" What kind of advancements can we expect from DLSS? Most people were expecting a DLSS 3.0, or, at the very least, something like DLSS 2.1. Are you going to keep improving DLSS and offer support for more games while maintaining the same version?
DLSS SDK 2.1 is out and it includes three updates:
New ultra performance mode for 8K gaming. Delivers 8K gaming on GeForce RTX 3090 with a new 9x scaling option.
VR support. DLSS is now supported for VR titles."
âŚ
I have no clue why this was not announced during the keynote as this is huge (even though it might take a very long time until we see supported titles)
Any predictions if this means we could enjoy the 8KX with 90Hz+ if the game implements DLSS ?
Iâm not too familiar with DLSS. Is the max resolution you can get from it equal to the max resolution of the panel ? So for the 8KX it would use a 1440p signal or even lower and upscale it to max 4k. And on the 5K+ it would use maybe 720p or 1080p and upscale to a maximum of 1440p, is this correct ?
" Improved VR support. Maintaining the 2880Ă1600 resolution of top-end VR head mounted displays while delivering the recommended 90 FPS has been made easier with DLSS."
not sure if thatâs the limit of what can be achieved at the moment for VR or just an example since the Index is referenced as the mainstream high end HMD.
8KX doesnât support over 75Hz. You can use a miracle video card which can render 1000FPS 16K but 8KX will handle 75Hz only.
8KX requires to render the picture over 8000 pixels per the eye. It will be 3840 after the image correction.
If we use DLSS than we can render like 4000px then the card will upscale it using AI very fast and sharp to 8000, then it will transformed to 3840. FPS will raise in a few times.
Assuming 9x is 3x3 super sampling, that should be enough to smooth out the jaggies in Elite Dangerous, on a 4K flat display. Hopefully, SteamVR will allow for that level of super sampling too.
Unfortunately, nVidia might be talking about 3x3 upscaling, which is far less interesting to me.
hmm the used bandwidth is gonna stay the same I didnât thought about that.
So the GPU calculates for example a 1080p signal, then the Ai upscales it to 4k and then sends it to the HMD which requires the same amount of bandwidth as if it was a native 4k rendered image
On top of that, DLSS demos I have seen thus far are hit-or-miss at best. About a quarter of things donât get supersampled, another quarter shows severe artifacts, and text never gets supersampled correctly. DLSS is more a cheap form of anti-aliasing than supersampling. When expanded to the size of VR, missed stuff and artifacts will be really obvious.
Really high DLSS might make some of the blocky stuff around edges of things in Elite Dangerous go away, but not sharpen up the HUD text.
I donât think you will be able to turn down real supersampling, turn up DLSS to compensate, and magically get better framerates at the same visual quality.
After all that, DLSS of course does not increase physical refresh rates of the headset.
So, no. DLSS will not allow better framerates. Very high values of DLSS combined with real supersampling might suppress a few more artifacts.
A: Register here to request access to the DLSS beta program. Once your application is approved, NVIDIA will enable you to access DLSS through the public RTX Unreal Engine 4 branch.
but besides the reddit post I havenât seen any other official information about DLSS 2.1 so I guess we have to wait and see if there are now faster ways to get game support
Why do you think it wonât work on the HUD Text?
Surley it wonât do magic, but getting more Frames by same visual quality is what it does. It might not be the best example, but the text on death Strandings backpack was also more readable after Dlss (the demo during the show). Sure they took an example that works rather well in a demo then taking something thatâs not working good and sure its more likley to work well trained games
All the examples I have found of DLSS show it does not significantly improve the visual quality of any text. Which makes sense. As I understand it, DLSS is basically using a neural network as a kind of âlookup tableâ pointing to textures that are composited on top of the actual rendered image. That would overlay well on top of vegetation textures drawn by artists with semi-regular patterns, but most often would not add the meaningful information provided by actually supersampling arbitrary shapes like text.
If the supersampling resolution is turned down enough, that might have already been possible. At a TotalSR of 1.5x with the 8kX⌠we often canât disable Smart Smoothing, which has far more severe artifacts and latency.
The other thing is, you just need to teach the network and it will learn text in any languageâŚ
But if it will work under all circumstances⌠That depends I guess
