DLSS 2 is AI. It does learn.  Given a certain frame, it predicts what the next frame will look like. When the next frame appears it measures how much the prediction was wrong. And then it changes the way it predicts by changing the coefficients of some of the factors and then tries to play the same game on the exact same original inputted image using  these new coefficients (numbers). Next it decides if the first guess or the second guess was better. 

2x2x2x2x2x2x2 is a large number. But there are more than 2 states for each factor and there's more of these factors. So it's more like 9x9x9x9x9x9x9... etc. That's a lot of trials just to test that one 'guess' on that one inputted image.

To do it right would take an humongous amount of computing power years to finish. And a different leaned scheme is needed for each individual sim or game. So it never gets to perfection. But it does get better as the testing moves along. Because it is learning. This learning is turned on by the sim producer (in this case probably Asobo) and it runs night and day for months. DLSS keeps getting better all the time. 

DLSS 2 may never get better. But how well DLSS 2 works on MSFS will get better until Asobo (or whoever they outsourced it to) turns off the learning scheme.

Sony made the exact same claim, and was later found out to be fibbing. Sony had said they had a neural network on a super computer to train a texture algorithm, bla bla. I don't think DLSS is trainable by the game developer, I looked in their SDK and didn't find it.

Instead, it simply says this:
DLSS uses the power of NVIDIA’s supercomputers to train and regularly improve its AI model. The latest models are delivered to your GeForce RTX PC through Game Ready Drivers. Tensor Cores then use their teraflops of dedicated AI horsepower to run the DLSS AI network in real-time. This means you get the power of the DLSS supercomputer network to help you boost performance and resolution. And it gets even better over time.

It doesn't say anything about being able to self-train the AI. The idea of running a training mechanism on a game for some set of numbers, how would that even work. How would the AI know if the resulting frame was better or not as perceived to the human eye. Even Tesla couldn't get their cars to drive just right, and that problem is basic image recognition compared to this problem, which is a scientific nightmare of combubulated factors.

 

Edited October 23, 20223 yr by Alpine Scenery

3 hours ago, GSalden said:

Myself I did not dig into it (yet) so in simple words if I am correct : first downscaling, then upscaling and adding an AI frame between 2 real frames so the framerate doubles …

What is important for me : at a large airport with many AI ac and bad weather where my 12900K cpu is the bottleneck and goes down to 37/38 fps , will DLSS 3 bring extra smoothness because of the extra AI frames ?

We all know the jumping from the scene when having just 10 fps or even with 30 fps and using a fast ac, making a fast turn left and look left out the window.

Just curious as I might pick up a Inno3D X3 4090 + Dellta GPS 1300W Ps today …

 

My humble advice, wait for the 7900x3d if you can, after that choose your graphic card.

Is is exactly like the same playbook of the last 15 years, page for page. Claim we have something superior due to our superior AI. Throw in terms like neural network and super computers. I don't doubt they added some dubious, likely useless code, to the algorithm just in case the regulators come after them (Sony didn't even do that though, and regulators left them alone, it wasn't big enough market to care). I also don't doubt they own super computers and probably even have an experimental neural network, but that's all it is, experimental. In marketing, you can pretty much say anything. Trying to calculate specific co-efficients based on imperfect imagery, impossible. The quality of the imagery varies, there is no benchmark that could be done by AI, barely could the human eye even do it (and I mean barely). Machine learning cannot train itself with imagery, it has to be human assisted. The aerial imagery varies tile to tile, and the cockpit looks completely different than the aerial imagery. There would be no standard or alteration of an algorithm that could withstand the scrutiny of such changes, and such algorithms are easily 25 to 40 years away. Any co-efficients that could improve one thing, would degrade another. Trying to find the perfect algorithm to do this properly is a life-long project by multiple genius-level individuals, and it just hasn't been done yet, not even in the research field have they had much success in improving image fidelity with AI, other than using anti-aliasing or search and replace type functionality (letting the AI find the same image online that is higher quality and then blending it). It kind of steams me just a little bit with just how far these companies are going with their AI baloney. 

OK, think about it like this, supposedly Nvidia has such incredible AI, yet Microsoft cannot even run the aerial imagery through a basic color correction AI to fix all the bad color?

Why?

Because AI is hard to program, it's tedious as heck, it's very very expensive and something that you think might take 3 months can take 3 years or longer to finish. 

Or Adobe doesn't have hardly any decent working AI that auto-fixes imagery?
This isn't even a real-time problem.

How about Maya, Blender, Substance Painter, 3DS Max?
Nothing?

Only Nvidia is special, not only can they do something that every scientist at MIT failed to do, that every programmer at all the aformentioned companies failed to do, but....

DRUM ROLL...

They can do it in REAL TIME and adhere to a millisecond latency issue, all why doing something no other company has been able to pull off.  They can make AI to do such incredible things, it defies belief. Exactly the same playbook Sony used, exactly.
 

 

Edited October 23, 20223 yr by Alpine Scenery

2 minutes ago, Alpine Scenery said:

Is is exactly like the same playbook of the last 15 years, page for page. Claim we have something superior due to our superior AI. I don't doubt they added some dubious, likely useless code, to the algorithm just in case the regulators come after them (Sony didn't even do that though, and regulators left them alone, it wasn't big enough market to care). In marketing, you can pretty much say anything. Trying to calculate specific co-efficients based on imperfect imagery, impossible. The quality of the imagery varies, there is no benchmark that could be done by AI, barely could the human eye even do it (and I mean barely). Machine learning cannot train itself with imagery, it has to be human assisted. The aerial imagery varies tile to tile, and the cockpit looks completely different than the aerial imagery. There would be no standard or alteration of an algorithm that could withstand the scrutiny of such changes, and such algorithms are easily 25 to 40 years away. Any co-efficients that could improve one thing, would degrade another. Trying to find the perfect algorithm to do this properly is a life-long project by multiple genius-level individuals, and it just hasn't been done yet, not even in the research field have they had much success in improving image fidelity with AI, other than using anti-aliasing or search and replace type functionality (letting the AI find the same image online that is higher quality and then blending it).

 

 

Is is exactly like the same playbook of the last 15 years..

Well without Nvidia and/or AMD there would not have been Flight Simulator as it is now !
And for now the 4090 (even without frame generation) is superior to all other cards.
The 4090 is not only here for FS2020

16 minutes ago, altenae said:

Is is exactly like the same playbook of the last 15 years..

Well without Nvidia and/or AMD there would not have been Flight Simulator as it is now !
And for now the 4090 (even without frame generation) is superior to all other cards.
The 4090 is not only here for FS2020

I agree, and Sony makes some of the best TV's, but it doesn't change the fact that their marketing departments are some of the worst perpetrators of outright baloney. The flood of AI papers you see on Nvidia's site that looks like it was written for doctors to read, it's all for show, it's all a psychological game in marketing. It's really sad to see them stoop to such a level.

 

 

Edited October 23, 20223 yr by Alpine Scenery

13 hours ago, FBW737 said:

That's what the Judder reduction filter in my Samsung 4KTV does.

Careful!  I got criticised for saying that. 
It's true though.  I played GTA 5 on my 4k Samsung TV with the built in frame interpolation, and it was dead smooth in most places and worked quite well.  
Apparently, Nvidia have licensed this technology and now call it DLSS 3 with frame generation.  {joke}

I DO understand that the two systems work slightly differently.  Samsung took two frames and 'averaged' one in the middle.  Very good quality, but there was a bit of lag

I am not knocking it.  Most people will like it, and it does work quite well.  I am just knocking the usual marketing hype that comes from Intel, AMD, and Nvidia.
Samsung just put the capability on a little graphics processing chip built into their TV's years ago. 
We could have had this kind of technology on AMD / NVidia graphics cards years ago as well; not sure why we didn't.
Even with the slightly different way Nvidia do it (which is inferior in quality actually), it is not like some massive technological breakthrough.  But hey, that's marketing and we have to expect it.

1 hour ago, pilotter said:

My humble advice, wait for the 7900x3d if you can, after that choose your graphic card.

I reckon the x3D chips might not see the big gains like we saw with the 5800x3D. Don't forget Intel has also increased their L2 and L3 cache as well as their clock speed with Raptor Lake.

It will boil down to heat (no pun intended). If AMD lower the clock speeds like they did with the 5800x3d chips to compensate for this, then they may struggle to beat Intels current 10% gain and impressive 1% lows. And lets give them the benefit of the doubt. By how much will they beat Intel given the likely power/heat restrictions? Maybe 5%? Is it really worth waiting? And if these x3d chips will be so great, and only a few months away, why are the newly released existing AM 5 chips so 'average' for the price amd hype that surrounded them? 

Currently I am leaning heavily towards just getting an Intel. 

1 hour ago, altenae said:

Who cares real or not real!!!!

What matters is that games play superb. 

Also very happy with frame generation on in combination with FS2020.  
Also F1 feels like the real deal with frame generation on.  
 

I don't care what all the vlogs say!!!
I care about my own experience!!

Same with food vlogs telling us a Big Mac is not very tasteful.  

Well I will decide about that. 

I also think that if the 4090 was 900 euro everyone would be happy and where not bashing against the 4090. 

 

 

Did you see what Alpine Scenery posted ?

Did you look what it showed at 11:15…?

I guess you bought the card the day it came out just because a new gpu came had been released and you only looked at the pros …

Myself and others here are looking not only at the pros but also at the cons, so we know what to expect if we would take the step. Or not ….

Edited October 23, 20223 yr by GSalden

1 minute ago, GSalden said:

Did you see what Alpine Scenery posted ?

Did you look what it showed at 11:15…?

I guess you bought the card the day it came out just because a new gpu came had been released and you only looked at the pros …

Myself I am looking not only at the pros but also at the cons, so I know what to expect if I am taking the step. Or not ….

Sure whatever.

2 hours ago, GSalden said:

I am curious what current owners experiences are. We have a few with a 4090 ….

I've shared my opinions in different threads. Some got moved to the monitor section, you might have a look there. I also made several videos but they don't do the 4090 justice for a number of reasons, some of which are elaborated on by Digital Foundry. 

But I'll reiterate: if you're not a visual quality aficionado that will actively look for the artifacts in every corner, you'll probably be as happy with it as I am. The fps increase is nothing short of amazing - basically it's doubled due to the extra frames generated, even when main thread limited. As it's the GPU that generates those extra frames. Even without DLSS3, there's a 25-30% fps improvement vs the 3090 according to my tests - can always turn off frame generation anytime. 

The biggest problem with this card is probably it's price. I can't recall that was much different when the 3090 was released though. 

If in doubt, watch Digital Foundry's and Hardware Unboxed reviews closely before emptying that bank account. 

3 hours ago, Alpine Scenery said:

This guy does a pretty good analysis starting at 11:15...

I'll just make some comments on the HU analysis. In the conclusion he lists three criteria he sees as important if DLSS3 will be of any benefit: 

1. "Need to be able to render at 100-120fps before enabling frame generation". Different opinions on this and I partly disagree. I've had a good experience also at lower frame rates. This reviewer disagrees with the high fps criteria. However - the image distortions could be more pronounced at lower fps, especially when panning very quickly around. As fewer frames are generated, the distortions are a bit more obvious. But in MSFS it's exactly at those tight low fps situations with heavy scenery, airliner addons and AI traffic that frame generation could be of benefit to avoid the familiar stutterfest. In my experience DLSS3 does not introduce stutters in the 30-40 fps range. This is also reflected in my CapFrameX benchmarks. My videos are very misleading though, as the video capture itself introduced stutters. That and the fact that in one of my videos the "real" fps at KLAX with the Fenix taking off plummeted to an embarrassing 17 fps. I've since tuned my settings and get an average fps >30 at KLAX without DLSS3.  

2. "You need to have a high refresh rate display, ideally 240Hz or higher". Completely disagree with regards to MSFS. I don't think any elaboration on this is necessary - it goes without saying. 

3. "You need to be playing a slower game that is not latency sensitive". True. As MSFS is one of those "slower games" input lag will not matter. 

1 hour ago, bobcat999 said:

I played GTA 5 on my 4k Samsung TV with the built in frame interpolation, and it was dead smooth in most places and worked quite well.  

But can you use your TV frame interpolation to achieve the same effect as DLSS3 in MSFS, effectively doubling the frame rate? Disclaimer: this is an honest question, no trap 🙂

Also, I find this topic particularly hard to discuss in a constructive way. It's the same over at the official forums. Not exactly sure why, but I suspect there are many things at play here. Lol, someone even got caught up with the stuttering frequency in one of the CapFrameX benchmarks I shared comparing the 3090 to the 4090 (0.2% with the 3090 vs 1.1% with the 4090). The whole thing was laughed off and the discussion ended there 😖

Edited October 23, 20223 yr by Cpt_Piett

12 minutes ago, steve310002 said:

I reckon the x3D chips might not see the big gains like we saw with the 5800x3D. Don't forget Intel has also increased their L2 and L3 cache as well as their clock speed with Raptor Lake.

It will boil down to heat (no pun intended). If AMD lower the clock speeds like they did with the 5800x3d chips to compensate for this, then they may struggle to beat Intels current 10% gain and impressive 1% lows. And lets give them the benefit of the doubt. By how much will they beat Intel given the likely power/heat restrictions? Maybe 5%? Is it really worth waiting? And if these x3d chips will be so great, and only a few months away, why are the newly released existing AM 5 chips so 'average' for the price amd hype that surrounded them? 

Currently I am leaning heavily towards just getting an Intel. 

Wait for benchmarks. They might not be best in class for every use case, but I'd expect them to perform extremely well in MSFS again, even if the clock speed is reduced by a few percent, again. 

The 5800x3d is not good in everything as well, just very good for the specific use case of MSFS.

I have nothing against the card itself, if someone wants it and especially if it really ends up giving a 25% to 30% increase in FPS. It is quite expensive though, and from the reviews I have seen, the DLSS3 isn't worth the degradation unless you have a specific airport or use case where you need that much power.

My only point was, I'm kind of worn out on the whole AI marketing game all these companies play. Like any big deception, there is some truth to it, and there are some useful AI tools out there, but as it currently stands, about 1% of it is useful and about 99% of it is marketing. The problem spans the entire global IT world right now, the hype of AI has gotten off the charts. In reality, most programmers aren't that good at AI, even some that specialize in it, it's tough to code in true AI frameworks and hit your end goals. Everything always ends up either under-fit or over-fit, buggy, and it's an endless circle.

 

1 hour ago, steve310002 said:

I reckon the x3D chips might not see the big gains like we saw with the 5800x3D. Don't forget Intel has also increased their L2 and L3 cache as well as their clock speed with Raptor Lake.

It will boil down to heat (no pun intended). If AMD lower the clock speeds like they did with the 5800x3d chips to compensate for this, then they may struggle to beat Intels current 10% gain and impressive 1% lows. And lets give them the benefit of the doubt. By how much will they beat Intel given the likely power/heat restrictions? Maybe 5%? Is it really worth waiting? And if these x3d chips will be so great, and only a few months away, why are the newly released existing AM 5 chips so 'average' for the price amd hype that surrounded them? 

Currently I am leaning heavily towards just getting an Intel. 

you are right, future will tell. Only thing they mentioned now, it will be overclockable and if you look at the 5800x3d results now.... L3 cache is much bigger also. Anyhow I hope everybody enjoys MSFS, world is crazy enough.  I thought AMD would be the best route but looking at my 6900XT and what AMD has done in terms of drivers for MSFS, not mentioning there are no shaders etc....I might reconsider my next purchase.  We will now next month when the new AMD card is coming.

1 hour ago, rka said:

Wait for benchmarks. They might not be best in class for every use case, but I'd expect them to perform extremely well in MSFS again, even if the clock speed is reduced by a few percent, again. 

The 5800x3d is not good in everything as well, just very good for the specific use case of MSFS.

my rig is purely for MSFS, nothing else. I even go the extra mile in having a stripped down W11, and download + internet on a other computer just to keep it as clean as possible. But yes, I will wait for the benchmarks, because I will have to update Mobo , memory, waterblock and cpu to go AMD gen5