Nvidia shows signs in [2023]

  • Thread starter Deleted member 2197
  • Start date
Status
Not open for further replies.
With Geometry Acceleration for DLSS4 I do wonder how Blackwell will maybe improve DLSS Super-Resolution? Like DLSS Quality now gives you DLSS Performance levels of performance improvement? Or will Geometry Acceleration be the big thing?

I'm assuming geometry acceleration is one of those "insider" leak things. I don't even know what that would be, but I wouldn't waste too much brainpower thinking about it. Those insiders are 99.9999% full of :poop:
 
As per Alex's rant in a recent DF Direct, I'd love to see something done about the state of video play-back in-game. A new codec which utilizes AI/compute to generate frames to ensure that video playback remains smooth despite varying frametimes and refresh rates. Also improving the compression ratio of said videos as well.

It's time we start knocking this sticking points down one by one.. because nothing is worse than coming off of beautifully rendered high resolution, high framerate in-game action into a low res, stuttery video playback. It's time that high quality, smooth playback becomes a priority.
 
With Geometry Acceleration for DLSS4 I do wonder how Blackwell will maybe improve DLSS Super-Resolution? Like DLSS Quality now gives you DLSS Performance levels of performance improvement? Or will Geometry Acceleration be the big thing?
Geometry acceleration? What is this?
 
In a parallel development, NVIDIA has unveiled a new software development kit (SDK) that conveniently includes support for DLSS 3.5. This was spotted by “emoose,” creator of DLSSTweaks, a tool that allows gamers to tinker with DLSS settings and profiles.

It’s crucial to note that DLSS 3.5 doesn’t automatically enable ‘Ray Reconstruction.’ This feature necessitates integration into a game’s graphics pipeline by the developer. For those using Unreal Engine 5, the process is expected to be easier, as NVIDIA offers a 3.5 UE5 plugin and Redditors have already found a readme file that explains how the plugin is structured.

However, as DLSS 3.5 comes with improved upscaling algorithms, some gamers have already swapped out libraries and reported notable improvements in image quality, along with a reduction in ghosting in games such as Last of Us or Need for Speed Unbound. Nevertheless, it’s essential to bear in mind that merely swapping the DLL won’t activate ray reconstruction. Developers must take deliberate steps to incorporate this feature into their games.
 
Yeah, this is why DLSS 3.5 isn't "just DLSS2" when it comes to super resolution component.
You can bet those same improvements are/will be in newest "DLSS2" implentations too - like NVIDIA has said several times, DLSS 2 = DLSS Super Resolution and it keeps getting updated.
 
You can bet those same improvements are/will be in newest "DLSS2" implentations too - like NVIDIA has said several times, DLSS 2 = DLSS Super Resolution and it keeps getting updated.

DLSS2 is dead. There are no new implementations of DLSS2. When companies support DLSS, they'll be using the 3.5 SDK. The can choose to support any of the three features: Super Resolution, Frame Generation and/or ray reconstruction. The Super Resolution model will keep getting updated, but it's DLSS 3.5 now. At least that's how I understand the new naming.
 
You can bet those same improvements are/will be in newest "DLSS2" implentations too - like NVIDIA has said several times, DLSS 2 = DLSS Super Resolution and it keeps getting updated.
So you still didn't look at the current version of DLSS super resolution, eh?
 
Yeah, this is why DLSS 3.5 isn't "just DLSS2" when it comes to super resolution component.
Yes it is. DLSS2/AI upsampling part of things was always getting improvements. This is just a continuation of that, not some totally new tech. Guarantee these improvements will be available for any developer who is only doing DLSS2 implementation.
 
So you still didn't look at the current version of DLSS super resolution, eh?
It's all just refinements of DLSS2. It's still DLSS2. If a dev isn't gonna implement frame generation or the new ray denoising, they will almost assuredly NOT call it DLSS3, but instead DLSS2, with all these same improvements to DLSS2 being made here.

This isn't like the jump from DLSS1 to DLSS2 which was a fundamentally different tech.
 
DLSS2 is dead. There are no new implementations of DLSS2.

Devs aren't gonna say they're implementing DLSS3 or 3.5 without having the features that are explicitly tied to that naming.

This is what I was saying before - Nvidia made this needlessly confusing. All they had to do was keep the DLSS naming for AI upsampling, while leaving Frame Generation and whatnot under their own names. OR, if they really wanted to have an umbrella naming for all these available technologies, they should have done a new name, not continued with 'Deep Learning Super Sampling'. What on earth does Frame Generation have to do with Deep Learning Super Sampling? Nothing.
 
Would it be easier if we said the 40 series support DLSS 3.5, earlier RTX cards supported DLSS 3.5 LE, and the 9 series support DLSS 3.5 MX? ;) Or can we stop bickering over marketing tents and start using the important qualifiers like FG or SR?
 
DLSS2 is dead. There are no new implementations of DLSS2. When companies support DLSS, they'll be using the 3.5 SDK. The can choose to support any of the three features: Super Resolution, Frame Generation and/or ray reconstruction. The Super Resolution model will keep getting updated, but it's DLSS 3.5 now. At least that's how I understand the new naming.
Nah.

DLSS1 = Spatial SR running on tensor cores.
DLSS2 = Temporal SR running on tensor cores.
DLSS3 = Frame Genration + DLSS2
DLSS3.5 = Ray Reconstruction + DLSS3

Blame Nvidia.
 
Nah.

DLSS1 = Spatial SR running on tensor cores.
DLSS2 = Temporal SR running on tensor cores.
DLSS3 = Frame Genration + DLSS2
DLSS3.5 = Ray Reconstruction + DLSS3

Blame Nvidia.
This would be my preferred shorthand. NV were nice enough to introduce a single standout feature for each number increase, but RR throws a wrench into the works by extending back to the 20 and 30 series. Maybe we could say those cards support DLSS3.5- -.
 
Nah.

DLSS1 = Spatial SR running on tensor cores.
DLSS2 = Temporal SR running on tensor cores.
DLSS3 = Frame Genration + DLSS2
DLSS3.5 = Ray Reconstruction + DLSS3

Blame Nvidia.

DLSS Super Resolution requires the 3.5 SDK and is called DLSS Super Resolution 3.5

1693075884727.png
1693075740673.png


NVIDIA DLSS 3.5 is a suite of AI rendering technologies powered by Tensor Cores on GeForce RTX GPUs for faster frame rates, better image quality, and great responsiveness. DLSS now includes Super Resolution & DLAA (available for all RTX GPUs), Frame Generation (RTX 40 Series GPUs), and Ray Reconstruction (available for all RTX GPUs). To get the full benefits of DLSS, download both DLSS Super Resolution and DLSS Frame Generation SDKs/Plugins.

NVIDIA DLSS Unreal Engine Plugin

NVIDIA DLSS 3 for Unreal Engine contains both the DLSS Super Resolution Plugin and the DLSS Frame Generation Plugin. Download the NVIDIA DLSS 3 Unreal Engine package to get access to both plugins.

DLSS 3 Features include:

  • DLSS Frame Generation*
  • DLSS Super Resolution
  • NVIDIA Reflex
  • DLAA
  • NVIDIA Image Scaling
DLSS is a suite and as they add major features the version increases. Some gpus support all of the features, some don't.
 
Nah.

DLSS1 = Spatial SR running on tensor cores.
DLSS2 = Temporal SR running on tensor cores.
DLSS3 = Frame Genration + DLSS2
DLSS3.5 = Ray Reconstruction + DLSS3

Blame Nvidia.
Don't forget Reflex which is part of DLSS3 and works on GeForce 900+ GPUs lol
 
DLSS is a suite and as they add major features the version increases. Some gpus support all of the features, some don't.
I'm aware of what that says.. but they messed up.. DLSS was never a suite to begin with it was only a super resolution upscaling technology.. and they should have kept it that way. That way, people would understand DLSS3... as the 3rd revision of DLSS.. not a stupid suite with a bunch of different technologies which do or do not work on various GPU series. It also would be extremely clear to say DLSS3 is supported on all RTX GPUs. DLSS3 should have been the 3rd iteration of SR... and not DLSS2+FG.

Imagine just how much simpler it would be...

RTX:
DLSS3.x ----> Scales up as tech improves (All RTX)
NVRR1.x ----> Scales up as tech improves (All RTX)
NVFG1.x ----> Scales up as tech improves (40 Series+)

"Oh, my 20/30 series cards supports DLSS3 and NVRR, cool"

Easy.
 
I'm aware of what that says.. but they messed up.. DLSS was never a suite to begin with it was only a super resolution upscaling technology.. and they should have kept it that way. That way, people would understand DLSS3... as the 3rd revision of DLSS.. not a stupid suite with a bunch of different technologies which do or do not work on various GPU series. It also would be extremely clear to say DLSS3 is supported on all RTX GPUs. DLSS3 should have been the 3rd iteration of SR... and not DLSS2+FG.

Imagine just how much simpler it would be...

RTX:
DLSS3.x ----> Scales up as tech improves (All RTX)
NVRR1.x ----> Scales up as tech improves (All RTX)
NVFG1.x ----> Scales up as tech improves (40 Series+)

"Oh, my 20/30 series cards supports DLSS3 and NVRR, cool"

Easy.

I'm guessing they started with DLSS, then changed the model so figured DLSS 2.0 was the way to go. They probably didn't really know how many different models they'd come up with. At some point naming them all 2, 3, 4, 5, 6, 7, 8 would just get super confusing, especially if they depricate or replace them. So now they've switched the naming. Makes sense from a developer standpoint. It's kind of like DirectX with feature levels for different gpus. No matter which way they name them for consumers, it'll probably be confusing. Most people aren't going to know what these features do, and if there was a list of ten of them they'd be lost.
 
I'm guessing they started with DLSS, then changed the model so figured DLSS 2.0 was the way to go. They probably didn't really know how many different models they'd come up with. At some point naming them all 2, 3, 4, 5, 6, 7, 8 would just get super confusing, especially if they depricate or replace them. So now they've switched the naming. Makes sense from a developer standpoint. It's kind of like DirectX with feature levels for different gpus. No matter which way they name them for consumers, it'll probably be confusing. Most people aren't going to know what these features do, and if there was a list of ten of them they'd be lost.
The original DLSS was quite different, frame upscalers like FSR1. It was called DLSS2 after the shift in design to motion vectors and incorporating the model they ended up with from all the training. It wasn't just a change in model which is being updated with each new sub version currently.
 
Status
Not open for further replies.
Back
Top