• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

NVIDIA DLSS 4.5 Dynamic Multi-Frame Generation and 6x Mode Officially Arrive

That's gross. I was looking forward to a new LEGO game to play with my kid, we've pretty much done everything in DC Villains and Skywalker Saga.
Everything about that game made me want to stay away when they announced it, saying that we would get SEVEN playable characters. I mean half the fun of Lego games is having the hundreds of characters to choose from (at least in free play mode)
 
No. When they used both it did not cut back on development costs, it increased them.

So it was not "always about cutting back on development costs".
You always have to pay to train somebody.

That's like saying paying the team in India and the team in California at the same time, only to later fire the California team and expand the India one, wasn't about cutting costs.
 
Frame Generation relies on your CPU/GPU to perform calculations in real time.
Actually, modern Nvidia FG implementations use very little CPU time. Almost all processing is done using the GPU's Tensor Core's, hence the reason why FG is a very effective way to overcome CPU bottlenecks.

I prefer performance increases like Rebar. It's surprisingly big for RTX 20 series cards.

View: https://youtu.be/vy2tREc6pVs?si=Y56VX2_wsL0F48zZ

With the exception of the fact that ReBar can actually result in performance regressions under certain titles.
 
Actually, modern Nvidia FG implementations use very little CPU time. Almost all processing is done using the GPU's Tensor Core's, hence the reason why FG is a very effective way to overcome CPU bottlenecks.


With the exception of the fact that ReBar can actually result in performance regressions under certain titles.
Of all the games NJ Tech tested, it was just Oblivion Remastered that lost 2 fps, and Last of Us Part 2 that lost 1 fps. In other games it's a positive increase with Dead Space Remaster being a massive 26% increase. All you gotta do is install NvStrapsReBar mod. AMD has done the same thing with their older GPU's, where mods have increased FPS for GPU's that weren't meant to have ReBar. If Nvidia was interested in genuine performance increase, then something like ReBar for old cards would be better.


View: https://youtu.be/XmOapWvI9mk?si=T7H_ouHdyepUcP2B
 
AMD has done the same thing with their older GPU's, where mods have increased FPS for GPU's that weren't meant to have ReBar.
Well this is the theory, the reality is you won't always get a performance boost with ReBar enabled, running DLSS or DLSS FG you'll always get a performance boost and the artifacting issue is way overblown. With ReBar enabled 'and' DLSS and DLSS FG performance could be even better.
 
Last edited:
Well this is the theory, the reality is you won't always get a performance boost with ReBar enabled, running DLSS or DLSS FG you'll always get a performance boost and the artifacting issue is way overblown. With ReBar enabled 'and' DLSS and DLSS FG performance could be even better.
Like I said and what Hardware Unboxed said, you don't get a performance boost using frame generation. It's a placebo performance increase. Bigger frame rates without the benefits of bigger frame rates.
 
Like I said and what Hardware Unboxed said, you don't get a performance boost using frame generation. It's a placebo performance increase. Bigger frame rates without the benefits of bigger frame rates.
Our metric for measuring performance is FPS, FG increases FPS - Therefore, by definition, you get a performance boost no matter what some tech tuber born yesterday may believe. That performance boost makes FPS feel smoother as you're getting...More FPS.

The reality is: If you want to enjoy Path based Ray Tracing with all the eye candy turned up to ultra, you're gonna be using a combination of DLSS and FG as well as Ray Reconstruction no matter what the GPU used if you want playable FPS.
 
Last edited:
Our metric for measuring performance is FPS, FG increases FPS - Therefore, by definition, you get a performance boost no matter what some tech tuber born yesterday may believe. That performance boost makes FPS feel smoother as you're getting...More FPS.

The reality is: If you want to enjoy Path based Ray Tracing with all the eye candy turned up to ultra, you're gonna be using a combination of DLSS and FG as well as Ray Reconstruction no matter what the GPU used if you want playable FPS.

"performance is shitty anyway so just enable it, bro"

Like Nvidia said. The 5070 is the same performance as a 4090. In no way is the experience different, 60FPS feels fine from 15 fps native, bro just enable it, bro, our game runs at 60FPS bro just enable it.
 
"performance is shitty anyway so just enable it, bro"

Like Nvidia said. The 5070 is the same performance as a 4090. In no way is the experience different, 60FPS feels fine from 15 fps native, bro just enable it, bro, our game runs at 60FPS bro just enable it.
This is literally a scarecrow argument while taking things to the extreme. Naturally a 15fps base isn't going to scale well to 60fps using FG, everyone knows that - if this is the best argument you can muster, it's an epic fail.

DukenukemX doesn't even have an Nvidia card and openly brags about using older AMD GPU's, therefore I take his perspective on the subject with a 'huge' grain of salt considering he can't even use the technology. At the end of the day, if you want to be a die hard and stick with traditional rendering because you believe it's somehow a more pure implementation - just state you're going to stubbornly refuse to budge from traditional rendering and intend to remain stuck in the past, and the rest of us gamers won't bat an eyelid and will continue with our lives and enjoy PB RT with all the eye candy enabled.

Your refusal to advance won't stop advancement I'm afraid.
 
This is literally a scarecrow argument while taking things to the extreme. Naturally a 15fps base isn't going to scale well to 60fps using FG, everyone knows that - if this is the best argument you can muster, it's an epic fail.

DukenukemX doesn't even have an Nvidia card and openly brags about using older AMD GPU's, therefore I take his perspective on the subject with a 'huge' grain of salt considering he can't even use the technology. At the end of the day, if you want to be a die hard and stick with traditional rendering because you believe it's somehow a more pure implementation - just state you're going to stubbornly refuse to budge from traditional rendering and intend to remain stuck in the past, and the rest of us gamers won't bat an eyelid and will continue with our lives and enjoy PB RT with all the eye candy enabled.

Your refusal to advance won't stop advancement I'm afraid.

The whole "a kids Lego game needs FG to hit 30FPS" situation disagrees with you.

FG as a smoothness filter is fine, I'd even enable it in some situations. But saying it is the same as actual performance is disingenuous at best and wholly misleading at worst.


Also Nvidia didn't just SHOW, but instead BRAGGED that they took a game running at 15 FPS and 1080p and made it run at 60FPS at 4k. That's some scarecrow argument.
 
The whole "a kids Lego game needs FG to hit 30FPS" situation disagrees with you.
Which is more of a crap game problem than specifically a FG problem. Basically, you can't polish a turd.

FG as a smoothness filter is fine, I'd even enable it in some situations. But saying it is the same as actual performance is disingenuous at best and wholly misleading at worst.
We gain a sense of smoothness as a result of more fps (naturally, assuming the base fps is adequate). More fps = more performance, therefore we gain performance and even overcome CPU bottlenecks as a result of FG - Ignoring tech tubers born yesterday looking for clicks using rage bait, once you use the technology it's actually quite impressive and improving all the time.

And that's coming from someone that started as a naysayer of the technology, until I used it.

Also Nvidia didn't just SHOW, but instead BRAGGED that they took a game running at 15 FPS and 1080p and made it run at 60FPS at 4k. That's some scarecrow argument.
Yes, and we all know it's BS, therefore scarecrow is scarecrow to the extreme.
 
Which is more of a crap game problem than specifically a FG problem. Basically, you can't polish a turd.


We gain a sense of smoothness as a result of more fps (naturally, assuming the base fps is adequate). More fps = more performance, therefore we gain performance and even overcome CPU bottlenecks as a result of FG - Ignoring tech tubers born yesterday looking for clicks using rage bait, once you use the technology it's actually quite impressive and improving all the time.


Yes, and we all know it's BS, therefore scarecrow is scarecrow to the extreme.


Then I don't know what you disagree with.

FG is being used as a crutch by shitty devs to make up for optimisation

We both agree

FG was advertised misleadingly by Nvidia to equal (not approximate) real performance and by your own admission, it was BS.

We both agree

You admit that 15 to 60 with FG is not the same as 60 native.

We both agree.

But yet you still want to "uhm actually" the fact that since more frames are presented, that it is EQUAL to more performance? Because performance is measured in FPS?

Then why was SLI bad?
 
Actually, modern Nvidia FG implementations use very little CPU time. Almost all processing is done using the GPU's Tensor Core's, hence the reason why FG is a very effective way to overcome CPU bottlenecks.
Nvidia would be thrilled if that were true—yes, with ReBAR, they bypass some CPU calculations.

I haven’t read anything yet about Tensor Cores performing CPU tasks; it would be interesting to look into this.
But here’s the thing: every CPU review highlights the difference between weak and powerful CPUs in terms of FPS, which in itself indicates that there is a bottleneck resulting from the CPU.

When I jump my 9800X3D, I ran multiple benchmarks. In one of them, I discovered that the driver for the 3D V-Cache was malfunctioning—it was marked with a yellow exclamation mark in Device Manager. The result? Major graphical artifacts in some games. After applying some workarounds to fix it, the artifacts became minor, as they should be.
 
FG is being used as a crutch by shitty devs to make up for optimisation
Which is, as stated, a crap game problem and in no way a FG problem. Once again, as stated, you can't polish a turd, and no one here is arguing the game isn't a turd. At the end of the day, ever increasing vram requirements are also being used as a crutch by shitty devs to make up for optimization, meaning hardware is just becoming more and more expensive as a result of developer marketing decisions and outright laziness.

FG was advertised misleadingly by Nvidia to equal (not approximate) real performance and by your own admission, it was BS.
And when used correctly on a game capable of the required minimum fps including FG overheads, FG increases 'real world' performance. Once again, you're attempting to spin the problem as a FG problem, when it's blatantly obvious the problem is a shitty game problem.

You admit that 15 to 60 with FG is not the same as 60 native.
I think that fact's so well known within the tech community it's hardly worth mentioning. Once again, your example is a shitty game problem, not a FG problem. When used correctly on a game capable of the required minimum fps including FG overheads, FG is a very effective performance booster, especially where CPU bottlenecks are a problem, especially considering high refresh rate monitors.

But yet you still want to "uhm actually" the fact that since more frames are presented, that it is EQUAL to more performance? Because performance is measured in FPS?
You're talking about a shitty game, I'm talking about FG. We all know 15fps can't scale to 60fps using FG, it's not even a point worth mentioning - but the issue highlighted in the Lego game isn't an inherent FG problem, this is a shitty game problem. You appear to be hellbent trying to spin the issue as a FG problem when it's obvious to anyone with half a clue it's not a FG problem at all.

Then why was SLI bad?
I ran 2 x 6800GT's in SLI, 2 x GTX 275's in SLI, 2 x GTX 570's in SLI and 2 x GTX 670FTW's in SLI - They all performed great in the day. Most people trying to claim SLI was ineffective are usually so young there was literally no way they used the tech first hand in the day, so they quote inaccuracies and misconceptions while claiming them to be the truth because it's what they want to believe - Really no different to DLSS and FG today, this thread is evidence that certain members are making the same claims based on second hand misinformation and rage bait tech tube video's intended solely to rack up clicks - they haven't used DLSS or FG at all, making their opinion largely moot.

I haven’t read anything yet about Tensor Cores performing CPU tasks; it would be interesting to look into this.
But here’s the thing: every CPU review highlights the difference between weak and powerful CPUs in terms of FPS, which in itself indicates that there is a bottleneck resulting from the CPU.

Frames are inserted at GPU level, they're not pushed at CPU level - Once again, we're working on the well established assumption that base fps is adequate considering FG overheads. Therefore, FG can overcome CPU bottlenecks, allowing for frame rates that match high refresh rate monitors, allowing for a more fluid gaming experience.

When I jump my 9800X3D, I ran multiple benchmarks. In one of them, I discovered that the driver for the 3D V-Cache was malfunctioning—it was marked with a yellow exclamation mark in Device Manager. The result? Major graphical artifacts in some games. After applying some workarounds to fix it, the artifacts became minor, as they should be.

OK...I'm struggling to see the relevance here, sorry.
 
Last edited:
When FPS is decoupled from how fast a game is actually running it is no longer a performance measure of a game.

This comment makes no sense. From what I gather, you believe that if the CPU isn't pushing frames to the GPU, than the main metric we have for measuring a system's gaming performance becomes irrelevant, which is also nonsense. It's like stating "I like CPU bottlenecks, without them performance cannot be measured" - Which is, naturally, ridiculous.
 
Frames are inserted at GPU level, they're not pushed at CPU level - Once again, we're working on the well established assumption that base fps is adequate considering FG overheads. Therefore, FG can overcome CPU bottlenecks, allowing for frame rates that match high refresh rate monitors, allowing for a more fluid gaming experience.


OK...I'm struggling to see the relevance here, sorry.
I'm not sure how accurate the AI's response is, but:

DLSS 4.5 Frame Generation: CPU and GPU Roles

DLSS 4.5 Frame Generation continues to rely almost entirely on the GPU, with the CPU playing a supportive, non-generative role. Here’s how it breaks down:
1. GPU-Driven Frame Generation
  • AI Network Runs on GPU:
    DLSS 4.5 Frame Generation uses Tensor Cores to execute the AI model that generates new frames. This process is entirely GPU-bound and does not offload work to the CPU for frame creation.
    The AI model analyzes motion vectors, depth data, and previous frames to interpolate new frames, all within the GPU’s Tensor Cores.
  • No CPU Assistance in Frame Creation:
    The CPU does notparticipate in the AI inference process. Its role is limited to:
    • Providing game engine data (e.g., motion vectors, scene information) to the GPU.
    • Managing driver-level synchronization between the CPU and GPU.
2. CPU’s Role in DLSS 4.5 FG
While the CPU doesn’t generate frames, it ensures the GPU can operate efficiently:
  • Synchronization with NVIDIA Reflex:
    DLSS 4.5 integrates with NVIDIA Reflex to synchronize the CPU and GPU, minimizing latency and ensuring smooth gameplay. This is critical for maintaining responsiveness, especially in competitive or fast-paced games.
  • Reducing CPU Bottlenecks:
    By offloading frame generation to the GPU, DLSS 4.5 reduces the CPU’s workload in rendering tasks. This can free up CPU resources for other tasks like physics, AI, or background processes, indirectly improving overall system performance.
  • Data Preparation:
    The CPU prepares and sends game engine data (e.g., motion vectors, HUD elements, UI) to the GPU. This data is essential for the AI model to generate accurate frames. However, the CPU itself does not perform any AI-based frame interpolation.

So, if the processor is struggling, the model generated by the AI will be less accurate—doesn't that mean bad data or artifacts?

Because I'm saying that there will be more artifacts when there's a problem with the operating system, i.e., broken or malfunctioning drivers or files.
 
I'm not sure how accurate the AI's response is, but:



So, if the processor is struggling, the model generated by the AI will be less accurate—doesn't that mean bad data or artifacts?

AI Frames are generated using the GPU's OFA & Tensor Cores, hence the reason FG isn't supported on the RTX 20 series as OFA/Tensor performance is lacking on the RTX 20 series. Any use of the CPU is minimal, essentially providing basic logic functions.

So, if the processor is struggling, the model generated by the AI will be less accurate—doesn't that mean bad data or artifacts?

To quote myself:

Frames are inserted at GPU level, they're not pushed at CPU level - Once again, we're working on the well established assumption that base fps is adequate considering FG overheads. Therefore, FG can overcome CPU bottlenecks, allowing for frame rates that match high refresh rate monitors, allowing for a more fluid gaming experience.
Essentially, we're not talking about 15 > 60 FPS due to poorly optimized games and extreme CPU bottlenecks here. We're talking about 60 FPS > 120 FPS to match the refresh rate of one's monitor.
 
AI Frames are generated using the GPU's OFA & Tensor Cores, hence the reason FG isn't supported on the RTX 20 series as OFA/Tensor performance is lacking on the RTX 20 series. Any use of the CPU is minimal, essentially providing basic logic functions.



To quote myself:


Essentially, we're not talking about 15 > 60 FPS due to poorly optimized games and extreme CPU bottlenecks here. We're talking about 60 FPS > 120 FPS to match the refresh rate of one's monitor.
I'm talking about artifacts (glitches in the game) that appear, not about how many FPS will be generated.

1777986015712.png
 
This comment makes no sense. From what I gather, you believe that if the CPU isn't pushing frames to the GPU, than the main metric we have for measuring a system's gaming performance becomes irrelevant, which is also nonsense. It's like stating "I like CPU bottlenecks, without them performance cannot be measured" - Which is, naturally, ridiculous.
What was unclear? It's visually preferable for some people or solving another issue like VRR flicker. Padding frames to the monitor isn't making the program run any faster unless I've missed something?
 
This is literally a scarecrow argument while taking things to the extreme. Naturally a 15fps base isn't going to scale well to 60fps using FG, everyone knows that - if this is the best argument you can muster, it's an epic fail.
Except, this is exactly what the Lego game wants you to do.
DukenukemX doesn't even have an Nvidia card and openly brags about using older AMD GPU's, therefore I take his perspective on the subject with a 'huge' grain of salt considering he can't even use the technology.
Firstly, I do have Nvidia cards. They're just not used in my main rig. Also, they're not RTX cards. The fact you keep mentioning this as an excuse to ignore my claims is amazing honestly. Not just my claims but what Hardware Unboxed is claiming. Second, this doesn't change the fact that ANY frame generation technology is going to increase input lag and create blurry images with artifacts. Doesn't matter if you're using AMD's, Intel's, or any other version of Frame Generation. You can argue that Nvidia has the best implementation, but it's still a flawed technology. You are making sacrifices for using it. KazeoHin words it best, it's a Smoothness boost.
At the end of the day, if you want to be a die hard and stick with traditional rendering because you believe it's somehow a more pure implementation - just state you're going to stubbornly refuse to budge from traditional rendering and intend to remain stuck in the past, and the rest of us gamers won't bat an eyelid and will continue with our lives and enjoy PB RT with all the eye candy enabled.
Calling yourself "gamers" is a stretch. Go ahead and use fake frames and gain that input lag while I use real frames and gain the upper hand with lower input lag. Remember, low input lag is king when it comes to gaming.
Your refusal to advance won't stop advancement I'm afraid.
Go ahead, but don't expect gamers to enable this to gain a stable 30 or even 60 fps, like in that Lego game.
age of fake frames.jpg
 
I'm talking about artifacts (glitches in the game) that appear, not about how many FPS will be generated.
See this is the problem with tech tubers and clickbait video's. The artifacts are exaggerated on a still frame captures like the one in your example, with comments like "yikes" and "does this matter?" to generate clicks. Actually playing the game and viewing more than a single frozen frame (or frame and a bit), such artifacting is rarely visible. I use FG where needed and I certainly don't experience such artifacting in the case base fps is adequate considering FG overheads.

Our own brain inserts images based on memory and prediction during rapid eye movements, the brain will suppress visual input to prevent motion blur which results in a gap in our vision. The brain then fills the gap based on memory and prediction of the expected scene, providing the illusion that no sensory input was lost - This is called Saccadic Suppression and Prediction and Anticipation. Neurons in the visual cortex are also remapped during eye movement as to maintain both the current and the expected location of an object to provide a transition that's perceived as seamless. Ironically enough, this all sounds very similar to what NV FG is doing in relation to frames output to your display.

The Lego game is a dev/game issue, I'm not focused on the Lego game. I'm focused on FG in general where the base fps is adequate considering FG overheads.

What was unclear? It's visually preferable for some people or solving another issue like VRR flicker. Padding frames to the monitor isn't making the program run any faster unless I've missed something?
I didn't state it was unclear, I stated it was nonsense.

Except, this is exactly what the Lego game wants you to do.
And that's a dev/game problem, not a FG problem.

Firstly, I do have Nvidia cards. They're just not used in my main rig. Also, they're not RTX cards. The fact you keep mentioning this as an excuse to ignore my claims is amazing honestly.
So you cannot use native FG, therefore you're really in no position to comment on it's effectiveness. Posting click bait video's apparently highlighting a problem via still frames or less than ideal situations (like when base fps is too low for FG to be used effectively) doesn't change the fact you've never used NV FG. Based on your own admission that you actually don't have an Nvidia GPU capable of FG, any claims from this point forward that you've used the tech firsthand are going to be met with extreme scepticism.


You are making sacrifices for using it. KazeoHin words it best, it's a Smoothness boost.
Except in cases where base fps is adequate considering FG overheads, you're really sacrificing nothing. "Smoothness" is obtained by boosting real world fps, it's that simple. Therefore, FG is an FPS booster.

Calling yourself "gamers" is a stretch. Go ahead and use fake frames and gain that input lag while I use real frames and gain the upper hand with lower input lag. Remember, low input lag is king when it comes to gaming.
Personally I'd say the stretch is having such a forceful opinion on a technology you've never used based on second hand clickbait opinion pieces. As stated, if you want PB RT with all the eye candy turned up to max at an fps that matches high refresh rate monitors, you're gonna be using either DLSS, FG, or Ray Reconstruction - likely you'll be using all at once. In relation to input lag, the one video you posted supposedly highlighting this massive problem was downright questionable and an outright joke - The fact is, using FG 2x and Nvidia Reflex, it's been shown that latency barely changes compared to the supposed pure render path the tech tribe expect.

Go ahead, but don't expect gamers to enable this to gain a stable 30 or even 60 fps, like in that Lego game.
No one in this thread has made such a claim.
 
Last edited:
See this is the problem with tech tubers and clickbait video's. The artifacts are exagerated on a still frame capture like the one in your example with comments like "yikes" and "does this matter?" to generate clicks. When actually playing the game, such artifacting is rarely visible, I use FG where needed and I've certainly not experienced such artifacting in the case base fps is adequate considering FG overheads.


I didn't state it was unclear, I stated it was nonsense.


And that's a dev/game problem, not a FG problem.


So you cannot use native FG, therefore you're really in no position to comment on it's effectiveness. Posting click bait video's apparently highlighting a problem via still frames or less than ideal situations (like when base fps is too low for FG to be used effectively) doesn't change the fact you've never used NV FG.


Except in cases where base fps is adequate considering FG overheads, you're really sacrificing nothing. "Smoothness" is obtained by boosting real world fps, it's that simple.


Personally I'd say the stretch is having such a forceful opinion on a technology you've never used based on second hand clickbait opinion pieces. As stated, if you want PB RT with all the eye candy turned up to max at an fps that matches high refresh rate monitors, you're gonna be using either DLSS, FG or Ray Reconstruction - likely you'll be using all at once.


No one in this thread has made such a claim.
Not going to bother refuting it then?
 
Not going to bother refuting it then?
I already have.

From what I gather, you believe that if the CPU isn't pushing frames to the GPU, than the main metric we have for measuring a system's gaming performance becomes irrelevant, which is also nonsense. It's like stating "I like CPU bottlenecks, without them performance cannot be measured" - Which is, naturally, ridiculous.
I think I've essentially covered all arguments against NV FG as a technology, especially as someone that actually makes use of it where needed. If you don't like it, don't use it and try to stick with rasterization under the belief it's pure, even though titles are starting to be released that only support Ray Tracing. At this point I'm gonna nope out as people are just rehashing the same tired old arguments with evidence based on tech tubers baiting for clicks.
 
See this is the problem with tech tubers and clickbait video's.
Pointing out YouTubers does not change the information.
The artifacts are exaggerated on a still frame captures like the one in your example, with comments like "yikes" and "does this matter?" to generate clicks. When actually playing the game, such artifacting is rarely visible, I use FG where needed and I've certainly not experienced such artifacting in the case base fps is adequate considering FG overheads.
Artifacts and input lag not something you see does not mean it's not there. Also, when did we ignore image quality? Whatever happened to scrutinizing companies who screwed with the image quality of games, like with ATI's Quake 3 Quack situation?
The Lego game is a dev/game issue, I'm not focused on the Lego game. I'm focused on FG in general where the base fps is adequate considering FG overheads.
This maybe the future of frame generation technology, as a method for developers to reach frame rate goals without having to do any actual optimization. You see it as a way to reach 240Hz, while developers see it as a cost effective way to reach 30 and 60 fps.
So you cannot use native FG, therefore you're really in no position to comment on it's effectiveness. Posting click bait video's apparently highlighting a problem via still frames or less than ideal situations (like when base fps is too low for FG to be used effectively) doesn't change the fact you've never used NV FG. Any claims that you've used the tech firsthand are going to be met with extreme scepticism.
If you don't own the hardware then you can't have an opinion, is a pretty stupid statement. What about Hardware Unbox's opinion on the matter? Do we just ignore his claims because he's a YouTuber?

No one in this thread has made such a claim.
Give it time, you will. If you ever played competitively online, then you won't be using something like frame generation. This is why I've said for nearly two decades that something like cloud gaming will never work, because of the latency and decreased image quality. Frame generation technology does the same thing, but instead pushes frame rates to the hundreds. Keep in mind I'm not just criticizing DLSS 4.5 Frame Generation, but AMD's, Intel's and even the Lossless Scaling you can buy for $7.
 
Last edited:
Pointing out YouTubers does not change the information.
I'll take it a step further...Tech tubers looking for clicks are the problem. Although that could be YouTube themselves monetizing such content...Or Google....Or Alphabet.

What about Hardware Unbox's opinion on the matter? Do we just ignore his claims because he's a YouTuber?
Yep.

Not interested in the rest of your post as it's just the same tired argument on repeat.
 
See this is the problem with tech tubers and clickbait video's. The artifacts are exaggerated on a still frame captures like the one in your example, with comments like "yikes" and "does this matter?" to generate clicks. Actually playing the game and viewing more than a single frozen frame (or frame and a bit), such artifacting is rarely visible. I use FG where needed and I certainly don't experience such artifacting in the case base fps is adequate considering FG overheads.

Our own brain inserts images based on memory and prediction during rapid eye movements, the brain will suppress visual input to prevent motion blur which results in a gap in our vision. The brain then fills the gap based on memory and prediction of the expected scene, providing the illusion that no sensory input was lost - This is called Saccadic Suppression and Prediction and Anticipation. Neurons in the visual cortex are also remapped during eye movement as to maintain both the current and the expected location of an object to provide a transition that's perceived as seamless. Ironically enough, this all sounds very similar to what NV FG is doing in relation to frames output to your display.
If this is just a problem in our minds, then why is it visible in the screenshots too? :)

No, this is entirely due to a lack of synchronization or data, because the AI cores aren’t receiving the data in time and are simply displaying whatever they have on screen.
This always happens in games, even with normally rendered frames. There are so many examples where the CPU isn’t fast enough and can’t deliver the data—so single textures are missing or appear with a 0.1–1.0 second delay.
The benchmarks for Shadow of the Tomb Raider and Horizon Zero Dawn are a good example of this—just throw a lot of small objects onto a small number of or slower cores, or simply data that isn’t cached, and there you go.

It’s the same with FG here, but it’s much more critical in this case.

Screenshot_20260506_113303.pngScreenshot_20260506_113413.png
 
Last edited:
No, this is entirely due to a lack of synchronization or data, because the AI cores aren’t receiving the data in time and are simply displaying whatever they have on screen.
Firstly, I never stated the issue was in our minds. Secondly, once again, the OFA and Tensor cores aren't pushed frames from the CPU in relation to FG, all additional frames are generated by the OFA and Tensor cores on the GPU itself.

Also, once again, the very process of taking screenshots creates artifacts in the instance you're seeing one rendered frame with half a predicted inserted frame. Play the game as intended and artifacts like the ones in your screenshots are exceedingly rare regarding NV FG. You've got a 5090, try it.

Arguing about this beyond poorly optimized Lego game in relation to FG is as pointless as arguing about the video supposedly highlighting frame latency in relation to FG. One is a game/dev issue to the extreme and not an issue specific to FG, and the other was oddly questionable to the extreme and basically unrealistic - not surprising, both video's were posted by the same user.
 
Last edited:
Also, once again, the very process of taking screenshots creates artifacts in the instance you're seeing one rendered frame with half a predicted inserted frame. Play the game as intended and artifacts like the ones in your screenshots are exceedingly rare regarding NV FG. You've got a 5090, try it.

Arguing about this beyond poorly optimized Lego game in relation to FG is as pointless as arguing about the video supposedly highlighting frame latency in relation to FG. One is a game/dev issue to the extreme and not an issue specific to FG, and the other was oddly questionable to the extreme and basically unrealistic - not surprising, both video's were posted by the same user.
Yes, I had similar issues with the 9800x3D + 5090 simply because I was using a very old Windows configuration—it’s always better to reinstall the system for new hardware, especially with this stupid Windows 11 (it just keeps the old configuration).

The problem with Lego is due to Unreal Engine 5 and some lazy devs, as well as bad marketing, because they want to cover a wide range of hardware to sell more.
Just look:

1778071471477.png


I'm not sure how they'll use DLSS/FG on the GTX 960—it just doesn't support them.
And UE5 is way too heavy for the R5 1600.
So yeah, 15 FPS on that hardware is pretty normal, but FG on it? Their marketing department is staffed by people who don’t know the heck about hardware :D
 
Wanna know what else thinks Frame Generation is a smoothness maker, and not a performance maker? Valve was so concerned with frame generation that they created an overlay to show you how many frames are base and how many are generated. What this tool shows is that when you turn on Frame Generation, you do lose base frame rate. Which is why when you enable 2X you don't get 2X frames, because you lost some just by enabling Frame Generation. You lose about 15% to 20% of your base frame rate when enabling this technology. Doom Dark Ages loses 33% of it's base frame rate. Some games lose even more base frame rate. This is why some games that had low input lag at 60fps, were suddenly getting much worse input lag because a substantial amount of base frame rate was lost. This is why 2X is more like 2/3rds frame rate increase.

vavle frame generation smoothness.png

744844f333555af1f1a9b07e766747e1d64e3351.png
 
Last edited:
Wanna know what else thinks Frame Generation is a smoothness maker, and not a performance maker? Valve was so concerned with frame generation that they created an overlay to show you how many frames are base and how many are generated. What this tool shows is that when you turn on Frame Generation, you do lose base frame rate. Which is why when you enable 2X you don't get 2X frames, because you lost some just by enabling Frame Generation. You lose about 15% to 20% of your base frame rate when enabling this technology. Doom- It's certainly better than running a poorly optimized game at 15 - 60 fps with or without FG enabled. Dark Ages loses 33% of it's base frame rate. Some games lose even more base frame rate. This is why some games that had low input lag at 60fps, were suddenly getting much worse input lag because a substantial amount of base frame rate was lost. This is why 2X is more like 2/3rds frame rate increase.

View attachment 801316
View attachment 801320
I play marvel Rivals with FG 2x enabled and I finish at or near the top of the ladder with Nvidia Reflex enabled because my latency is lucky to exceed 10ms according to MangoHud. What ever minuscule latency FG adds, Nvidia Reflex effectively takes away. In many cases latency running FG + Reflex enabled can be as good as latency running with FG + Reflex disabled assuming adequate base fps considering FG overheads.

Most kiddies running bluetooth mice and WiFi have absolutely no grounds whatsoever in stating that FG adds latency, when their latency using no FG whatsoever will be a diabolical mess considering WiFi is in almost all cases running as half duplex. As for your comments regarding FG not being an FPS booster due to FG overheads like you're stating something most aren't aware of, I want to highlight the following quotes for absolute clarity:


I use FG where needed and I certainly don't experience such artifacting in the case base fps is adequate considering FG overheads.

I'm focused on FG in general where the base fps is adequate considering FG overheads.

Except in cases where base fps is adequate considering FG overheads, you're really sacrificing nothing. "Smoothness" is obtained by boosting real world fps, it's that simple. Therefore, FG is an FPS booster.

And when used correctly on a game capable of the required minimum fps including FG overheads, FG increases 'real world' performance.

I'd also like to present the following table clearely highlighting that DOOM: The Dark Ages with DLSS Quality and FG 2x enabled is not only over twice as fast as the same resolution at native with no FG, but also has essentially identical latency:

Doom Dark Ages FG.png


Assuming adequate base fps including FG overheads, more FPS = FPS booster. If we consider your extreme example regarding some crappy Lego game, FG is not a 'smoothness maker' - in providing your extreme example of a poorly coded game, you've effectively totally contradicted yourself.

*Mic drop.
 
Last edited:
Most kiddies running bluetooth mice and WiFi have absolutely no grounds whatsoever in stating that FG adds latency, when their latency using no FG whatsoever will be a diabolical mess considering WiFi is in almost all cases running as half duplex. As for your comments regarding FG not being an FPS booster due to FG overheads like you're stating something most aren't aware of, I want to highlight the following quotes home for absolute clarity:

I'd also like to present the following table clearely highlighting that DOOM: The Dark Ages with DLSS Quality and FG 2x enabled is not only over twice as fast as the same resolution at native with no FG, but also has essentially identical latency:

View attachment 801475


Assuming adequate base fps including FG overheads, more FPS = FPS booster. If we consider your extreme example regarding some crappy Lego game, FG is not a 'smoothness maker' - in providing your extreme example of a poorly coded game, you've effectively totally contradicted yourself.

*Mic drop.
Yeah, cables FTW!

Yeah, I also use FG in every game I can.

The misunderstanding here is elsewhere. FG will add FPS that will be displayed, but it can’t reduce latency the way real / DLSS FPS can.
So, to use your picture, when the FPS hits 193, the latency won’t be 25 ms, but half that.
So yes, your monitor will help you see everything better—which is a big plus—but your response time will still be the same 25 ms.

Of course, it’s another story how much that 12 ms actually helps... A mouse click under 25-30 ms can be achieved with really hard and continuous work, and of course the right mouse (a magnet mouse probably helps with that, but that’s another topic) :D
One of the reasons for enforcing bans in certain games is based on milliseconds of input from the player...
 
Most kiddies running bluetooth mice and WiFi have absolutely no grounds whatsoever in stating that FG adds latency, when their latency using no FG whatsoever will be a diabolical mess considering WiFi is in almost all cases running as half duplex.
Yes, lets add more input lag. That'll show those kiddies.
I'd also like to present the following table clearely highlighting that DOOM: The Dark Ages with DLSS Quality and FG 2x enabled is not only over twice as fast as the same resolution at native with no FG, but also has essentially identical latency:

View attachment 801475


Assuming adequate base fps including FG overheads, more FPS = FPS booster. If we consider your extreme example regarding some crappy Lego game, FG is not a 'smoothness maker' - in providing your extreme example of a poorly coded game, you've effectively totally contradicted yourself.

*Mic drop.
Problem is I've seen Doom Dark Ages get much lower frame latency than mid 20's, like in the teens. Pretty sure I've linked videos on this in previous posts here. With Doom, when you turn up the Ray-Tracing then the frame latency is going to be high regardless of what you do. Compared to Doom Eternal where you could have frame latency as low as 3.5ms.

As it's been said many times before, in that Frame Generation is a smoothness increase, not a performance increase. Also, remember there is artifacting when using Frame Generation.

View: https://youtu.be/DKm2_D6RiXE?t=1007

View: https://youtu.be/x6Zlqy70r-A?si=tH8gB3sBmHgFWAvv
One of the reasons for enforcing bans in certain games is based on milliseconds of input from the player...
Elden Ring needs this. People with high network latency needs to be banned.

View: https://youtu.be/wtkXZ1zSI3s?si=3YMWSiaXPkycvY2n
 
The misunderstanding here is elsewhere. FG will add FPS that will be displayed, but it can’t reduce latency the way real / DLSS FPS can.
So, to use your picture, when the FPS hits 193, the latency won’t be 25 ms, but half that.
So yes, your monitor will help you see everything better—which is a big plus—but your response time will still be the same 25 ms.
The graph clearly shows that with DLSS set to Quality and with FG 2x enabled, latency is 25ms at 193 fps - Which is 1ms higher than native.

As stated, playing Marvel Rivals, my latency with FG 2x enabled is lucky to crack 10ms. I can and will provide screenshots if needed, something Duke can't do.

Problem is I've seen Doom Dark Ages get much lower frame latency than mid 20's, like in the teens.

And yet that table shows this isn't the case, with latency at 24ms with DLSS and FG disabled, running native, on a 5090. Discussing this topic with someone so committed to misunderstanding me who has never used the tech in question while responding with YouTube video's of tech tubers with stupid expressions on their faces is becoming somewhat tiring. I'm not even watching the video's.

Problem is I've seen Doom Dark Ages get much lower frame latency than mid 20's, like in the teens.

And you get that smoothness increase by boosting fps via adding frames, assuming base fps is adequate considering FG overheads. FG is an FPS booster first and foremost no matter what some green tech tuner wants to believe.
 
Last edited:
The graph clearly shows that with DLSS set to Quality and with FG 2x enabled, latency is 25ms at 193 fps - Which is 1ms higher than native.

As stated, playing Marvel Rivals, my latency with FG 2x enabled is lucky to crack 10ms. I can and will provide screenshots if needed, something Duke can't do.



And yet that table shows this isn't the case, with latency at 24ms with DLSS and FG disabled, running native, on a 5090. Discussing this topic with someone so committed to misunderstanding me who has never used the tech in question while responding with YouTube video's of tech tubers with stupid expressions on their faces is becoming somewhat tiring. I'm not even watching the video's.



And you get that smoothness increase by boosting fps via adding frames, assuming base fps is adequate considering FG overheads. FG is an FPS booster first and foremost no matter what some green tech tuner wants to believe.


I think if we look at FG as a "FPS booster" we can no longer determine performance by FPS and instead we need to measure performance differently using a metric that decouples itself from end-result-frames-to-the-screen.
Let me list out some postulations, and we can see if we agree here. This may be a bit long, but bear with me here.

My thesis is that a taking a 60 FPS native game and using FG to go to 120 FPS (hypothetically) would not give you the same positive uplift in experience as playing the game natively at 120 FPS. This is something that is not disputed. You can feel free to try and argue against this, but I think you and I agree on the this point so far. Therefore the metric of "FPS" is no longer a metric useful for associating with "performance" as we now have two different experiences with theoretically the same "FPS" but offering a notably different experience, and comparing that experience is the ultimate goal of the metric of "performance".

Allow me to extrapolate:

With my understanding is that 60 FPS using FG to reach 120 FPS would give you an experience more similar to a 60 FPS game, only with smoother motion and animation.

You could say "smoother motion and animation is the natural benefit of additional FPS" but I would say it's not the only benefit of additional FPS, or even the primary benefit for many experiences. The other benefit of higher (native) FPS is responsiveness.

I would say that playing with 120 FPS using FG from 60 FPS is a better experience than using 60 FPS natively in most cases. There are some outliers, and the tech is improving, but lets assume that there is no latency or performance penalty. That FG can take 60 FPS and add a generated frame inbetween each real one to reach 120 FPS with the latency identical to 60FPS. Responsiveness is not improved using FG.

Therefore I would say:

120 FPS via FG sits somewhere in-between the experience of 60 FPS native and 120 FPS native. It's better than playing 60 FPS native, but its not as good as 120 FPS native.

Depending on the game, and the gamer, that improvement in experience can be more akin to "almost as good" as 120 FPS native, in other circumstances it's more "only a little better than" 60 FPS native. It offers the responsiveness and "connected to the game" feeling at the level of 60 FPS, but with the smoothness of 120 FPS.

I would also argue that going up to, say 4X FG to 240 FPS exaggerates the issue, where you have an experience where 120FPS native is many ways better than 240 FPS via 4x FG.

Even though the animation smoothness and fluidity is higher with 240 FPS via 4x FG, a large number of gamers would probably enjoy the 120FPS native experience in most situations due to the increased responsiveness.

Which means one could argue 240 FPS via 4X FG is an inferior experience to 120FPS native, so once again the 'value' of the experience sits somewhere between 60 FPS and 120 FPS.

I would argue that even 6X FG to 360 FPS would STILL be a worse overall experience to 120FPS rendered natively.

In fact, I don't think there is any point where you take an otherwise native 60 FPS experience and use any amount of FG to make it abjectly better than a 120FPS native experience. It's animation will be smoother, sure. But anyone who's played on a laggy TV can tell you that the latency matters A LOT.

Going from 60 FPS native to 120 FPS native would still be an abjectly better experience in every conceivable way.

At any given FPS: Using FG is inarguably inferior to that same FPS at native rendering. We're not talking enabling FG versus not, we're talking about comparing two independent experiences. X Native FPS is better than X Generated FPS.

In a perfect world Frame Gen is a net benefit. If you can enable it without any latency penalty, I don't really see any downside: but if you think it's appropriate to say that the experience of taking 60 FPS native to 120 FPS using FG is appropriately called "120FPS" then I would say that "FPS" essentially becomes a worthless metric.

If we want to honour the semantics of "frames per second" meaning "frames per second" (and I can respect that) we run into a case with understanding that even now: up to 6X frame gen being available and undoubtably more exorbitant higher multiples to come.

Eventually, I guarantee we will see FG numbers in the double-digits, to the point where there is no real limitation to how far you can crank the setting. Say a future X090 can run Cyberpunk 2078 at 500FPS at max settings using frame generation, but the X060 can also run Cyberpunk 2078 at 500FPS at max settings using frame generation. Only the X090 is using 6x FG and the X060 is using 24X FG to get there. Essentially, the idea of FPS will become synonymous with 'refresh rate': all but the most lowly, old cards essentially outputting FPS to match the monitor's refresh rate, both are maxing out the adaptive sync refresh on the monitor, but one is a vastly superior experience. Essentially, the metric of FPS no longer matters to determine which video card is going to offer a better gaming experience. So in that situation, FPS =/= performance.


So how do you measure the performance? I'm not trying to be snarky, but instead actually asking. If we get to the point where you can crank the generated frames to whatever absurd number you want, what actually will be used as a numeric metric for quickly summing up the comparative performance between experiences or hardware configurations?
 
My thesis is that a taking a 60 FPS native game and using FG to go to 120 FPS (hypothetically) would not give you the same positive uplift in experience as playing the game natively at 120 FPS. This is something that is not disputed.

Agreed. However this isn't the point of FG where base FPS is adequate considering FG overheads.

With my understanding is that 60 FPS using FG to reach 120 FPS would give you an experience more similar to a 60 FPS game, only with smoother motion and animation.

By boosting FPS to gain the desired fluid frame rates suited to high refresh rate monitors = Agreed.

120 FPS via FG sits somewhere in-between the experience of 60 FPS native and 120 FPS native. It's better than playing 60 FPS native, but its not as good as 120 FPS native.

Agreed. It's also a great way to overcome CPU bottlenecks assuming the base FPS is adequate considering FG overheads as additional frames are not pushed by the CPU.

Even though the animation smoothness and fluidity is higher with 240 FPS via 4x FG, a large number of gamers would probably enjoy the 120FPS native experience in most situations due to the increased responsiveness.

Agreed. However in the case of all eye candy maxed out and PB RT enabled this simply isn't possible at this point in time no matter what the GPU, a problem compounded by high resolution (4k+) monitors.

Which means one could argue 240 FPS via 4X FG is an inferior experience to 120FPS native, so once again the 'value' of the experience sits somewhere between 60 FPS and 120 FPS.

This was pretty much covered by another comment of yours, however...Agreed.

I would argue that even 6X FG to 360 FPS would STILL be a worse overall experience to 120FPS rendered natively.

Agreed. The greater the frames generated, the worse the experience in relation to latency - However it's not quite as bad as some here would like people to believe.

At any given FPS: Using FG is inarguably inferior to that same FPS at native rendering. We're not talking enabling FG versus not, we're talking about comparing two independent experiences. X Native FPS is better than X Generated FPS.

Considering all eye candy maxed out as well as PB RT, it's the only option in relation to high refresh rate monitors. Assuming base fps is adequate considering FG overheads, running with DLSS and/or FG enabled is better than no FG at all.

So how do you measure the performance? I'm not trying to be snarky, but instead actually asking. If we get to the point where you can crank the generated frames to whatever absurd number you want, what actually will be used as a numeric metric for quickly summing up the comparative performance between experiences or hardware configurations?

Performance is still measured in terms of outright fps + latency. As stated via personal experience, and highlighted in my supplied table regarding Doom: The Dark Ages, the latency increase at least at FG 2x is well within the margin of error compared to no FG and native resolution, with over double the fps running DLSS and FG - Which is a win for users running high refresh rate monitors.

Marvel Rivals actually reports the average FG time, and basically, anyone claiming that they can notice 6ms extra is vastly overstating their ability to perceive latency:

Marvel Rivals 4900 best mod 2.jpg
 
Last edited:
OK.

How Frames Are Generated (The Technical Process)

When you play a game, the GPU and CPU work together to generate frames in a continuous loop. Here’s how it works step-by-step:

1. Game Engine Pipeline

Every frame goes through a rendering pipeline in the game engine. The steps are:
  1. Input Processing:
    • The CPU reads player inputs (keyboard, mouse, controller).
    • Example: Pressing the “W” key to move forward.
  2. Game Logic Update:
    • The CPU updates the game world based on inputs.
    • Example: Moving the player character, updating AI behavior, or calculating physics (e.g., collisions, gravity).
  3. Scene Preparation:
    • The CPU sends commands to the GPU about what needs to be rendered.
    • Example: “Draw the player character at position X, render the trees in the distance, and apply lighting.”
  4. GPU Rendering:
    • The GPU renders the frameby:
      • Vertex Processing: Converting 3D models into 2D shapes.
      • Rasterization: Converting shapes into pixels.
      • Pixel Processing: Applying textures, lighting, shadows, and effects (e.g., reflections, anti-aliasing).
      • Output: Combining all elements into a final image.
  5. Frame Display:
    • The rendered frame is sent to the monitor to be displayed.
    • The monitor refreshes at its refresh rate (e.g., 60Hz = 60 times per second).
2. Frame Time vs. FPS
  • Frame Time: The time (in milliseconds) it takes to render one frame.
    • Formula:
      Frame Time (ms)=FPS1000
    • Example:
      • 60 FPS → Frame Time = 60 / 1000≈16.67 ms per frame.
      • 120 FPS → Frame Time = 120 / 1000≈8.33 ms per frame.
  • FPS Calculation:
    • FPS is simply the inverse of frame time.
    • Formula:
      FPS=Frame Time (ms)1000
    • Example:
      • If a frame takes 8.33 ms to render, the FPS is 120.
3. What Happens When FPS Drops?
  • If the GPU or CPU can’t keep up, frame times increase, and FPS drops.
  • Example:
    • A frame takes 33.33 ms to render → FPS = 33.33 / 1000≈30 FPS.
4. Bottlenecks: Why FPS Isn’t Infinite
The FPS is limited by the slowest part of the rendering process:
  • GPU Bottleneck: The GPU can’t render frames fast enough (common in high-resolution or graphically intense games).
  • CPU Bottleneck: The CPU can’t prepare the scene fast enough (common in games with complex AI or physics).
  • RAM/Storage Bottleneck: Not enough memory to load assets quickly (causes stuttering).
  • Monitor Bottleneck: The monitor can’t display more than its refresh rate (e.g., 60Hz = max 60 FPS, even if the GPU renders 200 FPS).
5. VSync, FreeSync, and G-Sync
These technologies sync the GPU’s output with the monitor’s refresh rate to:
  • Prevent screen tearing (when parts of the screen show different frames).
  • Smooth out FPS fluctuations (e.g., if FPS drops below the monitor’s refresh rate).
  • VSync: Syncs FPS to the monitor’s refresh rate (e.g., 60 FPS on a 60Hz monitor). Can cause input lag.
  • FreeSync/G-Sync: Adaptive sync technologies that match the GPU’s FPS to the monitor’s refresh rate dynamically, reducing lag and stuttering.
Real-World Example: Generating 60 FPS
Let’s say you’re playing a game on a 60Hz monitor:
  1. The game engine updates the world every 16.67 ms (CPU).
  2. The GPU renders the frame in 16.67 ms.
  3. The monitor displays the frame.
  4. Result: 60 FPS.
If the GPU takes 20 ms to render a frame:
  • FPS = 20 / 1000=50 FPS.
  • The monitor will display 50 FPS, and you’ll experience stuttering if it drops further.
________________________________________________________________________________
________________________________________________________________________________
________________________________________________________________________________

When fake frames (also called interpolated frames or synthetic frames) are added to increase the perceived FPS, they are not rendered by the GPU like real frames. Instead, they are generated artificially using techniques like frame interpolation or upscaling. These fake frames are inserted between real frames to make motion appear smoother, even if the actual FPS is lower.


Here’s how they fit into the FPS formula and rendering pipeline:

1. How Fake Frames Are Generated
Fake frames are created using two main techniques:

A. Frame Interpolation (Motion Interpolation)
This technique generates new frames by analyzing motion between two real frames and creating intermediate frames.

How It Works:

  1. Two Real Frames are rendered by the GPU (e.g., Frame 1 and Frame 3).
  2. Motion Analysis: The system analyzes the movement between Frame 1 and Frame 3.
  3. Fake Frame Generation: A new frame (Frame 2) is created by blending or predicting the motion.
  4. Insertion: The fake Frame 2 is inserted between the real frames.
Example:
  • Real FPS: 30 FPS (33.33 ms per frame).
  • With Frame Interpolation: 60 FPS (16.67 ms per frame).
    • The system generates one fake frame between every two real frames.
Technologies That Use This:
  • NVIDIA’s Reflex + DLSS + Frame Generation (e.g., in supported games).
  • AMD Fluid Motion Frames (FMF) (used in AMD GPUs).
  • TV Motion Interpolation (e.g., “Motion Smoothing” on TVs).
  • Intel XeSS Frame Generation.
B. Upscaling + Frame Generation (Hybrid Approach)
Some technologies combine upscaling (e.g., DLSS, FSR) with frame generation to boost both performance and smoothness.

How It Works:

  1. The GPU renders a frame at a lower resolution (e.g., 1440p).
  2. The frame is upscaled to the target resolution (e.g., 4K) using AI (DLSS) or spatial upscaling (FSR).
  3. Fake frames are generated to fill in gaps, increasing the perceived FPS.
Example:
  • Real FPS: 45 FPS at 1440p.
  • With Upscaling + Frame Generation: 90 FPS at 4K (perceived).
    • The upscaled frame is duplicated or interpolated to create fake frames.
Technologies That Use This:
  • NVIDIA DLSS 3 (Frame Generation + DLSS Super Resolution).
  • AMD FSR 3 (Frame Generation + FSR upscaling).

2. How Fake Frames Affect the FPS Formula

The actual FPS (real frames rendered by the GPU) remains the same, but the perceived FPS (what the user sees) increases because of the added fake frames.

Formula Adjustment:
Perceived FPS=Real FPS×(1+Number of Fake Frames per Real Frame)

Examples:

Real FPSFake Frames AddedPerceived FPS
30 FPS1 fake frame60 FPS
40 FPS1 fake frame80 FPS
60 FPS0 fake frames60 FPS

Key Differences:

AspectReal FramesFake Frames
SourceRendered by the GPUGenerated by software/hardware
Motion Accuracy100% accurateMay introduce artifacts or “soap opera effect”
Input LagNoneCan increase input lag
GPU LoadHigh (real rendering)None (generated by dedicated hardware)
Perceived SmoothnessSmoothSmoother (but not always better)

3. Where Fake Frames Fit in the Rendering Pipeline

Here’s how fake frames are integrated into the traditional rendering pipeline:
  1. Real Frame Rendering:
    • GPU renders Frame 1 (real).
    • GPU renders Frame 3 (real).
  2. Frame Interpolation:
    • System generates Frame 2 (fake) by analyzing motion between Frame 1 and Frame 3.
  3. Frame Insertion:
    • The sequence becomes: Frame 1 → Frame 2 (fake) → Frame 3 → Frame 4 (fake) → …
  4. Display:
    • The monitor displays the sequence at a higher perceived FPS.

Visual Representation:

Time: 0ms 16.67ms 33.33ms 50ms
Real Frames: [Frame 1] [Frame 3] [Frame 5]
Fake Frames: [Frame 2] [Frame 4]
Perceived: [Frame 1] -> [Frame 2] -> [Frame 3] -> [Frame 4] -> ...
FPS: 30 FPS (real) -> 60 FPS (perceived)

4. Pros and Cons of Fake Frames

Pros:
✅ Smoother Motion: Reduces stuttering and makes gameplay feel more fluid.
✅ Higher Perceived FPS: Improves the gaming experience without requiring more GPU power.
✅ Better for High Refresh Rate Monitors: Useful for 120Hz+ displays where real FPS might not reach the monitor’s limit.

Cons:

❌ Artifacts: Can introduce visual glitches, ghosting, or the “soap opera effect” (unnaturally smooth motion).
❌ Input Lag: Adding fake frames can increase latency, which is bad for competitive gaming.
❌ Not True Performance: Doesn’t actually improve the game’s real performance (e.g., physics, AI, or loading times).
❌ Hardware Dependency: Requires specific hardware (e.g., NVIDIA RTX 40-series for DLSS 3, AMD RX 7000 for FSR 3).

5. When Should You Use Fake Frames?

Use CaseRecommended?Why?
Single-player games✅ YesSmoother visuals improve immersion.
Competitive gaming❌ NoInput lag can hurt performance.
High refresh rate (120Hz+)✅ YesHelps reach the monitor’s refresh rate even if real FPS is lower.
Low-end hardware✅ YesBoosts perceived performance without overloading the GPU.
Cinematic experience⚠️ MaybeCan make games look too smooth (like a soap opera).

6. Technologies That Use Fake Frames

TechnologyDeveloperHow It WorksHardware Required
DLSS 3 (Frame Generation)NVIDIAAI-generated fake frames + DLSS upscalingNVIDIA RTX 40-series GPUs
FSR 3 (Frame Generation)AMDSpatial upscaling + frame interpolationAMD RX 7000, RX 6000 (with update)
Intel XeSS Frame GenerationIntelAI upscaling + frame interpolationIntel Arc GPUs
NVIDIA Reflex + DLSSNVIDIALow-latency rendering + upscalingNVIDIA RTX GPUs
AMD Fluid Motion FramesAMDFrame interpolation (no upscaling)AMD RX 7000 GPUs
TV Motion InterpolationTV ManufacturersFrame interpolation for movies/TV showsModern TVs

7. Summary: How Fake Frames Affect FPS

  1. Real FPS = Actual frames rendered by the GPU.
  2. Fake Frames = Artificially generated frames inserted between real frames.
  3. Perceived FPS= Real FPS + (Real FPS × Number of Fake Frames).
    • Example: 30 real FPS + 30 fake frames = 60 perceived FPS.
  4. Trade-offs: Smoother motion vs. potential artifacts and input lag.
  5. Use Cases: Best for single-player, immersive gaming, or high-refresh-rate monitors. Avoid for competitive gaming.

Final Answer:

Fake frames are not part of the traditional FPS formula because they aren’t rendered by the GPU. Instead, they are inserted between real frames to increase the perceived FPS (what the user sees). The formula adjusts as:

Perceived FPS=Real FPS×(1+Fake Frames per Real Frame)

While this makes motion smoother, it doesn’t improve the game’s real performance and can introduce visual artifacts or input lag. Technologies like DLSS 3, FSR 3, and AMD FMF use this technique to boost perceived smoothness without requiring more GPU power.
 
Back
Top