10 minutes ago, Jetman67 said:

I actually did see a video tonight but I can't find it, it was a comparison between a 3080 ti verses a 4090 using dlss3 and in the 172 over new york 4 k ultra settings 186 fps butter smooth I'll try to find it, was a comparison of 10 games

Thanks, would be interesting to see. I assume though the base FPS was already very high if 186 FPS were achieved with DLSS3.

10 minutes ago, Jetman67 said:

I actually did see a video tonight but I can't find it, it was a comparison between a 3080 ti verses a 4090 using dlss3 and in the 172 over new york 4 k ultra settings 186 fps butter smooth I'll try to find it, was a comparison of 10 games

In my experience it’s not quite like that. To the original comment, adding AI frames to 80 or 90 FPS and getting 186 FPS you were already in a winning situation and those AI frames won’t be worth the frame generation side effects. 
 

First thing to note is Frame Generation is a totally separate option. You can use it with TAA, DLAA, and DLSS. You just need HAGs on and a 4000 series.

Second, your frame counter will not change in the sim. If you get 20 FPS it still says 20 FPS. If you run an external frame counter like CapFrameX it will register the extra AI frames. 
 

I loaded the Fenix at iniscene KLAX with 300 TLOD and AI. It’s not smooth at all. The dev mode counter says 25 FPS and capframex captures 50 FPS but it does not feel like it. The 1% lows are still bad. Bashing up against a CPU limite full throttle still stutters. It will not feel like a GPU limited 45-50 FPS. And yes, there is very clear artifacting when you pan around. It’s like your monitor is kind of dying or something. 

Can't find the actual video yet but here is a similar and shows some pretty impressive fps gains

but like everything there may be more to this

 

 

 

 

Here is some capframeX testing data.

This is the recorded FPS using a C172 around iniScene KLAX:

https://ibb.co/LS4y64P

The 1% and .2% lows are big swings, but sure you get the nice looking average FPS.  All in all it doesn't look that bad on the graphs, but that's not how it feels.

Take a look at the variances from CapFrameX from that test scenario.

https://ibb.co/9wwkdQQ

That's a lot of Red, Orange, and yellow.  This is what stuttering none smooth image looks like on a pie chart.

For example take a look at this test that was done at 4k, no frame generation and in a place where I got stable FPS, i.e. was not CPU limited

https://ibb.co/xjvxWNk

You can clearly see the difference. That's what a smooth no stutter pie chart looks like.

Edited October 15, 20223 yr by TravelRunner404

1 hour ago, TravelRunner404 said:

This is the recorded FPS using a C172 around iniScene KLAX:

Ok, but who is going to actually take the default C172 into LAX? I appreciate the benchmarking, but are you able to test more realistic and demanding scenarios? Maybe the FBW or PMDG at LAX with the building storms preset at dusk. Or if you’re more interested in GA, how about the Milviz 310, TBM, or a plane with a bit more complex simulation in an area that strains the GPU like LOWI at dusk with clouds/rain. 
 

The Fenix is extremely CPU bound, hence why I suggested the PMDG or FBW since those are much more balanced. 

Edited October 15, 20223 yr by V1ROTA7E

2 hours ago, TravelRunner404 said:

Bashing up against a CPU limite full throttle still stutters.

what's your CPU? your tests were run on a 4090?

I think the Hardware Unboxed DLSS3 analysis really proves that it's not a be-all end-all tech. I know the video has been shared here recently, but I'll share it again, starting from "final thoughts". 

Basically they're saying: 

• DLSS3 works best at high frame rates, preferably >120fps
• DLSS3 significantly introduces latency (not so important for slower-paced titles like MSFS)
• The visual artifacts from frame generation are much more noticeable at lower fps

So DLSS3 will not give any meaningful benefits when pairing a 4090 with a last-gen CPU and being main thread limited. Well, fps will probably be higher due to the frame generation, but at lower fps the visual artifacts might be severe. 

Also I think it very much depends on how sensitive a person is to these artifacts. Just my two cents. 

Edited October 15, 20223 yr by Cpt_Piett

1 minute ago, Cpt_Piett said:

I think the Hardware Unboxed DLSS3 analysis really proved that it's not a be all end all tech. I know the video has been shared here recently, but I'll share it again, starting from "final thoughts". 

Basically they're saying: 

• DLSS3 works best at high frame rates, preferably >120fps
• DLSS3 significantly introduces latency (not so important for slower-paced titles like MSFS)
• The visual artifacts from frame generation are much more noticeable at lower fps

So DLSS3 will not give any meaningful benefits when pairing a 4090 with a last-gen CPU and being main thread limited. Well, fps will probably be higher due to the frame generation, but at lower fps the visual artifacts might be severe. 

Also I think it very much depends on how sensitive a person is to these artifacts. Just my two cents. 

A 4090 needs a very modern cpu imo. 

My 3090 are using TAA as DLSS is less sharper in the far distance.

I have tried to like it but hagen gone back to TAA.  DLSS 3 might even be less sharp than DLSS 2 in the far distance.

For now it’s TAA for me…

1 hour ago, TravelRunner404 said:

You can clearly see the difference. That's what a smooth no stutter pie chart looks like.

great work, thanks very much. the proverbial "a benchmark picture is worth a 1.000 words" and definitely a lot more than the stereotype "it feels much smoother than xyz ..."

NVIDIA RTX 4080 12GB just 12% Faster than the RTX 3080, Slower than the 3090 Ti in Every Game

45 minutes ago, Ianrivaldosmith said:

A 4090 needs a very modern cpu imo. 

An opinion that I share!

51 minutes ago, Cpt_Piett said:

 

• DLSS3 works best at high frame rates, preferably >120fps
• DLSS3 significantly introduces latency (not so important for slower-paced titles like MSFS)
• The visual artifacts from frame generation are much more noticeable at lower fps

I guess even with a state-of-the-art CPU you won't achieve 120 FPS 90% of the time in MSFS, so it seems to me that there would be sort of a middle ground where the 4090 plus DLSS3 would yield a benefit albeit limited due to FPS not being high enough. Anything lower than that apparently results in artifacts and not much if any performance improvements while anything higher where you would get the most out of DLSS3 would be very rare.

The price surely isn't worth that middle ground, imo.

1 hour ago, turbomax said:

what's your CPU? your tests were run on a 4090?

5800x3D and, yes a 4090.

1 hour ago, V1ROTA7E said:

Ok, but who is going to actually take the default C172 into LAX?

Of course but it was a test of how well the frame generation works and looks. I threw one of the easier things I could think of at it. It wasn’t great and if you read the whole post and look at the charts the data shows it still stutters quite a bit even with the “50” fps. I say 50 with heavy quotations. 

6 minutes ago, TravelRunner404 said:

5800x3D and, yes a 4090.

Of course but it was a test of how well the frame generation works and looks. I threw one of the easier things I could think of at it. It wasn’t great and if you read the whole post and look at the charts the data shows it still stutters quite a bit even with the “50” fps. I say 50 with heavy quotations. 

I am wondering 50FPS on what monitor ?
What kind of refresh rate is the monitor ?
You cannot expect 50 FPS to be stutter free all the time on for example a 60HZ monitor. !!!.
There will be always frame dropping when FPS are not matching the refresh rate of a monitor without gsync and other solutions in this regard.