gtx970 vs. rx480 Doom.. peculiar image quality discrepency

[TheRyuu]

is that cheating? lower visual quality for higher fps numbers?

A better question is will overall performance drop once the bug if fixed. Either way the effect is ugly how have reviewers not noticed this?

The visual impact is only when first entering an area. The texture quality equals out over time. It's the discrepancy in the amount of time it takes to reach the highest quality which is the issue that he brings up in the video. It probably relates to how the megatexture stuff works in the idtech engine.

I can't tell you why this behavior is occurring but I doubt it will affect performance for the reasons mentioned. I swear half the people in this thread didn't bother to watch the video or don't understand how the idtech engine works. It's not that there's an image quality discrepancy per se but there's an issue on Nvidia cards which causes the megatexture's to stream more slowly than on AMD cards.

 

[sweetusernames]

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.

Now I'd usually agree with this but the differences don't look to be differences made by compression tbh.

It's a very static scene. Compression usually works on movement (i.e change). This is where most artifacts are seen. The less movement, the prettier the image will look since more bitrate is spent on the image quality itself. It probably still is compressed, but the visual quality difference far too large to simply chalk up to compression.

That being said though, the carpet quality is lower, but look at the characters clothes in the video. They're all wildly different in quality. This is probably just some weird texture streaming/filtering/whatever in Hitman's engine.

EDIT : Talking about this scene, sorry

 

[SirCanealot]

Now I'd usually agree with this but the differences don't look to be differences made by compression tbh.

It's a very static scene. Compression usually works on movement (i.e change). This is where most artifacts are seen. The less movement, the prettier the image will look since more bitrate is spent on the image quality itself. It probably still is compressed, but the visual quality difference far too large to simply chalk up to compression.

Not really. A decent encoder will encode the entire video at 'similar' quality — in a static scene it will pull the bitrate down to save it for the scenes with more motion. Unless Youtube uses constant bitrate encoding of course (which it doesn't).

So you don't know exactly what the Youtube encoder does, so you can't compare these images, end of story unfortunately

I used to encode anime a lot and it's quite impressive what video encoders will do to fine details when smooshed down to a low bitrate (and Youtube does not have a decent bitrate at all).

 

[sweetusernames]

Not really. A decent encoder will encode the entire video at 'similar' quality — in a static scene it will pull the bitrate down to save it for the scenes with more motion. Unless Youtube uses constant bitrate encoding of course (which it doesn't).

So you don't know exactly what the Youtube encoder does, so you can't compare these images, end of story unfortunately

I used to encode anime a lot and it's quite impressive what video encoders will do to fine details when smooshed down to a low bitrate (and Youtube does not have a decent bitrate at all).

Youtube doesn't use CBR? huh, you do learn something everyday.

 

[antihelten]

So you don't know exactly what the Youtube encoder does, so you can't compare these images, end of story unfortunately

Not really end of story at all.

a) There's a limit to how much the Youtube encoder will mess up a video like this.
b) The encoder should mess up the three clips more or less equally (you wouldn't get a difference like this due to the encoder)
c) All of the texture issues are clearly replicated in still images (1, 2, 3), so this issue is not created by the encoder.

Also it's worth noting that if anyone thinks this is an issue specific to Nvidia, that there are plenty of examples of AMD having texture issues as well. Texture streaming in Hitman is basically just a mess.

 

[zinfamous]

Not enough VRAM for high quality textures, perhaps?

I'd like to see this visual test repeated with a 980 and 980ti.

This

 

[SirCanealot]

Not really end of story at all.

a) There's a limit to how much the Youtube encoder will mess up a video like this.
b) The encoder should mess up the three clips more or less equally (you wouldn't get a difference like this due to the encoder)
c) All of the texture issues are clearly replicated in still images (1, 2, 3), so this issue is not created by the encoder.

Also it's worth noting that if anyone thinks this is an issue specific to Nvidia, that there are plenty of examples of AMD having texture issues as well. Texture streaming in Hitman is basically just a mess.

It's the end of the story if someone doesn't want to go and compare lossless imagery now — we can clearly see there is an issue with NVidia cards, but unless we actually look at lossless images now it's still useless to compare as we only know half the story. I mean, has any noise reduction been used at any stage of the process? We don't what know it could be doing to the image at all!

If anyone wants to have a discussion on half a story, be my guest

But I couldn't sit here without giving my 2 cents

Youtube doesn't use CBR? huh, you do learn something everyday.

Not that I know anything for a fact from having done zero research, but CBR encoding is insanely wasteful and would mean the quality of the videos would be completely all over the place, which it isn't (it's generally awful everywhere ). Nothing should be using CBR encoding, ever!

 

[antihelten]

It's the end of the story if someone doesn't want to go and compare lossless imagery now — we can clearly see there is an issue with NVidia cards, but unless we actually look at lossless images now it's still useless to compare as we only know half the story. I mean, has any noise reduction been used at any stage of the process? We don't what know it could be doing to the image at all!

If anyone wants to have a discussion on half a story, be my guest

But I couldn't sit here without giving my 2 cents

Yes we can indeed clearly see that there is an issue with the Nvidia cards, and it is plainly obvious from the nature of the issue that this is not an encoder issue.

We don't need lossless images to compare things, I don't know were in the world you got this idea from. We just need images of sufficiently high quality, and both the youtube video and the still images I linked above are decent enough for this.

Noise reduction wouldn't produce the kinds of texture issues that we are seeing in the video, so it is irrelevant.

 

[zinfamous]

seems like testing it on a 7970 or 280x would answer the question. good suggestion.

1060 doesn't use "proper" asynch scheduler, correct? It might be best to compare with 280x and 970 for a VRAM test,

and a current gen 480 vs 1060 to compare within class with similar VRAM to check if it's a compute issue?

 

[dacostafilipe]

DOOM's Vulkan implementation uses Async compute to speed up the MegaTexture texture streaming.

The GTX 970 is not running Async compute.

This would make sense.

 

[SirCanealot]

Yes we can indeed clearly see that there is an issue with the Nvidia cards, and it is plainly obvious from the nature of the issue that this is not an encoder issue.

We don't need lossless images to compare things, I don't know were in the world you got this idea from. We just need images of sufficiently high quality, and both the youtube video and the still images I linked above are decent enough for this.

Noise reduction wouldn't produce the kinds of texture issues that we are seeing in the video, so it is irrelevant.

Thank god the scientific community generally don't think like this

As I said, it's completely up to you if you want to have a discussion on images from a heavily compressed YouTube video, but I wouldn't underestimate in anyway how badly YouTube can compress things.

Clearly we can see there is an issue, but don't you want to see what kind of issue and how bad it is?

Otherwise, surely the discussion is over? Yes, there's an issue. What else is there to talk about? I'd rather know more details myself

 

[coercitiv]

As I said, it's completely up to you if you want to have a discussion on images from a heavily compressed YouTube video, but I wouldn't underestimate in anyway how badly YouTube can compress things.

Clearly we can see there is an issue, but don't you want to see what kind of issue and how bad it is?

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?

 

[realibrad]

Thank god the scientific community generally don't think like this

As I said, it's completely up to you if you want to have a discussion on images from a heavily compressed YouTube video, but I wouldn't underestimate in anyway how badly YouTube can compress things.

Clearly we can see there is an issue, but don't you want to see what kind of issue and how bad it is?

Otherwise, surely the discussion is over? Yes, there's an issue. What else is there to talk about? I'd rather know more details myself

Actually, your argument is the very issue that Science has to fight against. Because the situation is not perfect, you are saying all data is invalid. The fact that Youtube compresses video is a constant, which means that both images should be effected the same way if the images were the same. The fact that you can see a difference means they are not the same. You are basing your comments off of ignorance on how compression works.

 

[coercitiv]

Actually, your argument is the very issue that Science has to fight against. Because the situation is not perfect, you are saying all data is invalid.

Scientists never use a weaker method of measurement when the better one is at hand, with little or no added effort. This is not what Science has to fight against when experiment data is being challenged.

 

[zinfamous]

Actually, your argument is the very issue that Science has to fight against. Because the situation is not perfect, you are saying all data is invalid. The fact that Youtube compresses video is a constant, which means that both images should be effected the same way if the images were the same. The fact that you can see a difference means they are not the same. You are basing your comments off of ignorance on how compression works.

exactly

 

[zinfamous]

Scientists never use a weaker method of measurement when the better one is at hand, with little or no added effort. This is not what Science has to fight against when experiment data is being challenged.

Oh, that happens all the time, doesn't always make it right. But that doesn't mean it can't or shouldn't be challenged.

Besides, even if the chosen methodology is flawed, it doesn't mean useful data can't be gleaned from that method, much less make it invalid.

Better methods are always best, but it doesn't mean an easier method isn't useful.

 

[SirCanealot]

So what I'm saying is we can see an issue, but we should do further research rather than just loudly bashing company A or company B:

(thank you, coercitiv)
Scientists never use a weaker method of measurement when the better one is at hand, with little or no added effort. This is not what Science has to fight against when experiment data is being challenged.

So instead of vaguely saying there's an issue, let's examine lossless screenshots so we can say exactly what the issue is and how bad it is so someone can raise this up to NVidia

Edit: I guess something like this can be raised up as-is, but it's always better to provide all of the information rather than some of it.

 

[zinfamous]

So what I'm saying is we can see an issue, but we should do further research rather than just loudly bashing company A or company B:

(thank you, coercitiv)
Scientists never use a weaker method of measurement when the better one is at hand, with little or no added effort. This is not what Science has to fight against when experiment data is being challenged.

So instead of vaguely saying there's an issue, let's examine lossless screenshots so we can say exactly what the issue is and how bad it is so someone can raise this up to NVidia

Edit: I guess something like this can be raised up as-is, but it's always better to provide all of the information rather than some of it.

but coercitiv's statement isn't always true, and is exactly why peer review exists. Scientists use poor methods all the time. Sometimes this is addressed in peer review and sometimes it follows through to publication.

My years of weekly lab meetings and journal clubs are pretty much defined by "well, these guys did this the wrong way." You can't really be in science for any significant amount of time without having that experience. But again--it still doesn't mean a lesser method is wholly without merit.

The other thing to consider is that yes, while it would be best to compare across perfect uncompressed data, it's the reporting that is the issue, and not the method--as far as we can tell. Sit down and do it yourself? Maybe the dude can only post to youtube and does so poorly. But it is also not the same as what he is seeing, right?

But again--as long as the redactedcompression is applied to all images equally (is it?) then meaningful data is still extracted.



No profanity in tech.


esquared
Anandtech Forum Director

 

[MrTeal]

So what I'm saying is we can see an issue, but we should do further research rather than just loudly bashing company A or company B:

(thank you, coercitiv)
Scientists never use a weaker method of measurement when the better one is at hand, with little or no added effort. This is not what Science has to fight against when experiment data is being challenged.

So instead of vaguely saying there's an issue, let's examine lossless screenshots so we can say exactly what the issue is and how bad it is so someone can raise this up to NVidia

Edit: I guess something like this can be raised up as-is, but it's always better to provide all of the information rather than some of it.

Could you please provide some lossless images for us to inspect?

 

[realibrad]

So what I'm saying is we can see an issue, but we should do further research rather than just loudly bashing company A or company B:

(thank you, coercitiv)
Scientists never use a weaker method of measurement when the better one is at hand, with little or no added effort. This is not what Science has to fight against when experiment data is being challenged.

So instead of vaguely saying there's an issue, let's examine lossless screenshots so we can say exactly what the issue is and how bad it is so someone can raise this up to NVidia

Edit: I guess something like this can be raised up as-is, but it's always better to provide all of the information rather than some of it.

With the current method, we can see there is a difference. All your method will bring is a better quantification of the issue. That said, the current method does show that there is an issue.

So this...

So you don't know exactly what the Youtube encoder does, so you can't compare these images, end of story unfortunately

is wrong. You can 100% compare the images and see there is a problem. What you then said is correct that there is a problem, but we dont know what the problem is. Where you went wrong again is saying that comparing lossless images will help understand the problem, and that is not true.

 

[antihelten]

Thank god the scientific community generally don't think like this

As I said, it's completely up to you if you want to have a discussion on images from a heavily compressed YouTube video, but I wouldn't underestimate in anyway how badly YouTube can compress things.

Clearly we can see there is an issue, but don't you want to see what kind of issue and how bad it is?

Otherwise, surely the discussion is over? Yes, there's an issue. What else is there to talk about? I'd rather know more details myself

I always find it rather cute when people who have clearly never worked in academia pretend to know how people within the scientific community think.

The vast majority of people working within science won't waste time trying to improve the quality of their data if the quality is already sufficient to show what they are trying to show. Your time is much better spend working on new experiments for your next publication.

In this case the video is more than sufficient to show that there is a problem with textures, and furthermore the kind of discrepancies we are seeing will never ever be produced by encoder artifacts. We already know what the issue is (textures are not loading properly) and your continued insistence on encoder artifacts as a plausible explanation, quite frankly just makes you look ignorant, you might as well argue that the phase of the moon is to blame.

The question that remains at this stage is not whether the textures are loading properly or not (they aren't), but rather why the textures are not loading properly (although hitman is arguably off topic, since this thread is about DOOM).

Scientists never use a weaker method of measurement when the better one is at hand, with little or no added effort. This is not what Science has to fight against when experiment data is being challenged.

Problem is that a better method that requires little to no added effort is almost never available. So if the weaker method provides data of sufficient quality, it would be a waste of time to use a better method if said method requires more resources.

 

[gamervivek]

Wow, why does it looks so washed out on the 980Ti and 1080?! Like AF not functional or on Low Quality or something.

Messed up man. Look at the carpet, no clarity at all, just a blur. Likewise for the rest of the room's details.

Reviewers are blind or they just AFK bench?

The washed out look on nvidia cards is due to their color setting default being so. That can be changed but I doubt you can change LoD settings.

Though this texture loading bug is something different.

 

[coercitiv]

Problem is that a better method that requires little to no added effort is almost never available. So if the weaker method provides data of sufficient quality, it would be a waste of time to use a better method if said method requires more resources.

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?

 

[SirCanealot]

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 always find it rather cute when people who have clearly never worked in academia pretend to know how people within the scientific community think.

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

Could you please provide some lossless images for us to inspect?

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

 

[MrTeal]

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?

That one's actually interesting, because it's not just the barcode. The 1080 shows the barcode and the wrinkles on the suit in sharp focus, while the 980Ti and Fury X show them fuzzy. It's not until after he looks away that the wrinkles appear on the FuryX/980Ti, and the barcode is shown in focus when he looks straight again.

If this was actual video of reality I'd say it looks a lot like 1080 has foreground focus in the first part while the other two don't focus on the guy until he turns, but I have no idea if it's a depth of field issue or not or if that's the look they're going for. It definitely does look a little jarring to have the wrinkles pop in through.