Sort of a click-baity title, but the analysis is quite enlightening, and I would have loved to see him try out a few flight Sims.

 

I’ll check this out tomorrow.  Is it focused on 3x and 4x FG in the 50 series?

For me, 2x FG has been a life saver on a few games. Cyberpunk and Stalker 2 with all the eye candy turned on goes from about 40ish to 70-80FPS which makes a huge difference in perceived smoothness in an action/shooter. Even ARMA 3 which is so old it’s extremely CPU limited, using lossless scaling to double its frame rate makes a huge difference in smoothness. I don’t know how there can still be skeptics. I think any game that’s CPU bound at times is going to benefit and even games like CP2077 that use advanced ray-tracing can benefit. And I’ve never seen FG degrade image quality.

On MSFS 2024, I run FG with TAA with all settings maxed and get both amazing visuals and around 100FPS normally on my 4090, and when I land, it dips to 60, which means I never notice it.. unlike if it was going from 60-50 down to 40-30 FPS which would be very jarring. And with AutoFPS I can smooth that dip on landing to about 80FPS AND get even higher TLOD in flight at 100FPS.

I can’t go back to life without FG. 😛

Edited August 7, 2025Aug 7 by Virtual-Chris

3 minutes ago, Virtual-Chris said:

I’ll check this out tomorrow.  Is it focused on 3x and 4x FG in the 50 series?

It also goes into the 40 series.

It details how using FG cuts into your "real" frames, percentage-wise

The problem of all those "analyses" is that the baseline is entirely subjective but they usually claim it is a hard coded fact that something like "native" frames exist. They do not. Also their so called "native" frame is simply a digitally rendered image displayed on a monitor and this digital rendering includes a ton of algorithms, gpu processing etc. 

Now calling that what we had for the last decades "native" is entirely subjective. There is nothing more native in a non-frame generated image than it is in a frame generated image. Why should it, both are digitally rendered image displayed on a monitor. Just the algorithms differ. But they are constantly modified and "improved", no?

My bet is: in 10y, nobody talks about frame-generation vs. "native" anymore, frame-generation will simply be the "new normal" of how digital images are rendered and displayed on a monitor. 

Every frame in a computer simulation is a "fake" frame...the difference just lies in which processor (the CPU or GPU) generates it.

14 minutes ago, Bob Scott said:

Every frame in a computer simulation is a "fake" frame...the difference just lies in which processor (the CPU or GPU) generates it.

Not sure whether to take you literally or not, but I would answer that generated frames do have some deficits in regards to latency/input lag, especially if your base/real FPS is below a certain threshold.

To some, this is common knowledge, but to many it is not.

1 minute ago, HiFlyer said:

Not sure whether to take you literally or not, but I would answer that generated frames do have some deficits in regards to latency/input lag, especially if your base/real FPS is below a certain threshold.

To some, this is common knowledge, but to many it is not.

But then the issue is latency and not generated frames, no? I mean, in the end it is semantics, as I wrote, your computer generates frames using whatever approach and displays them on the screen. With the one or the other approach you get more or less latency, if this matters, you switch the way of how those frames end up on your screen. If you cant switch, discussion would anyway be obsolete, no? Or do you think about how your "native" image is processed in detail until it arrives on your screen? Do you think about the fact that unless you really do not use any antialiasing technique at all, the image is already rendered on a complete different resolution than you finally use and thus processed inbetween? This also adds latency but todays algorithms for it are so fast that it does not really matter (in most of the cases). Many however consider those frames as native...

That is why the upcoming Battlefield 6 will have the option to turn antialiasing completely off because of the latency it adds (no matter what technique is used) and because Battlefield considers itself a competitive shooter game, it might actually be valuable for some to have the lowest possible latency. For the sake of image quality of course...

For a flight sim running at 30FPS? Or even 60FPS? You anyway have 16ms (60FPS) to 33ms (30FPS) latency. And I even read somewhere that the joystick/yoke polling rate also is only 30Hz (33ms again). So I doubt that anyone will really see any drawback of latency due to frame generation in MSFS.

So I watched the video, and I think it offers some great insights into why you don't see a 2x increase in FPS when you turn on FG. The fact is, the generated frames aren't completely free, they require some GPU processing. So in situations where you are GPU limited or your CPU/GPU is evenly matched, turning on FG is going to "steal" some GPU performance from rasterized frames to generate frames. Makes perfect sense.

But what he doesn't show is what happens when you're CPU constrained. So your CPU can only generate 60FPS, but your GPU is sitting at 50% utilization.  In that situation, I suspect you will get much closer to 2x (120) FPS without hurting rasterized frame output.  

This is evident when I run ARMA3 for example, and my base frame rate is CPU limited, as it's an old and unoptimized single-threaded game where my 4090 is sitting at low GPU utilization but my CPU is running full out and I'm only getting 40FPS. If I turn on Lossless Scaling (as Arma doesn't support DLSS), I get a true doubling of 80 FPS because the GPU has plenty of headroom.

Now when I'm running Cyberpunk, with ray tracing on, my GPU is going full out and so a 2x FG setting is not going to give me a true 2x improvement, because some of the processing for those generated frames has to come at the cost of some rasterized frames which he shows in the video.

Now, what's the case with Microsoft Flight Simulator?  Well in my experience, the key problem with performance in MSFS is when you're on approach or land at a busy airport with traffic. That puts a huge load on the CPU and that's when your FPS tanks and you're likely to see stutters. So at that time, your GPU has plenty of headroom, because the CPU can't feed it frames fast enough, so you might as well use that extra GPU processing headroom to renders some generated frames to try and smooth out the bottleneck. The rest of the time, like when you're flying at 10,000 feet, your CPU has no problem keeping up with feeding your GPU all it can render, so frame generation is going to likely have to sacrifice some rasterized frames, to generate frames. But do you care? No. 

TLDR; Frame Generation is best used in CPU limited situations where you have GPU headroom to spare to generate the added frames and provide a smoother experience. In GPU limited situations, it will cost you raster frames for generated frames, but still net you a higher FPS as shown in the video.

 

Edited August 7, 2025Aug 7 by Virtual-Chris