?

"Repeating this doesn't make it any more relevant or correct. Lossless images are still not necessary, no matter how many times you keep claiming it."

I didn't say it was a necessity, I said I disagree with using them at this point in the discussion. Did I actually say this? If so, I apologise, that's not what I meant to imply.

"The only notable difference between video compression and image compression is the possibility of interframe compression with videos, but again that will never result in the kind of errors we are seeing here."

Exactly, so on the video are you comparing a I frame, B frame or P frame?

Again, let me quote myself:
"But then, again, if you all want to discuss this based on Youtube footage, then I don't have a problem with this. I just agree with the criticisms that now an initial investigation has been done with Youtube footage, someone should get looking at lossless pictures."

So you guys can keep arguing with me if you like, but it won't change my opinion on this discussion.

 

[sxr7171]

You can't use a youtube video as a valid comparison. The video compression artifacts make any comparison meaningless since it will taint the results. The video compression can make 3 identical scenes look different side by side, there's just now way to tell. You need to be comparing lossless screenshots.

Your point is valid but, what's happening with that handkerchief has nothing to do with video compression.