Nvidia DLSS support

Its likely because its Nvidia sponsored

2 Likes

This sharp NVidia logo with an excellent anti-aliasing algorithm where you can’t see any pixel, along with its unique shade of green and this white on black RTX label is looking nice to me. Sure, let’s hope.

1 Like

Which is a good news. I was afraid they wouldn’t work with Nvidia at all because Microsoft uses AMD GPU in Xboxes.

Yep, Jayne said during a community fly-in that the streams on the official channel are sponsored by NVIDIA because she is streaming using a system with an RTX 3080. They’ve also had an NVIDIA banner with a small note on the channel’s description for a while.

All computers that stream to this channel use NVIDIA graphics cards. We are a proud partner of NVIDIA.

Considering that NVIDIA have promoted the simulator on their own channels a few times, I think it’s baffling that DLSS has barely been considered so far. They obviously have more than a barebones partnership, which means implementing DLSS should be even easier.

1 Like

I do not agree with what you say as the xbox does not have a partnership with amd for FSM20, so here is Dlss2 possible and I can also say xbox games run in DLSS like the medium

1 Like

I’ve just seen the following video (posted in a topic which was immediately unlisted):

They are not showing anything better in their video than the CAS Shader already implemented in MSFS. It doesn’t look like super-res to me (aggregating multiple frames to create details):

  1. compare the character with the orange cape on both sides in the first shot below.

  2. compare the green plant to the left of the white bar on the first shot, with the same one to the right of the white bar in the second shot. Even when seen closer it doesn’t show more details at all and is more blurry



DLSS demonstrations/videos are looking much better…

…and given 85% of Simmers are using an NVidia (Navigraph survey) :face_with_symbols_over_mouth:


Look at this for example, it is not only faster, but also sharper with finer details with DLLS

7 Likes

I made a thread on the wishlist section about it.

1 Like

Quiz: find the missing title: :innocent:

3 Likes

DLSS in Battlefield V looks horrible. It’s not the 2.0 version though.

Fixed it! :smiley:

1 Like

Wait, what, RDR2 and Doom Eternal already have DLSS?!?!

Its coming to both games. Question is when

FSR is not really a replacement for DLSS at its current stage. It is a typical spatial upscaling solution, compared to the temporal reconstruction that is DLSS 2.x.

Of course, we all remember how terrible DLSS 1.0 was, and since AMD are calling this solution FSR 1.0, it means that it will be improved in the future (there are rumours that they will enable some enhancements specific to RDNA 3 when the time comes).

But there is really no reason for RTX owners to have to wait that long, GPUs will continue to be expensive for a long time and the next generations will not be coming out until late 2022 or so. They really need to support both solutions.

1 Like

I’m posting this here since it’s currently the most active DLSS/FSR conversation.
I feel like there’s a bit of intentional misleading going around, even on AMD’s part (more on that later). Maybe it’s their latest ‘underpromise and overdeliver’ mindset taking effect, who knows. Back when DLSS 1.0 was released, AMD’s RIS and/or CAS sharpening already pulled more detail out of an upscaled image than DLSS 1.0 did 90% of the time. I personally see no reason why AMD would create an upscaling method that would do worse than a simple sharpening filter they already have, but that’s what the current narrative seems to be.

On top of that, people only seem to have comments about image quality of the GTX1060 demo running at 1440P with FSR set at the second highest quality level, instead of the highest ‘Ultra Quality’ mode. Sure, it’s not looking great, but the native footage isn’t all that good to begin with, with motion blur cranked up to maximum. The higher the resolution and clearer image you’re upscaling from, the more detail you’re going to get. The same applies for DLSS.

Anyhow, to get to the bread and butter of my rambling. Here’s a few screenshots taken from AMD’s announcement video on Youtube. Why I personally think AMD has intentionally underpromised or ‘not shown the best they can do’; there seems to be a complete lack of any sharpening applied to the FSR footage, while Contrast Adaptive Sharpening (CAS) is part of the FidelityFX suite, and Radeon Image Sharpening (RIS) is a driver level feature. The demos just look ‘fuzzy’ as if it’s intentionally left unsharpened, which I find incredibly strange.

I’ve added a simple sharpening filter to a few screenshots below, showing what FSR could look like when CAS or RIS is running on top of it. Fingers crossed the forum doesn’t compress these even more. Open them in full screen and/or zoom in, and switch back and forth to see the differences.

Screenshots hidden here

Native vs FSR (Default)

Native vs sharpened FSR

Native vs FSR modes

Native vs sharpened FSR modes

GTX1060 Native (1440p) vs FSR

GTX1060 Native (1440p) vs sharpened FSR

We’ll have to wait for the first handful of games to arrive (or be updated, Asobo please take note) and see for ourselves how FSR gets implemented and what can be done to improve it even more client-side. From what I’ve seen so far, the quality loss at 4k using ‘Ultra Quality’ FSR and a bit of sharpening is negligible compared to native, and Quality mode still looks great while doubling the FPS. It can only improve with future updates, so I’m optimistic. It might not be for everyone, if you’re using FSR to upscale to 1080p you might end up with a blurry mess. It might not be it’s intended use. Only time will tell.

It’ll also run on the consoles, at least officially announced for XBOX. I have no doubt many (future) games will start to support it, including MSFS;
“At Xbox, we’re excited by the potential of AMD’s FidelityFX Super Resolution technology as another great method for developers to increase framerates and resolution. We will have more to share on this soon,” a Microsoft spokesperson told IGN.

4 Likes

@Matt987123

Your experiment with the sharpen filter is giving surprising good results indeed!

However the main point to me is not much about quality of FSR which I hope will be better than what they were showing off, but it is the fact DLSS is extra rendering power because it is using actual transistors on your video card which is doing nothing otherwise, whereas FSR will be using the same silicon that what is already used and probably fully busy by the rendering itself already. In other words:

DLSS:

  • extra fps by virtue of upscaling (rendering less but displaying as much)
  • upscaling is done with dedicated silicon just for this task
  • all the silicon meant for rendering (CU) can be fully used to rendering
  • it is also capable of creating new details that didn’t exist in the lower res frame.

FSR:

  • extra fps by virtue of upscaling
  • a balance must be set in allocating rendering silicon between rendering and FSR
  • it might just be able to enhance the existing details in the frame, not create new finer ones.

The effectiveness of FSR, besides the algorithm quality itself, might be affected by how much of the CU are used every frame. There are many games for which I doubt they are all 100% busy every frame and we don’t know how much computing FSR will require, but if you have a game renderer using 99% of the CU per frame already, with FSR you’ll have to trade off some pixel shader resources (lowering settings like AO, SSR etc…) whereas with DLSS this shouldn’t be a concern.

This is all speculation of course, but just to get a sense of it, I’ve tried No Man Sky in VR with DLSS yesterday for the first time, because I wanted to see how it goes, and I can say what I’ve seen is quite mind blowing. I’ve been pushing it at 300% render scale with SteamVR (about 3.2K x 3.6K pixels IIRC) at 45fps with motion smoothing 2:1, and at 200% render scale at 90fps. It is less demanding than FS2020 of course, but 300% upscaling in the Index is so legible you wish they would have supported DLSS in FS2020…

4 Likes

You’re right about the Tensor cores being able to offload the upscaling from the shaders, it does give you more processing power to work with. The upscaling itself is fairly efficient and is easy to do on shader cores though, at least for relatively modern GPUs (think GTX1060 and up, and AMD equiv. - basically what they’re supporting).

With DLSS 1.9 (first version of 2.0) running on ‘Control’, it ran on CUDA cores instead of Tensor cores. The performance was right where the current 2.0 runs on Tensor, so it seems the impact of upscaling on shader cores is easily offset by lowering the rendering resolution.

For now it’s just different techniques working towards the same goal, and both have their pros and cons. DLSS’s image clarity will undoubtedly be better off the bat, and FSR will be supported over a wide array of hardware and platforms (and to top it off, was never expected to exist on anything less than the RX6000 series, or even NVIDIA). I’m very happy with it’s potential, as upscaling from 1440p to 4k would pretty much double my framerates and lose very little detail in doing it, and so would it for almost everyone else. Allegedly.

I think it’s more realistic that FSR is coming to MSFS instead of DLSS (for now) as it’s already running part of FidelityFX. That, and being an Xbox title soon, they’ll want to improve framerates and resolutions as much as they can; DLSS wouldn’t help them much in that regard.

Adding DLSS would mean more development time on top of that, and only apply to people with RTX cards instead of anyone with a GPU from the last 6 years.
Who knows what mix of techniques they can come up with, the sim is already using CAS and temporal upscaling. Theoretically running FSR on top of that would give the best overall image clarity, although the current TAA is pretty blotchy when moving quickly.

In the end we’ll just have to wait and see, some improvement is still better than nothing at all ¯\_(ツ)_/¯

1 Like

Rainbow Six Siege has added DLSS 2.2 support, which has not been officially unveiled by NVIDIA yet. Apparently, it is possible to substitute the DLLs in games with DLSS 2.0/2.1 support with the new ones, and the result is reduced ghosting and artifacts.

2 Likes

DX12 support. That’s going to be game changing.
DLSS not so much as the game isn’t really GPU bound, but CPU bound, and DLSS doesn’t like to play with fine details… which is what MSFS is full of. (i.e. trees at a distance would look like paste)

2 Likes

Actually, DLSS does a lot better with fine details (like thin lines, which the simulator is full of) and trees. Take a look at these comparisons in Unreal Engine 4 (which already had good TAAU): Exploring DLSS 2.0 in Unreal Engine - Tom Looman

As for being CPU-bound, DirectX 12 will alleviate a lot of that. Don’t forget ray-tracing is coming too.

Those examples kind of look like it’s using a sharpening filter on the text, making it appear sharper than when full render resolution is used? Interesting choice.