Jump to content
Floppy_Sock

non-uniform rendering size across different resolutions

Recommended Posts

Posted (edited)

Hi all,

 

I've stumbled across something, which in my mind, is quite concerning for a large portion of the community who play this game at higher resolutions ( > 1440p) or rely on some form of supersampling (VR users). I'm not sure it's been discussed before so I felt I should share. 

 

Before I bore you with the details of my test let me be frank about my observation: 

 

Edit: thanks to @SharpeXB for the test, there are situations where this observation does not hold. As of now, I’m not sure when it happens or when it doesn’t. I need to test more. I’ll likely wait until after the patch and see if things change and add to this. 
 

Conclusion: 

The game does not render objects above ~2k equally across varying resolutions. At further distances, the NUMBER of pixels associated with a contact is uniform across the three resolutions I tested. Meaning that a contact at a distance of 5km which is 4 pixels wide for a 1080p screen is ALSO 4 pixels wide for a 4k display, meaning the contact appears half the size for the player in 4k.   This detrimental to any player attempting to compete in a multiplayer environment.

 

BF109 size at 1.05 km (width in pixels)

1080p: 7 pixels
1440p: 9 pixels (1.28x larger ~ 1.33)
2160p: 13 pixels (1.85x larger ~ 2)

Not shown in the method's section below but included in the google drive folder:

6880x2880: 23 pixels (3.28x larger ~3.6)

 

BF109 size at 5.35km (width in pixels)

1080p: 5 pixels
1440p: 4 pixels (1x larger =/= 1.33)
2160p: 4 pixels (1x larger =/= 2)

Not shown in the method's section below but included in the google drive folder:

6880x2880: 3 pixels....

 

Method:

Open the spoiler section for details

Spoiler

 

The way I tested this is by doing the following. I played back a replay of a track I use to test visibility. I set the game to three resolutions: 1080p, 1440p, 4k (2160p). I loaded each track and never unpaused the game and took a screenshot with each of the following resolution. Here is a full screen clip of the testing image at 1080p. 1080p_raw.thumb.png.5aba019c0bbdac0ebe339e75a7af5c85.png

 

Unfortunately the upload size limit prevents me from sharing the raw 4k image as well. All three raw image files are included in a google drive link at the bottom for anyone curious. 

 

What I expected to see was that for each contact on the 4k version, we should see twice the number of pixels (or approximately) as for the 1080 version. As it turns out, this theory holds, in most situations up close. As a test, I compared the bf109 at 1.05km away. In each image the object of interest is boxed in yellow. At the bottom left (boxed in red)  of each image is the size of the selection (what's bounded in the dashed white box inside the yellow box) in pixels followed by the size of the image I'm looking at. 

1080p

1080p_snip2.thumb.png.97d428292862fb712e87e734d2ff94b8.png

1440p

1440p_snip2.thumb.png.0aa50fd5ade20af436951f44fa7a5b13.png

2160p

4k_snip2.thumb.png.8c8bf407d34c0ddc8a29e697a99d1be0.png

What you see is the following:

1080p: 7 pixels
1440p: 9 pixels (1.28x larger ~ 1.33)
2160p: 13 pixels (1.85x larger ~ 2)

 

When you consider aliasing, LOD etc, I think these results are more than satisfactory. If you compare the raw images in full, both contacts are equally visible. 

 

Great.

 

Now on to the distant contacts. Same format as above. Now at 5.35km away

1080p

1080snip.thumb.png.f83118bfc6529b03520c699f843fab84.png

1440p

14440snip.thumb.png.90aecaa64247ea1771dff0f6ef3f1157.png

2160p

4ksnip.thumb.png.52539f0f4593903b3e731a2b22198e36.png

 

What you see is the following:

1080p: 5 pixels
1440p: 4 pixels (1x larger =/= 1.33)
2160p: 4 pixels (1x larger =/= 2)

 

This seems unacceptable to me in a multiplayer game. There can't be such huge discrepancies in contact sizes across different display resolutions. See for yourself, download the images and compare them side by side. Be sure you're not viewing them compressed. The least that can be done is use a bottom line for all displays, i.e. if an object is 4 pixels at 1080p, it should be 8 at 4k. 

 

 

Download raw images here: 

https://drive.google.com/drive/folders/1QlmpSIVybLlrGTJ_36kpIGDEVovUuCrA?usp=sharing

Edited by Floppy_Sock
  • Like 1
  • Upvote 4

Share this post


Link to post
Share on other sites

Yes, several people have noticed this and we have been recommending people to lower their resolutions in the mean time until the new render is released.

Share this post


Link to post
Share on other sites
17 hours ago, Geronimo553 said:

Yes, several people have noticed this and we have been recommending people to lower their resolutions in the mean time until the new render is released.

 

This is why I can see stuff from 25km away even without touching the zoom button. First gen Oculus, very low res. Hahaha. 4 pixels to me is huge.

Share this post


Link to post
Share on other sites
Posted (edited)

I understand why the Devs have done this.

If they didn't, some of those contacts might go below a pixel in size for lower resolutions or be too ill defined.

 

There are clearly problems with this approach, however there are problems with every approach to this issue.

 

I would do things a different way, but I understand why they have tried to do things this way.

Edited by [DBS]Browning

Share this post


Link to post
Share on other sites

The issue is you have to uniformly bound the spotting from below. By that I mean, if you make the smallest possible contact 4 pixels at 1080p. Then why not make it 8 pixels at 4k????? 

 

That's totally reasonable isn't it? 

Share this post


Link to post
Share on other sites
On 5/14/2020 at 8:54 PM, Floppy_Sock said:

The game does not render objects above ~2k equally across varying resolutions.

Are you using Alternate Visibility?

Share this post


Link to post
Share on other sites

No, in this testing I am not. Though I think alternative viz kicks in above 10k? I might be wrong though 

Share this post


Link to post
Share on other sites
Posted (edited)
12 minutes ago, Floppy_Sock said:

No, in this testing I am not. Though I think alternative viz kicks in above 10k? I might be wrong though 

As far as I understand it, Alternate only does make a difference beyond 10km

I’m using a 4K monitor and see no difference between Alternate and Normal Visibility.

What is your native monitor resolution?

Edited by SharpeXB

Share this post


Link to post
Share on other sites
Posted (edited)

I generally play in VR. If I'm playing in 2D I have a 3440x1440 ultrawide that I use. 

 

I actually went ahead and did the same test down to 800x600 and up to 6880x2880 (double 3440) and my conclusions were the same. 3 pixels at 800x600 and 4 pixels at 6880. 

 

You're at a really really big disadvantage in 4k with the way the rendering is right now I'm sorry to say. 

Edited by Floppy_Sock
  • Upvote 1

Share this post


Link to post
Share on other sites

Thanks for posting this!, ive been noticing some issues. im glad theres a temp fix for now, hope itll get better with the new update :)  

  • Like 1

Share this post


Link to post
Share on other sites
12 hours ago, Floppy_Sock said:

I generally play in VR. If I'm playing in 2D I have a 3440x1440 ultrawide that I use. 

 

I actually went ahead and did the same test down to 800x600 and up to 6880x2880 (double 3440) and my conclusions were the same. 3 pixels at 800x600 and 4 pixels at 6880. 

 

You're at a really really big disadvantage in 4k with the way the rendering is right now I'm sorry to say. 

It’s hard to make these kind of conclusions if you aren’t looking at native resolution. Looking at 1080p on a 1080p screen and 4K on an actual 2160p screen. Upscaling causes some odd effects. Looking at a native 4K screen, distant contacts are razor sharp and clear even if they don’t bloom out to cover more pixels. They are easily visible, especially if they’re against another texture like the ground. Yes blurring can make things appear “bigger” but that’s not really better. 
And the perceived sharpness doesn’t translate. A 1080x1920 output to a 1080p can look very sharp and clear but that same resolution sent to a 2160p looks very soft. Even 1440p upscaled looks soft on a 4K screen. I wouldn’t lower resolution anymore than I would play without my glasses on. 4K is definitely not a disadvantage. 

Share this post


Link to post
Share on other sites
Posted (edited)

I'm upscaling with DSR. The screenshots are pre-sample. As such, they're true native 4k images with zero processing on them.

 

If you're not familiar, that's how DSR works. The game, and windows for that matter, believes the signal going to the monitor is 4k. The final step in DSR is to sample the 4k image to drop it down to 1080p (in my case). But if I take a screenshot, or record, the recordings, screenshots etc all occur before this sampling happens.   I've disabled all smoothing filters on nvidia's side so there is no post-processing going on in my tests. Please check the raw images if you're curious. 

 

That being said, I agree, contacts up close are very clear in 4k. I'm not disputing that. The first sequence of 3 images in the methods sections make that very clear. You see the 109 at 1.05km is rendered with far more detail in the 4k image than in the 1080p image. This is how it SHOULD be, uniformly, across all distances. 

The issue is, that, after a certain distance, this "scaling," which should equate the relative size of objects across various resolutions stops happening. That planes are not rendered at the same relative size. This IS a disadvantage. The object is half the size on your screen as it is on mine.

 

I encourage you to download the screenshots I used for testing and do the comparison yourself if you don't believe me. At the least, go through the methods section contained in the spoiler tab carefully. 

 

Even better, you could run the same tests yourself. I would greatly appreciate it if you did.

Edited by Floppy_Sock

Share this post


Link to post
Share on other sites
45 minutes ago, Floppy_Sock said:

I'm upscaling with DSR. The screenshots are pre-sample. As such, they're true native 4k images with zero processing on them.

I’m not sure how you’re making screenshots which aren’t at your screen resolution. I have experimented with DSR on a 1080p screen. 
The bottom line is you can’t see what the game looks like in 4K without a 4K screen anymore than you can see what 1080p looks like on a 480p screen. 
And as far as I know about how visibility works in IL-2 the game uses LODs with a scaling algorithm applies (a larger factor of course is used in Alternate)

The game doesn’t use pixel sized sprites like DCS did, which were set at a certain number of pixels and therefore did indeed appear larger at lower resolutions. 

Share this post


Link to post
Share on other sites
2 minutes ago, SharpeXB said:

I’m not sure how you’re making screenshots which aren’t at your screen resolution. I have experimented with DSR on a 1080p screen. 
The bottom line is you can’t see what the game looks like in 4K without a 4K screen anymore than you can see what 1080p looks like on a 480p screen. 
And as far as I know about how visibility works in IL-2 the game uses LODs with a scaling algorithm applies (a larger factor of course is used in Alternate)

The game doesn’t use pixel sized sprites like DCS did, which were set at a certain number of pixels and therefore did indeed appear larger at lower resolutions. 

 

I don't mean to be rude, but I just told you how I'm doing it. DSR allows me to do it. There is no sampling going on. With all due respect, you clearly do not understand how DSR works. 

 

If you don't believe me, do the following. Enable DSR 4x in your nvidia control panel. You can find it under "manage 3d settings" -> "global settings" -> "dsr factors" -> check the "4x" option.   Make sure "dsr smoothness" is turned down to 0.  Fire up il2, make sure it's in full screen mode, you should be able to set the resolution in il2 to be 8k (7680x4320). Restart the game. If the game looks like you only see a quarter of your screen, alt tab a few times in and out until it adjusts to the proper scale. In the main menu, take a screenshot using printscreen. Open up paint and paste it in and look at the size of the image in pixels in the bottom left corner. You will see it will actually be 8k. 

 

I think the confusion might be in the fact that I'm not using what I see on my actual screen as a basis for comparison. What I see on my 1080p screen with DSR enabled is sampled.  Here I agree, I cannot compare what I see on my sampled 1080p screen to what you see on your 4k screen. There you are correct. However, the screenshots are not sampled. The downsampling to 4k (or to 1080p for me)  occurs somewhere in the graphics pipeline after you can take a screenshot. There are many threads on using DSR to take very high resolution screenshots in many games. Just do a bit of googling. 

 

I am well within the alternative visibility range. So that setting does not matter. 

 

 

 

 

Share this post


Link to post
Share on other sites
9 minutes ago, Floppy_Sock said:

 

If you don't believe me, do the following. Enable DSR 4x in your nvidia control panel. You can find it under "manage 3d settings" -> "global settings" -> "dsr factors" -> check the "4x" option.   Make sure "dsr smoothness" is turned down to 0.  Fire up il2, make sure it's in full screen mode, you should be able to set the resolution in il2 to be 8k (7680x4320). Restart the game. If the game looks like you only see a quarter of your screen, alt tab a few times in and out until it adjusts to the proper scale. In the main menu, take a screenshot using printscreen. Open up paint and paste it in and look at the size of the image in pixels in the bottom left corner. You will see it will actually be 8k.

I understand what you’re doing. I just don’t think you see a realistic depiction of how that looks to a player on an actual screen. And afaik the scaling in IL-2 isn’t set by a certain number of pixels like DCS was. If so then yes that’s quite wrong because it would indeed make targets larger at lower resolutions. And it would penalize higher resolutions and higher pixel density in VR. You may well be on to something because I agree that in 4K the Alternate Visibility effect isn’t noticeable and the same scaling algorithm is actually used in Normal as well, it’s just not as pronounced. I don’t feel like I could test this myself though without two monitors side by side running their native resolutions.

Share this post


Link to post
Share on other sites
Posted (edited)

 

57 minutes ago, SharpeXB said:

I understand what you’re doing. I just don’t think you see a realistic depiction of how that looks to a player on an actual screen. And afaik the scaling in IL-2 isn’t set by a certain number of pixels like DCS was. If so then yes that’s quite wrong because it would indeed make targets larger at lower resolutions. And it would penalize higher resolutions and higher pixel density in VR. You may well be on to something because I agree that in 4K the Alternate Visibility effect isn’t noticeable and the same scaling algorithm is actually used in Normal as well, it’s just not as pronounced. I don’t feel like I could test this myself though without two monitors side by side running their native resolutions.

 

Right - but I feel that what I showed clearly demonstrates that IL2 does set distant contacts, at least in this situation, by a specific number of pixels. 

 

Do you have a 1080p screen next to a 4k screen in your setup?

 

If you do, could you do the following: download this image: https://drive.google.com/open?id=1fFLCqB6zuZ-rvgvlihIwFf_pzpxSyywz and view it full screen on your 4k monitor and download this https://drive.google.com/open?id=1pBwv12vBtWVdUZQK8MNcElTtSKVZ97sW image and view it fullscreen on your 1080p monitor. 

 

View them side by side and go through contact by contact. From the closest one at 1.05km all the way out to the one at 7k. 

 

Do you perceive a difference in contact size across the two images. If you do, where do you see the discrepancy start.

Edited by Floppy_Sock
  • Like 1
  • Upvote 1

Share this post


Link to post
Share on other sites
Posted (edited)
On 5/16/2020 at 3:38 PM, Floppy_Sock said:

Do you perceive a difference in contact size across the two images. If you do, where do you see the discrepancy start.

When I download the actual images and look at them on a 4K screen, the aircraft in both (4K and 1080p) look the same size, but the 4K ones look "larger" because they are more sharply defined and aren't blurring at the edges. Look how the F-4 at 1.05 fills the space between the F and the 4 in 4K

1080P.JPG

4k.JPG

Edited by SharpeXB

Share this post


Link to post
Share on other sites
Posted (edited)

I believe that's the same conclusion as the original post for the closer contacts.

Edited by [DBS]Browning

Share this post


Link to post
Share on other sites
4 hours ago, SharpeXB said:

When I download the actual images and look at them on a 4K screen, the aircraft in both (4K and 1080p) look the same size, but the 4K ones look "larger" because they are more sharply defined and aren't blurring at the edges. Look how the F-4 at 1.05 fills the space between the F and the 4 in 4K

1080P.JPG

4k.JPG

Yes, as @[DBS]Browning said, the close contacts scale correctly.

 

Could you repeat for the distant contacts? 

Share this post


Link to post
Share on other sites

The distant contacts, I don’t think they read right in a still screenshot. You can see in the above, one might alias out a piece at the instant you take it so it’s hard to say. And any image is affected by the screen you look at it on. I can’t see a native 1080x1920 image on my screen, it will appear soft and upscaled. The same image at the native res will appear sharp. 

Share this post


Link to post
Share on other sites
Posted (edited)
19 minutes ago, SharpeXB said:

The distant contacts, I don’t think they read right in a still screenshot. You can see in the above, one might alias out a piece at the instant you take it so it’s hard to say. And any image is affected by the screen you look at it on. I can’t see a native 1080x1920 image on my screen, it will appear soft and upscaled. The same image at the native res will appear sharp. 

 

You can compare the image pixel for pixel. That is a valid comparison. 

 

For example, you can look at the relative size of the contacts across the image. Not across different image. 

 

By that I mean, on the 1080p screen, the distant contact is 70% the size of the close one.

 

On a 4k screen, the distant contact is 30% the size of the close one. 

 

I'm truly baffled how you do not see this as an issue. This clearly indicative of a non-uniformity in scaling wouldn't you agree?

 

Just look at what happens when I play in 720p. The distant contacts are HUGE

720p_raw.thumb.png.c0e7ec52deee7d4c40de68855634285b.png

Edited by Floppy_Sock
  • Upvote 1

Share this post


Link to post
Share on other sites

At 720p the distant contacts are roughly the same size as the close ones! At 1080p the effect is almost as bad.

 

I'm not especially concerned about this, but I can certainly see the issue and it's problems.

It would take wilful blindness for me not to see the issue here.

 

Share this post


Link to post
Share on other sites

The biggest issue for me - honestly, is it's effect on supersampling in vr. 

 

I don't like the AA implementation in this game. I have much better performance supersampling. But you can see, if I use it, the size of contacts after sampling is noticeably smaller. 

 

 

Share this post


Link to post
Share on other sites
32 minutes ago, Floppy_Sock said:

The biggest issue for me - honestly, is it's effect on supersampling in vr. 

 

I don't like the AA implementation in this game. I have much better performance supersampling. But you can see, if I use it, the size of contacts after sampling is noticeably smaller.

 

Yeah, that's an undeniable problem; a bug even.

Share this post


Link to post
Share on other sites
10 hours ago, Floppy_Sock said:

 

You can compare the image pixel for pixel. That is a valid comparison. 

I feel like I couldn’t evaluate it correctly without native signals on a native display. And I can’t look at the non-native image because by definition it’s being upscaled just trying to view it on my screen. I know from playing this game on both 1080p and 4K screens there’s a difference between the perceived “sharpness” and resolution. 

Share this post


Link to post
Share on other sites
3 hours ago, SharpeXB said:

it’s being upscaled just trying to view it on my screen

 

It doesn't have to be upscaled. If you displayed it full screen, it would be upscaled of course, but you can display a 320x480p image on a 4k screen with a 1:1 pixel ratio. It will just not take up the full screen.

It would not necessarily be a different size either as there is no standard size for screens. Anything from 16inch to 31inch screens are common and some screens are outside those bounds. Besides, lower resolution screens tend to be smaller anyway.

 

However, if you are still struggling to evaluate such things, perhaps it's time to bow out of the thread and leave it to people who are able to evaluate such things.

 

 

 

  • Upvote 1

Share this post


Link to post
Share on other sites
4 hours ago, SharpeXB said:

I feel like I couldn’t evaluate it correctly without native signals on a native display. And I can’t look at the non-native image because by definition it’s being upscaled just trying to view it on my screen. I know from playing this game on both 1080p and 4K screens there’s a difference between the perceived “sharpness” and resolution. 

 

 

I think we should agree do disagree at this point. I’m clearly not getting through to you.  This is not a subjective argument. It can, and has, been done as rigorously as we can without proper testing tools. 

 

I appreciate your feedback regardless!

Share this post


Link to post
Share on other sites
1 hour ago, [DBS]Browning said:

 

It doesn't have to be upscaled. If you displayed it full screen, it would be upscaled of course, but you can display a 320x480p image on a 4k screen with a 1:1 pixel ratio. It will just not take up the full screen.

Sure. But if you’re going to count pixels it would be better to use a magnifying glass than try to resize the image. 

47 minutes ago, Floppy_Sock said:

 

I think we should agree do disagree at this point. 

I’m not really disagreeing with you, I just can’t confirm it for myself. 

Share this post


Link to post
Share on other sites
Posted (edited)
34 minutes ago, SharpeXB said:

Sure. But if you’re going to count pixels it would be better to use a magnifying glass than try to resize the image. 

I’m not really disagreeing with you, I just can’t confirm it for myself. 

 

You understand that the number of pixels in a screenshot relate directly to the number of pixels displayed on a screen when the image is displayed at 100% size, right?

Edited by [DBS]Browning

Share this post


Link to post
Share on other sites
Posted (edited)
31 minutes ago, [DBS]Browning said:

 

You understand that the number of pixels in a screenshot relate directly to the number of pixels displayed on a screen when the image is displayed at 100% size, right?

Right. But the 1080p version I’m looking at is half my screen size. Again, I suppose I could take a magnifying glass to it... but I can’t post that result to anyone but myself. 

Edited by SharpeXB

Share this post


Link to post
Share on other sites

Ok, I dunno why this is going back and forth. This is a fairly straightforward and scientific experiment. This is clearly showing an issue where contacts at range are not scaling to resolution like they should be. There should be render targets they can set/adjust. 

  • Upvote 3

Share this post


Link to post
Share on other sites
21 minutes ago, SharpeXB said:

Right. But the 1080p version I’m looking at is half my screen size. Again, I suppose I could take a magnifying glass to it... but I can’t post that result to anyone but myself. 

 

Display the image at 2x size. Four pixels per pixel. It will be the same as having pixels twice the size. 

Share this post


Link to post
Share on other sites
14 minutes ago, SCG_Wulfe said:

Ok, I dunno why this is going back and forth. This is a fairly straightforward and scientific experiment. This is clearly showing an issue where contacts at range are not scaling to resolution like they should be. There should be render targets they can set/adjust. 

 

 I love my 4k monitor, eventho i am definitely dying more...to see someone first rarely happens..

 

i just can not give up the immersion eye candy

 

Also; it is super easy to ID the contacts in 4k as i am going down in flames

  • Like 1
  • Haha 2

Share this post


Link to post
Share on other sites
22 minutes ago, SCG_Tzigy said:

 

 I love my 4k monitor, eventho i am definitely dying more...to see someone first rarely happens..

 

i just can not give up the immersion eye candy

 

Also; it is super easy to ID the contacts in 4k as i am going down in flames

Yeah the visual ID in range is much more important. I’m still not sure that distant objects are bigger in 1080p. I used to have a 1080p monitor and don’t recall noticing any big difference. Are people in 1080p seeing contacts 32km away? If they’re 4x the size then they should

Ive never been killed by a pixel sized bandit either. The ones who get me are close enough to be plenty visible. 
I don’t know why it’s such a big deal seeing targets that far away. In WWII combat they might as well be on the moon. Anyone in relevant range is plenty easy to see. 

Share this post


Link to post
Share on other sites
Posted (edited)

Here's my own test, a 3840x2160 and 1920x1080 (running windowed) this is with 4xAA

I don't know if I can see a difference. They lose a little in compression to .jpg though

2020_5_19__19_53_9.jpg

2020_5_19__19_49_44.jpg

Edited by SharpeXB

Share this post


Link to post
Share on other sites

Can you provide a download link for the raw screenshots? 

 

The 4K image is too big for the uploader here.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×
×
  • Create New...