Support NeoGAF

[Corporal.Hicks]

The big problem with this post is that the windmill you are using for comparison is right in the region where it should begin being occluded by fog, so the 'more detail' as an assertion of correctness is tenuous. I would argue they all look wrong. TAA is tonally correct but under-sampled resulting in broken aliased edges, but the other two are too soft, with the quality DLSS being too tonally dark.

Then in the CNN vs New Transformer image, without marketing labelling guiding the viewer to the right image, both images have as many advantages and disadvantages over the other.

As for the comparison between native and DLSS in the Tomb raider video, native is still superior by the reviewer's own comment about the the canopy artefacts in the DLSS render at the very first comparison. Can ML AI look better than native in stills? most definitely. Is native typically superior in motion and latency and generally correct but under sampled at worst? Yes. On that basis are we at the point of saying ML AI can be superior to native in motion over a full game play through? no, not universally.

DLSS image isnt too dark, you just dont know, what you dont know. In RDR2, the lighting and colors changes literally every few seconds. One second the sun is shining brightly. A few seconds later everything changes as the clouds move. When I was taking screenshots, I saw this difference.

DLSS at default settings arnt using as strong sharperning filter as TAA in this game, I said it very clearly. When I play the game, I disable the in-game DLSS sharpening and use my own reshade filters and the sharpness looks much better. The TAA image also looks better with my own reshade settings, but during motion everything blurs (it's like watching 1080p resolution during motion). In comparison, even DLSS performance looks much sharper during motion. Having tested every single AA method in this game, I have to say that DLSS destroys every other AA method and clearly offers the best image quality even in performance mode.

As for the Shadow Of The Tomb Raider video, watch it to the end. This reviewer showed examples where the DLSS 2.0 image had significantly more detail than the native TAA.

 

[PaintTinJr]

DLSS image isnt too dark, you just dont know, what you dont know. In RDR2, the lighting and colors changes literally every few seconds. One second the sun is shining brightly. A few seconds later everything changes as the clouds move. When I was taking screenshots, I saw this difference.

DLSS at default settings arnt using as strong sharperning filter as TAA in this game, I said it very clearly. When I play the game, I disable the in-game DLSS sharpening and use my own reshade filters and the sharpness looks much better. The TAA image also looks better with my own reshade settings, but during motion everything blurs (it's like watching 1080p resolution during motion). In comparison, even DLSS performance looks much sharper during motion. Having tested every single AA method in this game, I have to say that DLSS destroys every other AA method and clearly offers the best image quality even in performance mode.

As for the Shadow Of The Tomb Raider video, watch it to the end. This reviewer showed examples where the DLSS 2.0 image had significantly more detail than the native TAA.

For a start I didn't say colour, I said tone, and I was specifically talking about your cherry picked three way comparison example of the windmill in the back of the frustum. It is at a distance in which it should be tonally lighter than the DLSS quality option regardless of colour because of depth cueing, which DLSS gets wrong by defeating the tonal change by making it a higher contrasted piece of the image than the TAA original, and is adding back some details that should be lost to atmospheric depth cueing at that multi kilometre distance from the camera .

As for you using reshade, you wouldn't convince me, as I've seen what that can do, and for every improvement you think you are getting you almost certainly throwing away detail/fidelity, so by all means enjoy them, but probably best to not hold them up as an improvement.

As for the review showing examples that had significantly more detail than TAA, that's nice and all, but for DLSS to be better than native, it can't just be better in cherry picked examples, it has to at a minimum in all scenarios be equal to native first and foremost or better, which the example at the beginning exposes that it isn't.

ML AI up-scalers can't be held up as superior to native while introducing noise to superior results in the native image.

 

[Corporal.Hicks]

For a start I didn't say colour, I said tone, and I was specifically talking about your cherry picked three way comparison example of the windmill in the back of the frustum. It is at a distance in which it should be tonally lighter than the DLSS quality option regardless of colour because of depth cueing, which DLSS gets wrong by defeating the tonal change by making it a higher contrasted piece of the image than the TAA original, and is adding back detail that should be lost to atmospheric depth cueing at that multi kilometre distance from the camera .

As for you using reshade, you wouldn't convince me, as I've seen what that can do, and for every improvement you think you are getting you almost certainly throwing away detail/fidelity, so by all means enjoy them, but probably best to not hold them up as an improvement.

As for the review showing examples that had significantly more detail than TAA, that's nice and all, but for DLSS to be better than native, it can't just be better in cherry picked examples, it has to at a minimum in all scenarios be equal to native first and foremost or better, which the example at the beginning exposes that it isn't.

ML AI up-scalers can't be held up as superior to native while introducing noise to superior results in the native image.

Both native TAA and DLAA/DLSS need sharpening to look reasonably sharp, as strong temporal AA softens the contrast of the edges (and the human eye perceives sharpness as contrast). Most developers use their own sharpening filters, but I prefer to use reshade with my own settings because I can always find the perfect balance between oversharpened and soft image, whereas developers will either oversharpen the image, or leave it too soft.

BTW. TAA in RDR2 at default settings has a little bit stronger sharpening than DLSS.

I don't know how you can say I'm cherry picking when this reviewer has shown several examples of the improvements DLSS 2.0 has over native TAA. He said DLSS has supperior detailing on the distant buildings, textures and trees. He also showed examples of where three branches were missing on the TAA, even at close range. DLSS has also clear advantage in motion. When I played the game, even DLSS 2.0 was clearly superior to TAA, which to my eyes was twice as blurry in motion. Even 2.0 DLSS looked better to me than TAA and now we have 3.8.1 DLSS.

Dude, you say I'm cherrypicking, but I remember your posts and I know you're the grand master of cherrypicking . You looked at hypothetical performance estimates on the wikipedia for PS2 console and tried to tell me that Sony's console was faster at rendering polygons than Xbox. I had to show you internal Sony documents to shut you up. Real PS2 games did not use nearly as many polygons.

 

[b0bbyJ03]

The stacked cache bandwidth discussion tells me it has already won out because when 1ms is the fraction of utilisation the silicon gets in a 8ms, 16ms or 33ms frame and to divide and conquer the bandwidth requirement, and handle the latency constraints the CNNs need to tile across WGP caches it just sounds counter productive in terms of design to be building more powerful, more expensive and more wasteful ML accelerators with more wasteful highspeed interconnects when you can generalise the functionality with WGPs and a more complex software interface.

Nvidia are going big on frame-gen but in the end they are still stuck with 33ms latency(optimal 30fps double buffered latency) from generating fake frames even when the frame-rate is north of 200fps, and at that frame rate the latency should be approaching 1-5ms and have motion fidelity feedback latency of 1-5ms. As the performance of the WGPs eventually allow for higher frame-rates, the ML benefit and the benefit of low latency with native high frame-rate should be delivered together on Sony/AMD because the WGP parallelism will give more stacked bandwidth and more TOPs per ms at a granular level to efficiently interleave with standard rendering which will grow in bandwidth and scale proportionally.

idk what any of this means, but it sounds incredible!

 

[PaintTinJr]

Both native TAA and DLAA/DLSS need sharpening to look reasonably sharp, as strong temporal AA softens the contrast of the edges (and the human eye perceives sharpness as contrast). Most developers use their own sharpening filters, but I prefer to use mine because I can always find the perfect balance between oversharpened and soft image.

BTW. TAA in RDR2 at default settings has stronger sharpening than DLSS.

I don't know how you can say I'm cherry picking when this reviewer has shown several examples of the improvements DLSS 2.0 has over native TAA. He said DLSS has supperior detailing on the distant buildings, textures and trees. He also showed examples of where three branches were missing on the TAA, even at close range. DLSS has also clear advantage in motion. When I played the game, even DLSS 2.0 was clearly superior to TAA, which to my eyes was twice as blurry in motion. Even 2.0 DLSS looked better to me than TAA and now we have 3.8.1 DLSS.

Dude, you say I'm cherrypicking, but I remember your posts and I know you're the grand master of cherrypicking . You looked at hypothetical performance estimates on the wikipedia for PS2 console and tried to tell me that Sony's console was faster at rendering polygons than Xbox. I had to show you internal Sony documents to shut you up. Real PS2 games did not use nearly as many polygons.

If you prefer your sharpening and your version of how a previous conversation played out, that's all good, you do you.

But with regards to this topic at hand, it feels like you automatically think more detail is the correct option, when game rendering goes to great lengths to model natural optical and lighting phenomena which DLSS is wrongly defeating in places or you are defeating with more different sharpening methods. So I doubt we can have this conversation if your stance is: "I like what I like and that's the way I like it!"

 

[Corporal.Hicks]

If you prefer your sharpening and your version of how a previous conversation played out, that's all good, you do you.

But with regards to this topic at hand, it feels like you automatically think more detail is the correct option, when game rendering goes to great lengths to model natural optical and lighting phenomena which DLSS is wrongly defeating in places or you are defeating with more different sharpening methods. So I doubt we can have this conversation if your stance is: "I like what I like and that's the way I like it!"

I cant take screenshots in this game with reshade, to show you how big difference reshade can make, because RDR2 is using vulkan, and DX12 API doesnt work on my PC (for some strange reason).

I can however show what difference reshade can make in BMW.

4K DLAA, default ingame sharpening settings

Reshade with my own sharpening settings

Without sharpening

IMO the game has too strong sharpening at default settings, I can see ringing, and ant like noise. Why my reshade settings there's none of that, but the image still look reasonably sharp. Without sharpening mask at all however the image is too soft despite running the game at 100% resolution scale. You want to tell me you would like to play with such blurry image, without any sharpening mask? Not even my old CRT has such blurry picture.

I'm not going to play this game at 4K DLAA anyway because my GPU is too weak for that, however DLSSP with my reshade settings still offers very good image quality to my eyes (far better than the PS5 / Pro versions), and the game is very playable thanks to DLSS (around 55fps and with FG on top of that around 80-100fps). I cant wait to test new DLSS (transformel model), because it should improve image quality even further.

 

[PaintTinJr]

I cant take screenshots in this game with reshade, to show you how big difference reshade can make, because RDR2 is using vulkan, and DX12 API doesnt work on my PC (for some strange reason).

I can however show what difference reshade can make in BMW.

4K DLAA, default ingame sharpening settings

Reshade with my own sharpening settings

Without sharpening

IMO the game has too strong sharpening at default settings, I can see ringing, and ant like noise. Why my reshade settings there's none of that, but the image still look reasonably sharp. Without sharpening mask at all however the image is too soft despite running the game at 100% resolution scale. You want to tell me you would like to play with such blurry image, without any sharpening mask? Not even my old CRT has such blurry picture.

I'm not going to play this game at 4K DLAA anyway because my GPU is too weak for that, however DLSSP with my reshade settings still offers very good image quality to my eyes (far better than the PS5 / Pro versions), and the game is very playable thanks to DLSS (around 55fps and with FG on top of that around 80-100fps). I cant wait to test new DLSS (transformel model), because it should improve image quality even further.

I don't know what you think I'm supposed to find better in the images you want me to like, but from a quick cursory look at them in Gimp, as expected the unsharpened image has the most fidelity with the fullest version of the histogram, and even after adjusting them all edited with their own duplicated layer, that is then high pass filtered, with the same settings, and layer grain merged with the original layer the unsharpened source image still wins out for clarity and still has the most detail in its histogram post-edit

So I don't really see much between the three images, even though I prefer the unsharpened image, and a numerical histogram analysis and comparison of them all offline enhanced and compared still supports that. So.....

 

[Zathalus]

I cant take screenshots in this game with reshade, to show you how big difference reshade can make, because RDR2 is using vulkan, and DX12 API doesnt work on my PC (for some strange reason).

I can however show what difference reshade can make in BMW.

4K DLAA, default ingame sharpening settings

Reshade with my own sharpening settings

Without sharpening

IMO the game has too strong sharpening at default settings, I can see ringing, and ant like noise. Why my reshade settings there's none of that, but the image still look reasonably sharp. Without sharpening mask at all however the image is too soft despite running the game at 100% resolution scale. You want to tell me you would like to play with such blurry image, without any sharpening mask? Not even my old CRT has such blurry picture.

I'm not going to play this game at 4K DLAA anyway because my GPU is too weak for that, however DLSSP with my reshade settings still offers very good image quality to my eyes (far better than the PS5 / Pro versions), and the game is very playable thanks to DLSS (around 55fps and with FG on top of that around 80-100fps). I cant wait to test new DLSS (transformel model), because it should improve image quality even further.

Not sure what Game Science was thinking with default sharpening pass they have, it looks horrendous. Your reshade preset looks great so well done on that. The no sharpening image just goes to show, that despite the benefits, TAA absolutely murders fine detail. Even DLAA struggles with fixing that.

 

[FireFly]

I cant take screenshots in this game with reshade, to show you how big difference reshade can make, because RDR2 is using vulkan, and DX12 API doesnt work on my PC (for some strange reason).

I can however show what difference reshade can make in BMW.

I think it would have been more interesting to just present the images without specifying whether reshade had been applied to them, and ask which looks better. Since that makes it a lot harder to respond in bad faith by inventing random reasons full of undefined jargon to "explain" why the original image is better.

 

[Corporal.Hicks]

Not sure what Game Science was thinking with default sharpening pass they have, it looks horrendous. Your reshade preset looks great so well done on that. The no sharpening image just goes to show, that despite the benefits, TAA absolutely murders fine detail. Even DLAA struggles with fixing that.

I completely agree. Unfortunately, both TAA / DLAA softens the image and there are only two ways to combat this problem, either use a sharpening mask or downsample. Of course, you can also use both methods at the same time to get a clean but also razor-sharp image. It will be interesting to see if the new DLSS improves natural sharpness.

I never used sharpening when I was playing games with MSAA, or even SMAA, but I absolutely cannot stand TAA blur. What's more, some games like Hellblade 2 take the TAA blur to another level. I tried to play Hellblade 2 with downampling and the image still looked a little soft, that's how blurry this game is. Luckily I found some ini parameters to remove the ugly blur. People buy high resolution monitors and want to play at ultra high resolutions to get sharper and more detailed image, but some developers still think people want to see blur .

I absolutely hate TAA blur. I have never seen such soft edges in real life, not even on my photos, so I cannot explain why some people prefer soft and blurry images. Here's one of my photos, and I can't wait to hear comments like "real life doesn't look this sharp" because it absolutely does , at least my eyes can see that sharp.

 

[sendit]

Currently have a 4090. I think I'm settling for a 5080. 120 FPS is my target at 4K (DLSS). This is dependent on how multi-frame gen works in the real world. Lossless scaling on anything above 2x has way too many artifacts.

 

[Bojji]

Currently have a 4090. I think I'm settling for a 5080. 120 FPS is my target at 4K (DLSS). This is dependent on how multi-frame gen works in the real world. Lossless scaling on anything above 2x has way too many artifacts.

Why you want worse GPU?

 

[sendit]

Why you want worse GPU?

Multi-frame gen. Less component heat output > raw frames.

 

[Corporal.Hicks]

Currently have a 4090. I think I'm settling for a 5080. 120 FPS is my target at 4K (DLSS). This is dependent on how multi-frame gen works in the real world. Lossless scaling on anything above 2x has way too many artifacts.

Unfortunatley DLSS FG x4 also has artefacts. Linus showed them in his video, and also digital foundry.

x2 FG is perfectly usable, but anything higher is big no no, and if people want to know why I recommend to watch Daniel Owens video:

 

[sendit]

Unfortunatley DLSS FG x4 also has artefacts. Linus showed them in his video, and also digital foundry.

x2 FG is perfectly usable, but anything higer is big no no, and if people want to know why I recommend to watch Daniel Ovens video:

 

[Bojji]

Multi-frame gen. Less component heat output > raw frames.

I get that, BUT:

- you can undervolt/power limit 4090
- 24GB of VRAM will become handy at some point
- you still have 2x FG that will be improved with update

More fake frames = higher chance of visible artifacts. But of course we will see what MFG is worth with reviews.

 

[3liteDragon]

Do we have any idea when DLSS4 is going to to be available?

When the 5090 launches, so Jan 30th.

 

[PaintTinJr]

I think it would have been more interesting to just present the images without specifying whether reshade had been applied to them, and ask which looks better. Since that makes it a lot harder to respond in bad faith by inventing random reasons full of undefined jargon to "explain" why the original image is better.

It is easy for you to say it is jargon, but here is those three images to draw out whatever detail is in each of them, with all getting the same sharpening and no alteration in any other way - apart from all consistently downsized to 1080p to allow me to upload them to GAF.

So A) which has the best clarity? B) which has the best contrast? C) which is the image that best represents that scene data?

 

[sendit]

I get that, BUT:

- you can undervolt/power limit 4090
- 24GB of VRAM will become handy at some point
- you still have 2x FG that will be improved with update

More fake frames = higher chance of visible artifacts. But of course we will see what MFG is worth with reviews.

My 4090 is already undervolted in a SSF case. If Multi-frame gen can generate 8x the frame rate with minimal artifacting, I would much prefer that over raw frames with more power draw.

 

[Corporal.Hicks]

It is easy for you to say it is jargon, but here is those three images to draw out whatever detail is in each of them, with all getting the same sharpening and no alteration in any other way - apart from all consistently downsized to 1080p to allow me to upload them to GAF.

So A) which has the best clarity? B) which has the best contrast? C) which is the image that best represents that scene data?

Is this a joke? You destroyed the original detail by resizing my screenshots and compressing everything beyond belief, and then as if that wasn't good enough, you ruined the image with an insane amount of sharpening mask. I know what you are doing dude, obviously if you apply the same amount of sharpness to three images with different amount of sharpness, images that had previously perfect amount of edge contrast (sharpness) will be now oversharpened, whereas the most blurry one will look the best because it will have the least amount of excessive sharpening. There's nothing unsurprising in this comparison dude. You have done everything you could to MISLEAD people that the worst image look the best.

 

[kevboard]

It is easy for you to say it is jargon, but here is those three images to draw out whatever detail is in each of them, with all getting the same sharpening and no alteration in any other way - apart from all consistently downsized to 1080p to allow me to upload them to GAF.

So A) which has the best clarity? B) which has the best contrast? C) which is the image that best represents that scene data?

you can't resize an image, upload it as a freaking JPEG to imgur of all places, and expect anyone to be able to tell make any real judgements about these

 

[PaintTinJr]

Is this a joke? You destroyed the original detail by resizing my screenshots and compressing everything beyond belief, and then as if that wasn't good enough, you ruined the image with an insane amount of sharpening mask. I know what you are doing dude, obviously if you apply the same amount of sharpness to three images with different amount of sharpness, images that had previously perfect amount of edge contrast (sharpness) will be now oversharpened, whereas the most blurry one will look the best because it will have the least amount of excessive sharpening. There's nothing unsurprising in this comparison dude. You have done everything you could to MISLEAD people that the worst image look the best.

No, you are being very disingenuous, and I would have upload at the 7MB original sizes for each image if GAF would let me upload them. The idea that for comparison 1080p is vastly inferior to 4K is a clear example of you being disingenuous, and no, I didn't over sharpen at all. I used a tiny std dev of just 1.7 in the high pass filter.

You clearly have no idea if you think I applied the same 'sharpening' to each image. No sharpening was actually done, each image was merly enhanced by its own high pass filtered data, added as a subtle grain merge to give proportional clarity to its existing discontinuities.

 

[Corporal.Hicks]

No, you are being very disingenuous, and I would have upload at the 7MB sizes altered sizes for each image if GAF would let me upload them. The idea that for comparison 1080p is vastly inferior to 4K is a clear example of you being disingenuous, and no, I didn't over sharpen at all. I used a tiny std dev of just 1.7 in the high pass filter.

You clearly have no idea if you think I applied the same 'sharpening' to each image. No sharpening was actually done, each image was merly enhanced by its own high pass filtered data, added as a subtle grain merge to give proportional clarity to its existing discontinuities.

Dont BS me. It's very easy to upload screenshots without reducing resolution and image quality, and if you didn't know that, you're dumber than I thought.

Whatever you done to my screenshots, they all look oversharpened now, and obviously screenshot that had the least amount of oversharpening look the best now (unsurprisingly).

What you have done in this dissusion is beyond pathetic. First you tried to tell people that the mill example with the missing details (TAA image) looked the best because there was supposed to be some kind of fog there according to your vivid imagination, so it made sense that the mill was visible only in half . Then you tried to tell people that the PCworld reviewer thought he preferred the TAA image, even though he found more imperfections in TAA images (it seems you havent even watched entire video).

Dude, I'm not surprised by your disingenuous behaviour. I remember your comments and you did the same thing in our argument about PS2 vs Xbox when you tried to present facts in such a way to make the inferior console superior. You wanted to tell people that the PS2 hardware could render the same effects in software and outperform the more powerful console in polycounts even though it was RAM limited and you couldnt even grasp that you need memory even to be able to use higher polycounts. Even developers who made games on all platforms said that Xbox had the upper hand when it came to hardware and polycounts, whereas the PS2 has the upperhand in other areas (fillrate) but you still thought Sony's console was superior. I had to link internal reports done by sony to show people the real poly counts of PS2 games and the numbers were nowhere near what you claimed and that finally shut you up.

I think I have wasted enough of my time on you, but enough is enough. This is my last reply to you. If you wish to entertain others with your vivid imagination, be my guest, but I'm out .

 

[mckmas8808]

nvidia is 3 (frames) generation ahead of competition

Nvidia is like Tesla, multiple gens and years ahead of everybody else that it's insane. Everybody else will be okay, but WOW!

 

[mckmas8808]

He’s most likely the person who will create Skynet…
But yeah I think he’s currently the absolute top dog in the tech world. He’s owning the stage at events like nobody else. At one time everybody listened to Steve Jobs, some watched Apple’s presentations even though they didn’t plan to buy any of their products. Now it’s Jensen and Nvidia.

You're forgetting.....

 

[Fess]

You're forgetting.....

Lol no I can’t stand him but I included him in another post speculating about who in the corporate elite who’ll most likely start their villain arch.

 

[mckmas8808]

Lol no I can’t stand him but I included him in another post speculating about who in the corporate elite who’ll most likely start their villain arch.

Jensen and Elon are in love with one another. I honestly believe they are going to team up alot more in the future.

 

[DoubleClutch]

He’s most likely the person who will create Skynet…
But yeah I think he’s currently the absolute top dog in the tech world. He’s owning the stage at events like nobody else. At one time everybody listened to Steve Jobs, some watched Apple’s presentations even though they didn’t plan to buy any of their products. Now it’s Jensen and Nvidia.

Leather jacket wishes he was Steve Jobs.

 

[llien]

He’s most likely the person who will create Skynet…
But yeah I think he’s currently the absolute top dog in the tech world. He’s owning the stage at events like nobody else. At one time everybody listened to Steve Jobs, some watched Apple’s presentations even though they didn’t plan to buy any of their products. Now it’s Jensen and Nvidia.

Nobody cares what Huang J. thinks outside of GPU/data center GPU world.

On top of it, a lot of things coming from his mouth is an echo of unhinged NV marketing.

He's more of a popular clown than influential techie.

That, and I've never listened to SJ, or watched Birne's presentations.

 

[Zathalus]

Dont BS me. It's very easy to upload screenshots without reducing resolution and image quality, and if you didn't know that, you're dumber than I thought.

Whatever you done to my screenshots, they all look oversharpened now, and obviously screenshot that had the least amount of oversharpening look the best now (unsurprisingly).

What you have done in this dissusion is beyond pathetic. First you tried to tell people that the mill example with the missing details (TAA image) looked the best because there was supposed to be some kind of fog there according to your vivid imagination, so it made sense that the mill was visible only in half . Then you tried to tell people that the PCworld reviewer thought he preferred the TAA image, even though he found more imperfections in TAA images (it seems you havent even watched entire video).

Dude, I'm not surprised by your disingenuous behaviour. I remember your comments and you did the same thing in our argument about PS2 vs Xbox when you tried to present facts in such a way to make the inferior console superior. You wanted to tell people that the PS2 hardware could render the same effects in software and outperform the more powerful console in polycounts even though it RAM limited. Even developers who made games on all platforms said that Xbox had the upper hand when it came to hardware and polycounts, whereas the PS2 has the upperhand in other areas (fillrate) but you still thought Sony's console was superior. I had to link internal reports done by sony to show people the real poly counts of PS2 games and the numbers were nowhere near what you claimed and that finally shut you up.

I think I have wasted enough of my time on you, but enough is enough. This is my last reply to you. If you wish to entertain others with your vivid imagination, be my guest, but I'm out .

Your original images actually each contain the same amount of detail, as sharpening doesn’t actually add any detail (how could it?), all it does is makes sure edges are more defined because our eyes see sharpness as contrast (as you rightfully pointed out). So anybody editing those screenshots and then claiming X is objectively superior is flat out wrong, as each image actually contains the same amount of detail. I'm not really sure what the intent behind that is, sharpening an already sharpened image and then blending with the original image would obviously be best on the soft TAA image. That doesn't mean the soft image is visually more pleasing, unless everyone walks around with auto-GIMP in their eyes. Most people would agree that adding some sharpening to a TAA based image does enhance some detail, due to the contrast difference you mentioned before, but that is a subjective opinion on how each person views such things. I like some subtle sharpening myself, as you did yourself with reshade. Untreated TAA is simply way too soft, and oversharpening is bad as well.

 

[PaintTinJr]

Your original images actually each contain the same amount of detail, as sharpening doesn’t actually add any detail (how could it?), all it does is makes sure edges are more defined because our eyes see sharpness as contrast (as you rightfully pointed out). So anybody editing those screenshots and then claiming X is objectively superior is flat out wrong, as each image actually contains the same amount of detail. I'm not really sure what the intent behind that is, sharpening an already sharpened image and then blending with the original image would obviously be best on the soft TAA image. That doesn't mean the soft image is visually more pleasing, unless everyone walks around with auto-GIMP in their eyes. Most people would agree that adding some sharpening to a TAA based image does enhance some detail, due to the contrast difference you mentioned before, but that is a subjective opinion on how each person views such things. I like some subtle sharpening myself, as you did yourself with reshade. Untreated TAA is simply way too soft, and oversharpening is bad as well.

You clearly didn't read my original comment in which the histograms of his images are not equal, because he's scrubbed data with his reshade nonsense. If you don't believe me, go check his original images yourself and compare the histograms in which the unsharpened one contains the most detail even though the file size is the smallest of the three at just 3.8MB.

And at no point did I say the enhancement added detail, I said the resulting histogram still contained the most detail after editing. A process merely used to expose that just because the unsharpened image was soft didn't mean it was inferior in stored detail, quite the reverse actually and it contains the most information/bandwidth/fidelity both in detail and luminance for a TV/monitor to display.

Corporal.Hicks might prefer his option - you probably do too - but it is an inferior degraded signal, and that is a point of fact.

 

[Zathalus]

You clearly didn't read my original comment in which the histograms of his images are not equal, because he's scrubbed data with his reshade nonsense. If you don't believe me, go check his original images yourself and compare the histograms in which the unsharpened one contains the most detail even though the file size is the smallest of the three at just 3.8MB.

And at no point did I say the enhancement added detail, I said the resulting histogram still contained the most detail after editing. A process merely used to expose that just because the unsharpened image was soft didn't mean it was inferior in stored detail, quite the reverse actually and it contains the most information/bandwidth/fidelity both in detail and luminance for a TV/monitor to display.

Corporal.Hicks might prefer his option - you probably do too - but it is an inferior degraded signal, and that is a point of fact.

All reshade does is add a sharpening filter that simply enhances edge contrast. No detail is added nor is any detail removed. It’s a simple processing trick that appears to make an image more detailed as localised contrast like that tricks the human eye into thinking there is more detail. It’s a simply pass to make the final image more pleasing to most people. Regular TAA does it, DLSS does it, PSSR does it, and even a games own post processing past can do it.

 

[PaintTinJr]

All reshade does is add a sharpening filter that simply enhances edge contrast. No detail is added nor is any detail removed. It’s a simple processing trick that appears to make an image more detailed as localised contrast like that tricks the human eye into thinking there is more detail. It’s a simply pass to make the final image more pleasing to most people. Regular TAA does it, DLSS does it, PSSR does it, and even a games own post processing past can do it.

The composited image (capture) he uploaded is permanently degraded by reshade's scrubbing of detail, this is a point of fact you can check in the histogram versus the superior unsharpened image.

Whether people like TV shop mode effects like reshade really has no bearing on its lack of value and detriment to signal processing.

 

[Zathalus]

The composited image (capture) he uploaded is permanently degraded by reshade's scrubbing of detail, this is a point of fact you can check in the histogram versus the superior unsharpened image.

Whether people like TV shop mode effects like reshade really has no bearing on its lack of value and detriment to signal processing.

Sharpening passes like Filmic Anamorphic Sharpening or AMD’s Contrast Adaptive Sharpening enhance edges via a localised contrast boost. They are actually similar to the high pass filter in GIMP you mentioned earlier, just in real time.

What actually scrubs detail is the underlying temporal scaler and/or anti-aliasing. If “signal processing” is truly that important to you then you should be championing games with no anti-aliasing at all, as TAA does far, far more damage to the underlying detail compared to any impact a mere sharpening pass can do. Sharpening alters the TAA image to restore edge contrast, TAA is the technique that brutally murdered it in the first place.

 

[PaintTinJr]

Sharpening passes like Filmic Anamorphic Sharpening or AMD’s Contrast Adaptive Sharpening enhance edges via a localised contrast boost. They are actually similar to the high pass filter in GIMP you mentioned earlier, just in real time.

What actually scrubs detail is the underlying temporal scaler and/or anti-aliasing. If “signal processing” is truly that important to you then you should be championing games with no anti-aliasing at all, as TAA does far, far more damage to the underlying detail compared to any impact a mere sharpening pass can do. Sharpening alters the TAA image to restore edge contrast, TAA is the technique that brutally murdered it in the first place.

You are moving the goal posts because reshade damages image quality, and I wasn't advocating for a high pass filter as a artistic choice solution for this game, merely using it to disprove that the unsharpened image had inferior information.

As for championing no AA in games, it is a false equivalency with using reshade/sharpening, and I'm guessing you don't understand how rasterization at a fundamental level works if you would think that was an option.

Rasterization dithers edges across pixel boundaries even in a pure super sampling solution so AA is baked into rasterization so why would anyone be okay with rasterization dithering, but be against techniques designed to remove the artefacts of rasterization that help get the rendered representation closer to the geometrical description of the scene?

 

[Zathalus]

You are moving the goal posts because reshade damages image quality, and I wasn't advocating for a high pass filter as a artistic choice solution for this game, merely using it to disprove that the unsharpened image had inferior information.

As for championing no AA in games, it is a false equivalency with using reshade/sharpening, and I'm guessing you don't understand how rasterization at a fundamental level works if you would think that was an option.

Rasterization dithers edges across pixel boundaries even in a pure super sampling solution so AA is baked into rasterization so why would anyone be okay with rasterization dithering, but be against techniques designed to remove the artefacts of rasterization that help get the rendered representation closer to the geometrical description of the scene?

Here we go again, ad hominem attacks, straw man fallacies, and obfuscation. Same playbook every single time. I’m out.

 

[PaintTinJr]

Here we go again, ad hominem attacks, straw man fallacies, and obfuscation. Same playbook every single time. I’m out.

Pretty sure I haven't done any of that. Want to substantiate each of those allegations so I can fix the offending sentences, so you can actually reply to the technical aspect of my comment?

 

[Zathalus]

Pretty sure I haven't done any of that. Want to substantiate each of those allegations so I can fix the offending sentences, so you can actual reply to the technical aspect of my comment?

“ As for championing no AA in games, it is a false equivalency with using reshade/sharpening, and I'm guessing you don't understand how rasterization at a fundamental level works if you would think that was an option.” - Clear ad hominem attack, instead of addressing the content of my argument, you are trying to focus on or claim I don’t have the required knowledge for this argument. It also ties into a straw man argument, trying to spin as if I claimed that no AA was an option. I never once claimed it was, I merely pointed out that if the supposed signal quality of an image was that important, TAA is responsible for a far more egregious drop in image quality.

As for obfuscation, your last paragraph is a clear example of this. It lacks clarity, instead burying the underlying point (that AA is supposed to more accurate reflect the underlying geometry) under a convoluted structure. Maybe this is unintentional, but it is not easy to parse at first glance.

I’m not going to argue the underlying point about sharpening any further.

 

[PaintTinJr]

“ As for championing no AA in games, it is a false equivalency with using reshade/sharpening, and I'm guessing you don't understand how rasterization at a fundamental level works if you would think that was an option.” - Clear ad hominem attack, instead of addressing the content of my argument, you are trying to focus on or claim I don’t have the required knowledge for this argument. It also ties into a straw man argument, trying to spin as if I claimed that no AA was an option. I never once claimed it was, I merely pointed out that if the supposed signal quality of an image was that important, TAA is responsible for a far more egregious drop in image quality.

As for obfuscation, your last paragraph is a clear example of this. It lacks clarity, instead burying the underlying point (that AA is supposed to more accurate reflect the underlying geometry) under a convoluted structure. Maybe this is unintentional, but it is not easy to parse at first glance.

I’m not going to argue the underlying point about sharpening any further.

Most of what was in the comment I last replied to was bolted on to the discussion by you, and yes it felt like rather than just acknowledging my original point about Corporal.Hicks' pictures was both genuine and factual, it felt like you were trying to segue the things I'd touched on into some stance I was to defend. And you did specifically press the issue of AA or no AA, which I drew a wrong conclusion from, which wasn't intended to be an attack but merely a fast track to align our discussion of the issue, and was tipped by you wrongly using sharpening and AA interchangeably, so it wasn't unreasonable to conclude that you might have not appreciated the difference, or the origins of AA, which you clearly do from your reaction.

 

[OmegaSupreme]

God bless Nvidia

 

[PaintTinJr]

..First you tried to tell people that the mill example with the missing details (TAA image) looked the best because there was supposed to be some kind of fog there according to your vivid imagination, so it made sense that the mill was visible only in half.....

Here is your own image with real detail lost to atmospheric depth cueing at that multi kilometre distance capture with infinitesimally small photons passing through extremely small molecules and captured a orders more fidelity than a 4K render game and this is what happens to solid parts of tree branches when quantised at that range.

Does the windmill render in DLSS quality mode mirror that? Or does that real photo support my original point?

 

[Corporal.Hicks]

Here is your own image with real detail lost to atmospheric depth cueing at that multi kilometre distance capture with infinitesimally small photons passing through extremely small molecules and captured a orders more fidelity than a 4K render game and this is what happens to solid parts of tree branches when quantised at that range.

Does the windmill render in DLSS quality mode mirror that? Or does that real photo support my original point?

Obviously you have no idea how camera and lenses work. You cant capture everything clearly if you focus on the foreground objects (due to DOF). If I were to point my camera at this tree, it would not be out of focus.

Also photo will never look good when viewed at 1:1 (the camera sensor is never perfect and neither is the lens), yet alone when zoomed in so much. That's the original 1:1 crop. You had to resize my image multiple times to find some imperfections (chromatic aberration).

Dude, with every post you make a bigger idiot out of yourself . Cant you see that? You should ar least know when to stop.

 

[FireFly]

Here is your own image with real detail lost to atmospheric depth cueing at that multi kilometre distance capture with infinitesimally small photons passing through extremely small molecules and captured a orders more fidelity than a 4K render game and this is what happens to solid parts of tree branches when quantised at that range.

Does the windmill render in DLSS quality mode mirror that? Or does that real photo support my original point?

Your image has what looks like compression artifacts on the edges. Nevertheless the core structure of the tree branches seems to be present, just like on the DLSS Quality mode image. By contrast the shape of windmill seems to be distorted in the native TAA image, and the frame connecting the windmill to the vane is almost entirely gone. We don't see the same level of detail loss on the frame of the structure itself, despite the fact that it looks to be about the same thickness.

 

[JohnnyFootball]

No one has ever said that DLSS is flawless, only that it is impressive and getting better and better.

I hate the fact that nvidia has such a monopoly, but that monopoly is mostly earned by being the best.

I like DLSS so much, that I often use performance mode instead of balanced or quality since the visual impact is minor and I get to take better advantage of my 240 Hz QD-OLED.

I do have very high hopes for FSR4. My bazzite living room PC uses Bazzite (SteamOS) and since nvidia support on such a system is a ways off I would love for FSR4 to come within the ballpark of DLSS.

 

[PaintTinJr]

Your image has what looks like compression artifacts on the edges. Nevertheless the core structure of the tree branches seems to be present, just like on the DLSS Quality mode image. By contrast the shape of windmill seems to be distorted in the native TAA image, and the frame connecting the windmill to the vane is almost entirely gone. We don't see the same level of detail loss on the frame of the structure itself, despite the fact that it looks to be about the same thickness.

It isn't my image. It is a zoomed unedited subsection of Corporal.Hicks' jpg photo to show even his own photo taken with a camera in RL - that his RDR2 shots are in effect trying to mimic - exhibit atmospheric depth cueing and under sample.

So like I said originally the windmill shots all had issues, with the DLSS quality shot getting the windmill tonally wrong, and it recovers detail that should have been lost to depth cueing, illustrating that adding in detail with ML AI reconstruction isn't automatically 'better' than native as Corporal.Hicks has espoused.

 

[FireFly]

So like I said originally the windmill shots all had issues, with the DLSS quality shot getting the windmill tonally wrong, and it recovers detail that should have been lost to depth cueing, illustrating that adding in detail with ML AI reconstruction isn't automatically 'better' than native as Corporal.Hicks has espoused.

I guess the developers must have specially placed a cloud of magical fog that cuts off most of the frame connecting the vane to the windmill – leaving floating pixels – but retains the frame of structure itself.

 

[PaintTinJr]

I guess the developers must have specially placed a cloud of magical fog that cuts off most of the frame connecting the vane to the windmill – leaving floating pixels – but retains the frame of structure itself.

I have no idea, every other interaction I have with you and Corporal.Hicks are reading his or your childish verbal insults or some strawman - like this one. You know the 'A' in GAF is for adult, yeah?

All three images of the windmill had problems, I think that's the third time I've made the same statement, so where is the quote of me saying otherwise like you are inferring?

 

[FireFly]

I have no idea, every other interaction I have with you and Corporal.Hicks are reading his or your childish verbal insults or some strawman - like this one. You know the 'A' in GAF is for adult, yeah?

All three images of the windmill had problems, I think that's the third time I've made the same statement, so where is the quote of me saying otherwise like you are inferring?

I already made a good faith attempt to engage with your argument, by pointing out that the detail loss you highlighted in the photo occurs at the edge of the branches, leaving the core structures intact. Where as the detail loss in the TAA image compromises the actual core structure (the frame connecting the vane to the windmill and the struts of the windmill itself).

Rather than respond to me by identifying the actual details that you think shouldn't have been retained in the DLSS Quality image, you simply restated your position, without providing any further clarification. So yes, my statement may be a strawman, but since your position seems to be essentially mysterious and resistant to clarification, there is no way of knowing.

 

[PaintTinJr]

I already made a good faith attempt to engage with your argument, by pointing out that the detail loss you highlighted in the photo occurs at the edge of the branches, leaving the core structures intact. Where as the detail loss in the TAA image compromises the actual core structure (the frame connecting the vane to the windmill and the struts of the windmill itself).

Rather than respond to me by identifying the actual details that you think shouldn't have been retained in the DLSS Quality image, you simply restated your position, without providing any further clarification. So yes, my statement may be a strawman, but since your position seems to be essentially mysterious and resistant to clarification, there is no way of knowing.

Okay fine, I'm happy to draw a line under you starting our discussion in this thread like below and thumbing up Corporal.Hicks childish verbal insults if you actual want to discuss.

I think it would have been more interesting to just present the images without specifying whether reshade had been applied to them, and ask which looks better. Since that makes it a lot harder to respond in bad faith by inventing random reasons full of undefined jargon to "explain" why the original image is better.

So discussion your points about the quality DLSS windmill.

In my view, in the RL photo the depth cueing foliage/trees are about half or a third of the effective distance in the background compared to the distance of the windmill in RDR2, so tonally the whole thing should be far more depth cued in all three renders and starting to lose its strong outline far more, similar to the bright green foliage in the RL zoom photo subsection to the right of the roof vent is.

Is that partly a draw distance issue for accurate depth cue/fog control with RDR2's engine being a PS4 game, so probably can't use a frustum cascade to provide far more precision, but that still begs the question why doesn't DLSS quality mode intelligently fix that tonally and by coverage when inferencing the windmill and the other objects proportionally at distance? Which I believe was a big difference and done slightly better using PSSR in R&C vs DLSS.

Regardless of render output resolution and ML AI algorithm my view is no normal view of the windmill in those screenshots with that scene camera should have the contrast or detail shown in any of those zoomed in captures.

 

[Corporal.Hicks]

Okay fine, I'm happy to draw a line under you starting our discussion in this thread like below and thumbing up Corporal.Hicks childish verbal insults if you actual want to discuss.

So discussion your points about the quality DLSS windmill.

In my view, in the RL photo the depth cueing foliage/trees are about half or a third of the effective distance in the background compared to the distance of the windmill in RDR2, so tonally the whole thing should be far more depth cued in all three renders and starting to lose its strong outline far more, similar to the bright green foliage in the RL zoom photo subsection to the right of the roof vent is.

Is that partly a draw distance issue for accurate depth cue/fog control with RDR2's engine being a PS4 game, so probably can't use a frustum cascade to provide far more precision, but that still begs the question why doesn't DLSS quality mode intelligently fix that tonally and by coverage when inferencing the windmill and the other objects proportionally at distance? Which I believe was a big difference and done slightly better using PSSR in R&C vs DLSS.

Regardless of render output resolution and ML AI algorithm my view is no normal view of the windmill in those screenshots with that scene camera should have the contrast or detail shown in any of those zoomed in captures.

you know what time it's?

Spoiler: ?????