rjohnson11
EVGA Forum Moderator
- Total Posts : 102227
- Reward points : 0
- Joined: 2004/10/05 12:44:35
- Location: Netherlands
- Status: online
- Ribbons : 84
https://www.kitguru.net/components/graphic-cards/dominic-moass/nvidia-clarifies-dlss-and-how-it-works/ DLSS is a post-processing technique that improves performance over traditional anti aliasing (AA) methods in two main ways. First of all, it simply takes less samples per pixel than current AA methods – meaning demand on the GPU is lessened. Secondly, DLSS is executed on the Tensor Cores within the Turing GPU, rather than CUDA cores. This frees up the shaders to focus on rendering instead of dealing with as much anti aliasing. DLSS can increase game performance because traditional GPU shaders are not being leveraged for AA, while DLSS is also using less samples per pixel. Lots of good information in this article if you want to know more about DLSS.
|
the_Scarlet_one
formerly Scarlet-tech
- Total Posts : 24581
- Reward points : 0
- Joined: 2013/11/13 02:48:57
- Location: East Coast
- Status: offline
- Ribbons : 79
Re: Nvidia clarifies DLSS and how it works
2018/09/21 09:39:19
(permalink)
In short, dlss renders parts of a 4k scene at 1440p then upscales to 4k. Some parts of the scene never get Anti Aliasing if it is not required.
The first comment on the article explained that KItGuru misinterpreted part of the information, and KItGuru was quick to update it, which is great to see.
|
Cool GTX
EVGA Forum Moderator
- Total Posts : 30963
- Reward points : 0
- Joined: 2010/12/12 14:22:25
- Location: Folding for the Greater Good
- Status: offline
- Ribbons : 122
Re: Nvidia clarifies DLSS and how it works
2018/09/21 10:13:32
(permalink)
more interesting reading 
|
redleader00
SSC Member
- Total Posts : 906
- Reward points : 0
- Joined: 2009/03/26 11:02:11
- Status: offline
- Ribbons : 7
Re: Nvidia clarifies DLSS and how it works
2018/09/21 12:00:06
(permalink)
the_Scarlet_one
In short, dlss renders parts of a 4k scene at 1440p then upscales to 4k. Some parts of the scene never get Anti Aliasing if it is not required.
The first comment on the article explained that KItGuru misinterpreted part of the information, and KItGuru was quick to update it, which is great to see.
I don't see it that way. DLSS is not limited to 4K, but they show an example with a 4K image because there is where you can notice the difference the most. Then, the game still needs to render the "raw" image at the same quality you are displaying (1080,2K,4K, whatever) but the 2 key differences are: - It only needs to use 1 sample per pixel to get an anti-aliasing result that is BETTER than a 64 sample per pixel TAA.
- All this is done using the tensor cores, not the cuda cores that are used to render the image. Pascal has to use cuda cores for rendering and anti-aliasing, while Turing can use all cuda cores ONLY for rendering and then apply the AA using the tensor cores.
In short, you get great anti-aliasing without losing rendering performance. Source: DLSS: What does it mean for game developers? (nVidia page)
post edited by redleader00 - 2018/09/21 12:02:27
|
the_Scarlet_one
formerly Scarlet-tech
- Total Posts : 24581
- Reward points : 0
- Joined: 2013/11/13 02:48:57
- Location: East Coast
- Status: offline
- Ribbons : 79
Re: Nvidia clarifies DLSS and how it works
2018/09/21 16:05:22
(permalink)
redleader00
I don't see it that way.
DLSS is not limited to 4K, but they show an example with a 4K image because there is where you can notice the difference the most.
Then, the game still needs to render the "raw" image at the same quality you are displaying (1080,2K,4K, whatever) but the 2 key differences are:- It only needs to use 1 sample per pixel to get an anti-aliasing result that is BETTER than a 64 sample per pixel TAA.
- All this is done using the tensor cores, not the cuda cores that are used to render the image. Pascal has to use cuda cores for rendering and anti-aliasing, while Turing can use all cuda cores ONLY for rendering and then apply the AA using the tensor cores.
In short, you get great anti-aliasing without losing rendering performance.
Source: DLSS: What does it mean for game developers? (nVidia page)
Digital Foundry looked into very close. To get the higher 4k frame rates, it is rendering in 1440p, or half the resolution. It also does not project previous frame information into the next frame where as TAA does. So, even if you “don’t see it that way” it is that way. Also, from the article you linked:
We are able to use the newly inferred information about the image to apply extremely high quality “ultra AA” and increase the frame size to achieve a higher display resolution.
“Increase the frame size” from a lower resolution to display as a higher resolution.
post edited by the_Scarlet_one - 2018/09/21 16:09:17
|
redleader00
SSC Member
- Total Posts : 906
- Reward points : 0
- Joined: 2009/03/26 11:02:11
- Status: offline
- Ribbons : 7
Re: Nvidia clarifies DLSS and how it works
2018/09/22 06:02:07
(permalink)
the_Scarlet_one
redleader00
I don't see it that way.
DLSS is not limited to 4K, but they show an example with a 4K image because there is where you can notice the difference the most.
Then, the game still needs to render the "raw" image at the same quality you are displaying (1080,2K,4K, whatever) but the 2 key differences are:
- It only needs to use 1 sample per pixel to get an anti-aliasing result that is BETTER than a 64 sample per pixel TAA.
- All this is done using the tensor cores, not the cuda cores that are used to render the image. Pascal has to use cuda cores for rendering and anti-aliasing, while Turing can use all cuda cores ONLY for rendering and then apply the AA using the tensor cores.
In short, you get great anti-aliasing without losing rendering performance.
Source: DLSS: What does it mean for game developers? (nVidia page)
Digital Foundry looked into very close.
To get the higher 4k frame rates, it is rendering in 1440p, or half the resolution.
It also does not project previous frame information into the next frame where as TAA does. So, even if you “don’t see it that way” it is that way.
Also, from the article you linked:
We are able to use the newly inferred information about the image to apply extremely high quality “ultra AA” and increase the frame size to achieve a higher display resolution.
“Increase the frame size” from a lower resolution to display as a higher resolution.
I think Digital Foundry were talking about their experience with the available demos. Looking at the NGX SDK page, it looks to me that those demos are actually using 2 technologies: - DLSS to apply "ultra-AA" to the image.
- AI Up-res to increase the image resolution.
I think you can use those techs separately, meaning that you can, for example, render a game at 2K and apply "ultra-AA" using DLSS without doing any upscaling and display it on your 2K monitor. Or you can pick a picture/video and upscale it without doing any AA. Or do both as they do in those demos.
post edited by redleader00 - 2018/09/22 06:05:08
|