Guys, gamers... The more you drink the more frames you generate(hallucinate). Sometimes you wont even renember the frames the next day. Thank you for pointing out this tho. I like details.
GREAT GREAT VIDEO well done! thank you for that, excellent noticing of the falling apart of the detail. i wonder how it would look in a brighter environment.. again great eye and detail
The ominous music fits the video so well. Very good video my friend. I havent delved too much into this topic yet as image quality comparisons arent really my strong suite, but you've done a great job here.
This is more of a preview look at DLSS 4 from Digital Foundry's preview video which I have linked in the description. It's important to remember that it was NOT* (EDIT) final code according to DF and it's just me being curious about the technology and examining the generated frames and sharing my thoughts with you all so feel free to share yours below.
Thnx for this analysis. Don't forget (M)FG has major trouble with games like CP 2077 maxed out, since the base fps is around 26-27 @4K with everything maxed out (on a 5090). Games like the new(er) Doom generate much more native frames, thus (M)FG will look better. Anyway I rarely use FG on my 4090, cause the image always looks with less depth.
Yeah good point. We don't really know what the base framerate is although they were using DLSS upscaling in this video but the base framerate should play a role in it all.
I hope they can fix this with the new DLSS. I prefer something a bit less smooth over something smoother but with some extra flickering. Also the input lag can be an issue. Maybe Reflex 2 will bring a solution there.
the issue with videos like these is 1. RUclips Compression 2. Nobody really cares unless you are pixel hunting like this. Sure everyone wants Native but when push comes to shove, people are gonna use that MFG, DLSS, FSR etc.
Even after seeing this, it's crazy how good it looks. If they fix these very small details and they solve they reduce latency even more, then I'm not going to mind using this at all.
You should get everything but multi frame generation. For now anyway although the transformer model DLSS will have a higher tensor computational cost resulting in less FPS vs the CNN model of DLSS with a 4080 Super you should be pretty good vs all the other RTX cards.
Any chance you’re part of the NVIDIA creator program so you can give us an early preview of the 5090? This is the best analysis yet I’ve seen on youtube analyzing these new fake frames. Great work.
I haven't been sent any RTX 5000 GPU's. If I get one though in the future I will for sure make similar comparisons as it is something I do find interesting.
Nobody plays games in 0.5x playback speed, right? I selected 2x playback ratio and see no problem in the video. Maybe some blurry effects can be a little disgusting and input lag might be distinguishable sometimes. However, all these have solutions and probably will be fixed in further releases.
This is pretty much the first thing I pointed out when I saw people on RUclips showcasing this. I could see the trees had issues.. A recording of screen. It was that obvious. I saw also halo around characters, then the ms spikes that was pretty bad. we are talking like 80ms while walking or running. Fur, was also problematic.
I noticed that too and a reason why I made this to point it out but to also point out the majority of the screen looked really good as you would expect from a user input frame. I think over time or maybe with the final code a lot of this could be mitigated. Some games will probably lend themselves a bit better than others. Time will tell.
gone are the days of pure raster performance gains, especially 80 class or bellow. The 90 series still seems to have good gains, but you pay the premium.
It doesn't. Generators start to work well when you already have decent fps (~60) WITH satisfying graphic settings, so rather not the case with midrange cards running RT in 4K.
@2MichalChojnowski2 It does. Some getting used to might be required I guess. That being said FrameGen aside,at 4K DLSS Performance looks quite nice. Quite a bit far from true 4K True , but better than 1440p Native.
It looks horrible in this preview driver. I would say unusable. This was cool to see for an advanced analysis so thanks for the video. Looking forward to the full launch review. Likely it won't be very different. Maybe 2 years from now it will be about 50% better but still not worthwhile. That's my guess. When the full launch happens I'd love to see a comparison between AMD, NVIDIA, and lossless scaling.
I think it sucks who have no idea and haven’t used it but base their opinions on 50% speed videos specifically pointing things out. Frame Gen has been awesome on my 4090 and can’t wait to get my 5090
Another great vid, as per usual. Juggling whether I should upgrade to 50 series has been mental gymnastics. Frame generation is not all that impressive, I play most games with controller and 60 fps is very easy for me to adjust to if the frame pacing is good. Seeing as though frame gen needs baseline 60 to be optimal, the appeal gets even less for me.
Thanks. I understand what you mean. If you already have a 4070/Ti/4080 the gains aren't what we had last gen. The 5090 seems to have the most gain but even that is less than half what previous gen was and noticeably more expensive. It's a very hard sell for sure.
Hold on, let me grab my massive spectacles, 500x zoom magnifying glass, and slow my framerate to 1 fps so that I can call out Frame-gen for being useless garbage we don't need (Still buys the 5090 anyway and uses FrameGen). This is the majority of people clowning on Nvidia with the whole "fake frames" narrative... they do this every year with any new technology. People forgetting how everyone treated DLSS 1 at launch, but now everyone is using it as the standard/preferred option.
You are correct, it does need a bit of work and I feel it will get better. As for paying, only pay if something is worth it to you. There's plenty of content out there to get an overall feel of things but there's some hope it could arrive to previous RTX cards too :)
2:12 I think is because the moving texture behind it which Brian Catanzaro said was an issue with fg and AI had to be taught to "discard" it kinda..so the leaves in front of a moving texture are an issue.
Yeah I've seen all those interviews. They'll adress these things with time but it is similar problem areas for FG I have seen in other games too. Now you have 3 generated frames vs 1 which stands out more when it comes to the flaws.
Yes, the flickering is annoying, but this is not the final version, and I'm sure Nvidia knows what they did. I listened to the interview of Alex with the Nvidia guy from the other day, and he explained it pretty well.
ruclips.net/video/uyxXRXDtcPA/видео.html **Following the announcement of the RTX 5000 and its anticipated release, I’m at the Font Blue hotel in Las Vegas, sitting down with Brian Karon, Vice President of Applied Deep Learning Research at Nvidia. The last time we spoke was over a year ago when Ray Reconstruction was first introduced. Now, with the RTX 5000 and a new DLSS model, we’re seeing significant changes for the first time since 2020. It’s exciting, and I hope to learn more about it.** --- ### **1. Why switch to this new Transformer model for super-resolution?** Over the past five or six years, we’ve been evolving the super-resolution model, and it’s become increasingly challenging to make it smarter within the same constraints. We needed to innovate and try something new. Transformer models have been revolutionary for language modeling and image generation, as seen in technologies like ChatGPT and Stable Diffusion. These models are highly scalable and can be trained on large amounts of data, allowing them to make smarter decisions about what's happening in an image and what to generate. This results in breakthrough image quality with our new Ray Reconstruction and super-resolution models in DLSS 4. --- ### **2. What key image characteristics have improved with the new Transformer model?** The main issues with super-resolution are stability, ghosting, and detail. These aspects usually trade off against each other. For example, more detail can lead to ghosting, and stability problems can cause shimmering and flickering. The new model makes better trade-offs, significantly improving these areas. Additionally, the bigger model with more compute capacity is smarter and makes fewer poor choices, reducing issues like ghosting and blurring. --- ### **3. Do you see greater potential for advancement with the new model?** Absolutely. Bigger models trained on more high-quality data generally yield better results. With DLSS or any real-time graphics algorithm, we have a strict compute budget. The Blackwell GPU’s powerful tensor cores allowed us to try a more ambitious Transformer-based algorithm, leading to significant improvements. --- ### **4. How does the new model compare in performance to the old one?** The new super-resolution model has four times more compute than the old one but doesn’t take four times as long to execute, especially on Blackwell, thanks to co-designing the algorithm with the tensor cores for high efficiency. --- ### **5. How does the Transformer model improve specific areas like skin rendering in games like Cyberpunk?** The improvements come from iterative development, better data, and a more intelligent model. The new model has more capacity to learn and make smarter decisions, reducing issues like ghosting and blurring. It also better handles animated textures, which have been challenging for previous models. --- ### **6. Why did you remove hardware-based optical flow for frame generation?** We achieved better results with a fully AI-based solution. Hardware implementations are difficult to improve, and we needed a smarter neural network to address the failures of hardware optical flow. The new frame generation algorithm is more tensor core-heavy, uses less memory, has better image quality, and is more efficient, especially for multi-frame generation. --- ### **7. Why doesn’t the new frame generation run on RTX 3000?** This is primarily an optimization and engineering issue. We’re launching the best multi-frame generation technology with the 50 Series, and we’ll see what we can do with older hardware in the future. --- ### **8. What has changed with frame pacing in DLSS 4?** Blackwell has an updated flip metering system with much lower variability, reducing display frame time variability by a factor of five or ten. This is especially important for multi-frame generation, where frame time variability can significantly affect the experience. --- ### **9. Will there be differences in frame pacing between RTX 4080 and 4090?** DLSS 4 is expected to perform better on the 40 Series, so there may be differences. --- ### **10. Why doesn’t DLSS 3 support standard vsync?** This is an engineering problem. If standard vsync becomes a priority for our customers, we could likely solve it. --- ### **11. What is new with Reflex 2?** Reflex 2 is a new tool for lowering latency. It renders the scene normally but warps the image to the new camera position right before finalizing it, dramatically reducing latency. AI plays a big role in filling in the disoccluded holes that appear when the camera moves. --- ### **12. What is the future of frame generation with ultra-high refresh rate monitors?** Frame generation will be crucial for achieving ultra-high refresh rates. Neural rendering is shifting graphics towards more realistic and convincing representations, overcoming the limitations of traditional polygon-based graphics. This will open up new possibilities for more interesting and fun games. --- **That’s all my questions. Thank you for speaking with me today.** It’s an honor to speak with you. I’m a huge Digital Foundry fan, so thank you for your work.
I would be sceptical about MFG as a feature worth getting Blackwell alone. For now we need to see how it works in bunch of games, because such tech is always presented in ones it works best. Remember PSSR? Be ready for FSR4 outside Ratchet. Plus I'm personally not a fan of generators - they are not a pass to real smoothness - responsiveness - nor prettier graphics as many see them. They tend to work well when your framerate is already not that bad (~60 fps) WITH satisfying graphic settings. And then give you less responsive kind of frames with addition of artifacts. Useful in old games with framerate caps, but it's trying hard.
@@TerraWare These are the times when performance related things get complicated and we need more than ever scepticism, technical analysis and reminding such knowledge whenever the matter is talked about. These are the times when people get crazy after Jensen said 5070 offering 4090 performance, but it's like they don't want to notice that just one sentence later he added it's impossible without AI. What could only mean help of DLSS-related tech. Or some guy here in the comments thinks FG is a pass to RT in 4K on midrange cards.
Are you using the actual RUclips video or did you sign up for DF’s Patreon and downloaded the better quality version of the video? If you’re doing the former, it’s utterly pointless to look for visual artifacts in individual frames in a 60fps stream outside of key frames, as 99% of the stream is heavily compressed and uses motion vectors to reconstruct the image (in a way not too dissimilar to the way TAA and FSR work). You could still potentially discern between compression artifacts and DLSS artifacts, but it’s really hard to tell. I don’t know what compression settings DF use for their “premium” downloads but it’s definitely worth checking out if you want something closer to the ground truth.
Not to mention the fact that in reality, perceived motion clarity of 120fps with interpolation is very different from what you get by running a 2x slowed down video at 60fps, let alone frame by frame. Sure, analyzing still frames is still a robust way of identifying artifacts, but perception of movement is equally if not more important than the quality of individual “physical” images, so things like frame pacing might actually play a more significant role when doing multi frame generation than the image quality (which is honestly quite acceptable even with cheap solutions like Lossless Scaling when using the base x2 mode, and when I play the DF footage sped up 2x I honestly can’t tell any of it is generated). And then of course the question of input lag, but that’s a separate discussion, and we no it pretty much makes this unusable in any low latency esports scenarios but is okay when playing games with a controller or in slower paced mouse controlled games.
The DF footage it's actually 25% speed, they said they can only capture 4K 120FPS while the fps were actually 240FPS so judging the quality slowed down to 25% for me it's a little bit pointless. U won't notice those artifacts in actual gameplay unless u look super close to the screen pin-pointing each one
it is a Preview build of Dlls4 running on a Preview build of Cyberpunk. Also DF did preface this by saying it looks better IRL Due to limitations of 200+ FPS being cut down to 4k 120 then shown at 50% speed. So Some artifacts might not be really perceptible and its not actually panning that slowly. Granted there does seem to be very persistent effects that are very notable. ruclips.net/video/lA8DphutMsY/видео.htmlsi=S6iapoThAx88QTzk&t=690 However, there is some pretty crazy ghosting here which doesn't seems to happen in the CNN model. Also very interesting DF doesn't have any footage of Cyberpunk's main DLSS problem which still is driving and where this person found some major ghosting issues. But, dunno all of this could be fixed or in a much better state by the time it comes out, hopefully!
AMD is only launching RX 9070 in March. Will be going with the RTX 5070 if 3rd party reviews for AMD is not out by the time the 5070 is out. I've been without a GPU since December. I loved using FG with the 4070 Super I returned so I am looking forward to maxing out my 165hz monitor with Multi FG in singleplayer games
As a 4080 OG owner, I am trying to find the silver lining to all of this but most of the time I am exasperated with the emphasis on framegen and AI and what it means as a typical end user for these cards. Higher prices, at best a "sideways" not even upgrade b/c AI, and the usual crew on youtube praising multgen like a puppy for totally not ruining the experience or pooping the bed, a very low standard indeed. Hope I am wrong about this and it works out b/c I never felt so disillusioned with PC gaming.
The interpolated frames should have the same quality. IE, the 2nd generated frame is not generated from the 1st generated frame. It is generated from the same two real frames as the 1st generated frame. It just uses a different time point for its interpolation. That said, technically those generated frames can be generated at the same time but are presented to the viewer at their appropriate time.
Yes I know the generated frames are made from the two real frames. It did look like they became progressively flawed though as you made your way to the next real frame. Maybe it was just coincidence and will have to look at more examples in the future.
dlss is at a point where the image is so soft those generated frames that introduce extreme image flicker jumps straight to the eye. i was really annoyed of that, and dont like how df foundry didnt critize it. when i use fluid motion frames2 on amd for test purpose and fps from 60 to 120 the difference is not like if i render 120fps.
I noticed this in the DF video and I think MFG looks much worse than FG did in the previews. I'm sure Nvidia will be making MFG way better, but wether it wil be good at the 50-series release or if it will take until the launch of the 60-series before that remains to be seen. Honestly, most of the new tech Nvidia showed us at CES, while cool and probably a big deal in the long run, are things that won't matter much for the next 3-5 years and people buying a Blackwell GPU need to be aware of that.
I find distracting any small square flickering and even AA not beigh smooth bothers me. Concealing 3 frames is asking a LOT ! No amount of trained AI can cover that. But yeah - you can say it works good -enough
I don't know when, why or how, but I have this feeling Nvidia are going to release a new version of G-Sync (AI-Sync maybe?) in the next few years that will be designed to work directly with Frame Gen to help with the pacing of the generated frames, which in turn would result in a cleaner final 'fake' frame.
They switched from hardware optical flow accelerator to ai optical flow. So the optical flow accelerator isn't excuse now. So im curious if they'll unlock fg for older gpus, at least for rtx 3000. There's interview with Bryan Catanzaro from Nvidia on Digital Foundry channel, and Alex asked him about it. The response isn't clear, but he said "we'll see what we're able to squeeze out of older hardware". I hope they'll finally unlock fg for older gpus, at least for rtx 3000.
The reason why his response wasn't "clear" was due to the fact that he was there mainly to promote 50 series cards. Kinda defeat the purpose of giving a solid confirmation that FG will come to RTX 20 and 30 series. Although, I am more curious to know if MFG is gonna be available for RTX 40 series (maybe 3x option will do).
I understand since 90% of NVIDIA users have 10, 20 and maybe low end 30 series everyone is worried if FG will be available for those models. But as a 40 series user I'm worried more about the quality of the new FG x2, since they changed the model from optical flow to something which is unxplored
@ashralph9903 If 30 series can run fg x2, then 40 series should run mfg x3. So if they will give fg x2 for 30 or even 20 series, then people will ask for at least mfg x3 for 40 series. But I still hope they will give fg x2 for at least 30 series.
@mike_miller Well. Probably they wouldn't switch from hardware optical flow accelerator to ai optical flow if ai model would be worse than hardware solution. They said that ai optical flow is faster than hardware optical flow, and can provide more accurate informations, or something like this if i remember correctly. So i guess fg x2 will have no issues.
If you’re nitpicking this much then you’re not even trying to enjoy the game.. and this is where I just disregard. You’re also seeing it slowed down compared to full speed.
I don't quite get what switching "from optical flow to AI" means... So basically there was a good reliable tech, hardware supported, and now they just switch to some unstable, unreliable AI model for this? Honestly, I use FG x2 currently and in most games i don't notice any artefacts unless I want to stare at the screen and seek for those artifacts. But with new FG it seems like they now have artifacts that are really hard not to notice... If i'm using FG x2, will i get the same quality that we have now with optical flow?
Good point. It looks like the whole model has been revamped and doesnt use optical flow. They say its improved and I wouldn't worry too much about it not using optical flow. FSR 3 essentially uses a software method and works really well. In this video we don't really know what the base framerate was. They said it was 4K with path tracing and DLSS upscaling. If the base framerate was 40 FPS for example it could result in more artifacting with the problem areas.
I need to spend more time with LS. I've only tried the 1X frame gen in Kingdom Come and it worked pretty good with a base of 85FPS. That said I'd expect DLSS 4 to look better absolutely.
@TerraWare I'm messing with it now on cyberpunk and I got it to work but still way off to much artifacts. The trick is to limit ur framerate to lowest so it's not jumping up and down.use g sync and vrr.
Im so confused! My pc has a 1650 super. So I'm looking to upgrade this year. At first i was going to go with the 7800xt. But then i decided to go with either a 7900 xtx, or a 4080 super. I refuse to pay the insane price of a 4090 or 5090! Now, with all this dlss vs fsr stuff, i don't know what to do 🙄
If it were I, I'd wait for the 5080 at this point since its supposed to be similar price to the 4080 Super and while not much faster, the DLSS package is too good to miss out on. This is what I would do. I love AMD, I have a 7900XTX and am excited for FSR 4 it looks good but DLSS is in way too many games and its gotten even better. I will have a video up later today comparing the old CNN with Transformer DLSS model. Its really good.
@Darkthorn55 For sure. This is more of a closer look and theres no way to really truly show FG on YT the way it would look on your screen. Its kind of like VR
These examples are maybe obviously visible on slowed down videos or when showing super low framerate content. This will be 100% different scenario when the user has a 240Hz monitor and the native fps runs for example average of 80 fps. When they play the content at 240 fps level with MFG, the difference between frames are so insanely tiny that any small detail differences are not an issue.
Users won't use the MFG if it causes major screen tearing type of major visual issues. If all the frames have smooth frametimes, no tearing semi low latency, and good native quality, then the MFG experience would be great. Some might think that some of these small game artifacts are a big issue. In reality, users won't focus these between insanely fast frames, but they do see massive artifacts that pop up all the time. I just hope there are no massive game breaking artifacts/issues.
The issues are more in your face slowed down absolutely but I was able to notice some of the problems even at regular speed but there are many factors, the base framerate, your tolerance and if something stands out or not. Problem with making content about FG is trying to capture footage and display it in a format people can see because of the amount of frame rate you are dealing with, capture limitations and RUclips's 60FPS max format.
@@TerraWare I 100% agree that it's hard or just impossible to make 60 fps YT videos to show the MFG. These video examples that Nvidia uses are all low native fps content that uses MFG, because it's impossible to show the real life difference when the game runs, for example, native 80 fps at 3x multi frame gen 240fps. This is something people have to test. Even if someone had that 240 fps video, they couldn't show the viewers how it feels in real life. Because 99.9% of viewers can't ever even test MFG, they can easily just say negative stuff about it. Basic internet behavior that always goes the same way. It's about the same thing as reading news headlines and making opinion based by that.
@@PhotoJohn80 bruh 2k is surely too much but don’t act like “fake frames” are the only thing you’re paying for 😭. You’re still buying a graphics card that can do all that previous ones can do and even more, so stop with this bullshit
@@AWeGPUs5090 MSRP is 25% higher and it uses 25% more power to provide just about +30% performance-wise, compared to 4090. Fake frames or not, 5090 is probably the least exiting gaming GPU on the market. Almost no improvement in terms of price/performance ratio. This is even worse than just a "meh" product.
@ I absolutely agree. And I also thing that the 90s high end RTX cards are not really a product that actual gamers should buy. Of course it’s not a value product, but I think that enthusiasts would still buy it even if it costed 3k $ and more. Regardless of its flaws, you still can’t really say that you’re paying 2k just for fake frames. I think that in general this “fake frames” topic that people are complaining so much about is just not worth talking so much about. Amongst all the new techs that Nvidia as announced MFG is the most insignificant and least exciting, still people that are too lazy to do some research are only talking about this. What about DLSS 4 (only upscaler) amazing quality and improved performance (less VRAM used) and its backward compatibility with 30 and 20 series? What about neural shaders, faces and materials with textures’ compression (even lesser VRAM used when developers implement them) ? What about Reflex 2? What about G-Assist? What about Nvidia Ace? Are we gonna sleep on the fact that now videogame bosses have the possibility to learn from the player’s behavior and modifying their movesets by themselves instead of having scripted moveset and input reading? AMD can never compete with this immense suite of techs. And I’m not an Nvidia fanboy, I just think that rn Nvidia is offering so much more for the future of gaming compared to its competitors (this is kinda a fact and not an opinion). But people are only talking about fake frames, that is the stupidest thing to talk about rn
honestly there is extremely small chance to see those issue when actualy gaming you never move that slow i am super curious hopefully will get my hands on 5090 next month
Yeah. I have a video that should be available in about an hour or less doing a deep dive on the DLSS 4 transformer model and I do point out this issue and why it happens. It seems to be more unstable than CNN when not using RT/PT in the game. They'll probably fix it but I know exactly what you mean.
Ok. So it's crap and nVidia fanboys try to justify this crap in comments, because it's nVidia crap. And DF is sponsored by nVidia totally so they're as objective as convicted in their own trials.
It can be quite a challenge to analyze FG as technology and make a video about it. With multi frame generation it's even harder because of the amount of frames and only being able to capture at 120 FPS and YT only playing at 60 FPS. It's more of an academic look at it than what it would look like while playing if that makes sense.
Ngl all of this is cool but yall wasting yall time now as reviewers has done unboxing and testing as we speak. Benchmarks will be out in a few days so all this kinda seeing from a video, guessing will come to a end The full truth will be out then that’s when we really can compare when the fps actual numbers are up and when they have proper experience with frame gen and dlss 4 and the 5090 all together
Sure but the quality of the generated frame is quite important and not many reviewers spend as much time diving into the frame by frame part of it except for a few maybe. Ultimately this was just something I was curious about which is the motivation for most my vids.
Yeah. Tiny flickering on certain lights and writing showed on super slowed down gameplay. Absolutely unplayable. That’s definitely a con I’m willing to take to achieve 3x fps
Those aren't even competing products but depending on how they price the 9070XT it could be a pretty good product in the market. FSR 4 looks pretty good from the limited amount we have seen so far.
The best is to play after a few glasses of vodka that way you won't notice the problems
i don't think additional brain latency is good? :D
@@wanshurst2416 The latency in your brain caused by the vodka will make you feel like the game has no latency at all, trust me.
I do that every time I start up a game after work. It helps synergize with the silicone in my pc
Guys, gamers... The more you drink the more frames you generate(hallucinate). Sometimes you wont even renember the frames the next day.
Thank you for pointing out this tho. I like details.
GREAT GREAT VIDEO well done! thank you for that, excellent noticing of the falling apart of the detail. i wonder how it would look in a brighter environment.. again great eye and detail
Will be interesting to see how it does in other games and CBP 2077 too once the code is final.
The ominous music fits the video so well. Very good video my friend. I havent delved too much into this topic yet as image quality comparisons arent really my strong suite, but you've done a great job here.
It's the Cyberpunk 2077 OST. There's some good stuff in it.
Honestly I'm pretty impressed with the previews of MFG so far. Looking forward to how it improves as time goes on.
Yeah me too. Nvidia have a history with ironing out issues and this is an early look at the tech.
UE5 Devs - "60FPS at 1080P with MFG"
"With a 9090 TI Super"
This is more of a preview look at DLSS 4 from Digital Foundry's preview video which I have linked in the description. It's important to remember that it was NOT* (EDIT) final code according to DF and it's just me being curious about the technology and examining the generated frames and sharing my thoughts with you all so feel free to share yours below.
Would be very interesting to put this to the max, like a game where 5090 gets 3 fps, can it really make it 20-30 fps and how would it look?
@@svgeier With that low of a framerate it would look awful lol. It would be interesting though from an academic perspective sure.
Superb analysis
Thank you
Thnx for this analysis. Don't forget (M)FG has major trouble with games like CP 2077 maxed out, since the base fps is around 26-27 @4K with everything maxed out (on a 5090). Games like the new(er) Doom generate much more native frames, thus (M)FG will look better. Anyway I rarely use FG on my 4090, cause the image always looks with less depth.
Yeah good point. We don't really know what the base framerate is although they were using DLSS upscaling in this video but the base framerate should play a role in it all.
CP2077's FPS are not 26-27 lmao, that's without DLSS at all, the base fps should be around 70-80 and then MFG up to 240-260
I also have a 4090 and never use frame gen too. Too many artifacts, and I fell the motion is somewhat weird, specially in fps games.
I hope they can fix this with the new DLSS. I prefer something a bit less smooth over something smoother but with some extra flickering. Also the input lag can be an issue. Maybe Reflex 2 will bring a solution there.
the issue with videos like these is
1. RUclips Compression
2. Nobody really cares unless you are pixel hunting like this. Sure everyone wants Native but when push comes to shove, people are gonna use that MFG, DLSS, FSR etc.
Even after seeing this, it's crazy how good it looks. If they fix these very small details and they solve they reduce latency even more, then I'm not going to mind using this at all.
Ill just be happier with my 4080 super when 50 series launches. And we get enhanced frame generation and upscaling as well.
You should get everything but multi frame generation. For now anyway although the transformer model DLSS will have a higher tensor computational cost resulting in less FPS vs the CNN model of DLSS with a 4080 Super you should be pretty good vs all the other RTX cards.
Ok Ok.. they give us fake frames ... Sooooo can we give them fake money ?
🤣
Keep up the good work ❤
Any chance you’re part of the NVIDIA creator program so you can give us an early preview of the 5090? This is the best analysis yet I’ve seen on youtube analyzing these new fake frames. Great work.
I haven't been sent any RTX 5000 GPU's. If I get one though in the future I will for sure make similar comparisons as it is something I do find interesting.
Nobody plays games in 0.5x playback speed, right? I selected 2x playback ratio and see no problem in the video. Maybe some blurry effects can be a little disgusting and input lag might be distinguishable sometimes. However, all these have solutions and probably will be fixed in further releases.
This is pretty much the first thing I pointed out when I saw people on RUclips showcasing this. I could see the trees had issues.. A recording of screen. It was that obvious. I saw also halo around characters, then the ms spikes that was pretty bad. we are talking like 80ms while walking or running. Fur, was also problematic.
I noticed that too and a reason why I made this to point it out but to also point out the majority of the screen looked really good as you would expect from a user input frame. I think over time or maybe with the final code a lot of this could be mitigated. Some games will probably lend themselves a bit better than others. Time will tell.
gone are the days of pure raster performance gains, especially 80 class or bellow. The 90 series still seems to have good gains, but you pay the premium.
Interesting video, good job. FG is awesome technology that unlocks 4k with RT for mid-range cards but we need to be aware of it limitations.
Sure and this video is more of an academic approach anyway. If you turn it on and it looks good to you, you're the final judge.
It doesn't. Generators start to work well when you already have decent fps (~60) WITH satisfying graphic settings, so rather not the case with midrange cards running RT in 4K.
@2MichalChojnowski2
It does.
Some getting used to might be required I guess.
That being said FrameGen aside,at 4K DLSS Performance looks quite nice.
Quite a bit far from true 4K True , but better than 1440p Native.
@ What you discuss and dissagree with?
@@GameslordXY yes, I much prefer dlss performance at 4k compared to native 1440p, 4k is just more detailed even with few artifacts.
It’s only going to get better. That’s what people don’t understand. This tech is legit AF. Let’s just hope Nvidea doesn’t get super greedy.
It's really only useful on 240hz and higher displays to go from 60 fps to 240 or
*They should make AI generate specific localized section of the frame based on it's velocity then, joining them to create the whole frame.*
It looks horrible in this preview driver. I would say unusable. This was cool to see for an advanced analysis so thanks for the video. Looking forward to the full launch review. Likely it won't be very different. Maybe 2 years from now it will be about 50% better but still not worthwhile. That's my guess. When the full launch happens I'd love to see a comparison between AMD, NVIDIA, and lossless scaling.
Love the thumbnail 🤣
Sometimes they come out the way I envision it in my head, a lot of times they don't haha.
I think it sucks who have no idea and haven’t used it but base their opinions on 50% speed videos specifically pointing things out. Frame Gen has been awesome on my 4090 and can’t wait to get my 5090
Another great vid, as per usual.
Juggling whether I should upgrade to 50 series has been mental gymnastics. Frame generation is not all that impressive, I play most games with controller and 60 fps is very easy for me to adjust to if the frame pacing is good. Seeing as though frame gen needs baseline 60 to be optimal, the appeal gets even less for me.
Thanks. I understand what you mean. If you already have a 4070/Ti/4080 the gains aren't what we had last gen. The 5090 seems to have the most gain but even that is less than half what previous gen was and noticeably more expensive. It's a very hard sell for sure.
Hold on, let me grab my massive spectacles, 500x zoom magnifying glass, and slow my framerate to 1 fps so that I can call out Frame-gen for being useless garbage we don't need (Still buys the 5090 anyway and uses FrameGen).
This is the majority of people clowning on Nvidia with the whole "fake frames" narrative... they do this every year with any new technology. People forgetting how everyone treated DLSS 1 at launch, but now everyone is using it as the standard/preferred option.
It feels like the technology will eventually be perfected but we gotta pay for it 😂
You are correct, it does need a bit of work and I feel it will get better. As for paying, only pay if something is worth it to you. There's plenty of content out there to get an overall feel of things but there's some hope it could arrive to previous RTX cards too :)
People are the beta testers who pay for doing it.
@@Deuxiityeah that’s what I get from this
The more you buy....the more you save -AI Jenson
2:12 I think is because the moving texture behind it which Brian Catanzaro said was an issue with fg and AI had to be taught to "discard" it kinda..so the leaves in front of a moving texture are an issue.
Yeah I've seen all those interviews. They'll adress these things with time but it is similar problem areas for FG I have seen in other games too. Now you have 3 generated frames vs 1 which stands out more when it comes to the flaws.
Are they using depth of field. I always turn that off. Moderate performance hit but worth it.
Yes, the flickering is annoying, but this is not the final version, and I'm sure Nvidia knows what they did. I listened to the interview of Alex with the Nvidia guy from the other day, and he explained it pretty well.
ruclips.net/video/uyxXRXDtcPA/видео.html
**Following the announcement of the RTX 5000 and its anticipated release, I’m at the Font Blue hotel in Las Vegas, sitting down with Brian Karon, Vice President of Applied Deep Learning Research at Nvidia. The last time we spoke was over a year ago when Ray Reconstruction was first introduced. Now, with the RTX 5000 and a new DLSS model, we’re seeing significant changes for the first time since 2020. It’s exciting, and I hope to learn more about it.**
---
### **1. Why switch to this new Transformer model for super-resolution?**
Over the past five or six years, we’ve been evolving the super-resolution model, and it’s become increasingly challenging to make it smarter within the same constraints. We needed to innovate and try something new. Transformer models have been revolutionary for language modeling and image generation, as seen in technologies like ChatGPT and Stable Diffusion. These models are highly scalable and can be trained on large amounts of data, allowing them to make smarter decisions about what's happening in an image and what to generate. This results in breakthrough image quality with our new Ray Reconstruction and super-resolution models in DLSS 4.
---
### **2. What key image characteristics have improved with the new Transformer model?**
The main issues with super-resolution are stability, ghosting, and detail. These aspects usually trade off against each other. For example, more detail can lead to ghosting, and stability problems can cause shimmering and flickering. The new model makes better trade-offs, significantly improving these areas. Additionally, the bigger model with more compute capacity is smarter and makes fewer poor choices, reducing issues like ghosting and blurring.
---
### **3. Do you see greater potential for advancement with the new model?**
Absolutely. Bigger models trained on more high-quality data generally yield better results. With DLSS or any real-time graphics algorithm, we have a strict compute budget. The Blackwell GPU’s powerful tensor cores allowed us to try a more ambitious Transformer-based algorithm, leading to significant improvements.
---
### **4. How does the new model compare in performance to the old one?**
The new super-resolution model has four times more compute than the old one but doesn’t take four times as long to execute, especially on Blackwell, thanks to co-designing the algorithm with the tensor cores for high efficiency.
---
### **5. How does the Transformer model improve specific areas like skin rendering in games like Cyberpunk?**
The improvements come from iterative development, better data, and a more intelligent model. The new model has more capacity to learn and make smarter decisions, reducing issues like ghosting and blurring. It also better handles animated textures, which have been challenging for previous models.
---
### **6. Why did you remove hardware-based optical flow for frame generation?**
We achieved better results with a fully AI-based solution. Hardware implementations are difficult to improve, and we needed a smarter neural network to address the failures of hardware optical flow. The new frame generation algorithm is more tensor core-heavy, uses less memory, has better image quality, and is more efficient, especially for multi-frame generation.
---
### **7. Why doesn’t the new frame generation run on RTX 3000?**
This is primarily an optimization and engineering issue. We’re launching the best multi-frame generation technology with the 50 Series, and we’ll see what we can do with older hardware in the future.
---
### **8. What has changed with frame pacing in DLSS 4?**
Blackwell has an updated flip metering system with much lower variability, reducing display frame time variability by a factor of five or ten. This is especially important for multi-frame generation, where frame time variability can significantly affect the experience.
---
### **9. Will there be differences in frame pacing between RTX 4080 and 4090?**
DLSS 4 is expected to perform better on the 40 Series, so there may be differences.
---
### **10. Why doesn’t DLSS 3 support standard vsync?**
This is an engineering problem. If standard vsync becomes a priority for our customers, we could likely solve it.
---
### **11. What is new with Reflex 2?**
Reflex 2 is a new tool for lowering latency. It renders the scene normally but warps the image to the new camera position right before finalizing it, dramatically reducing latency. AI plays a big role in filling in the disoccluded holes that appear when the camera moves.
---
### **12. What is the future of frame generation with ultra-high refresh rate monitors?**
Frame generation will be crucial for achieving ultra-high refresh rates. Neural rendering is shifting graphics towards more realistic and convincing representations, overcoming the limitations of traditional polygon-based graphics. This will open up new possibilities for more interesting and fun games.
---
**That’s all my questions. Thank you for speaking with me today.**
It’s an honor to speak with you. I’m a huge Digital Foundry fan, so thank you for your work.
Doom the dark ages shouldn't even need frame generation if it's as well optimized as the last 2 were.
I would be sceptical about MFG as a feature worth getting Blackwell alone. For now we need to see how it works in bunch of games, because such tech is always presented in ones it works best. Remember PSSR? Be ready for FSR4 outside Ratchet. Plus I'm personally not a fan of generators - they are not a pass to real smoothness - responsiveness - nor prettier graphics as many see them. They tend to work well when your framerate is already not that bad (~60 fps) WITH satisfying graphic settings. And then give you less responsive kind of frames with addition of artifacts. Useful in old games with framerate caps, but it's trying hard.
@2MichalChojnowski2 I agree with a lot of it.
@@TerraWare These are the times when performance related things get complicated and we need more than ever scepticism, technical analysis and reminding such knowledge whenever the matter is talked about. These are the times when people get crazy after Jensen said 5070 offering 4090 performance, but it's like they don't want to notice that just one sentence later he added it's impossible without AI. What could only mean help of DLSS-related tech. Or some guy here in the comments thinks FG is a pass to RT in 4K on midrange cards.
Hope you get a 5090 or 5080 to try all those areas out
I'd love to get a 5090 as that is the only card that really makes sense having a 4090 to compare against but am not sure I can afford it lol.
Are you using the actual RUclips video or did you sign up for DF’s Patreon and downloaded the better quality version of the video? If you’re doing the former, it’s utterly pointless to look for visual artifacts in individual frames in a 60fps stream outside of key frames, as 99% of the stream is heavily compressed and uses motion vectors to reconstruct the image (in a way not too dissimilar to the way TAA and FSR work). You could still potentially discern between compression artifacts and DLSS artifacts, but it’s really hard to tell. I don’t know what compression settings DF use for their “premium” downloads but it’s definitely worth checking out if you want something closer to the ground truth.
Not to mention the fact that in reality, perceived motion clarity of 120fps with interpolation is very different from what you get by running a 2x slowed down video at 60fps, let alone frame by frame. Sure, analyzing still frames is still a robust way of identifying artifacts, but perception of movement is equally if not more important than the quality of individual “physical” images, so things like frame pacing might actually play a more significant role when doing multi frame generation than the image quality (which is honestly quite acceptable even with cheap solutions like Lossless Scaling when using the base x2 mode, and when I play the DF footage sped up 2x I honestly can’t tell any of it is generated).
And then of course the question of input lag, but that’s a separate discussion, and we no it pretty much makes this unusable in any low latency esports scenarios but is okay when playing games with a controller or in slower paced mouse controlled games.
The DF footage it's actually 25% speed, they said they can only capture 4K 120FPS while the fps were actually 240FPS so judging the quality slowed down to 25% for me it's a little bit pointless.
U won't notice those artifacts in actual gameplay unless u look super close to the screen pin-pointing each one
This was more of a preview of a preview? Will definitely have to look at it and other games when I finally get my hands on an RTX 5000 GPU.
it is a Preview build of Dlls4 running on a Preview build of Cyberpunk. Also DF did preface this by saying it looks better IRL Due to limitations of 200+ FPS being cut down to 4k 120 then shown at 50% speed. So Some artifacts might not be really perceptible and its not actually panning that slowly. Granted there does seem to be very persistent effects that are very notable.
ruclips.net/video/lA8DphutMsY/видео.htmlsi=S6iapoThAx88QTzk&t=690
However, there is some pretty crazy ghosting here which doesn't seems to happen in the CNN model. Also very interesting DF doesn't have any footage of Cyberpunk's main DLSS problem which still is driving and where this person found some major ghosting issues.
But, dunno all of this could be fixed or in a much better state by the time it comes out, hopefully!
AMD is only launching RX 9070 in March. Will be going with the RTX 5070 if 3rd party reviews for AMD is not out by the time the 5070 is out. I've been without a GPU since December. I loved using FG with the 4070 Super I returned so I am looking forward to maxing out my 165hz monitor with Multi FG in singleplayer games
4:00 I feel that anything lacking a motion vector is having difficulty.
As a 4080 OG owner, I am trying to find the silver lining to all of this but most of the time I am exasperated with the emphasis on framegen and AI and what it means as a typical end user for these cards. Higher prices, at best a "sideways" not even upgrade b/c AI, and the usual crew on youtube praising multgen like a puppy for totally not ruining the experience or pooping the bed, a very low standard indeed. Hope I am wrong about this and it works out b/c I never felt so disillusioned with PC gaming.
"Smooth motion" its not the fast "motion" like WE want ! Comon man its simple (((
The interpolated frames should have the same quality. IE, the 2nd generated frame is not generated from the 1st generated frame. It is generated from the same two real frames as the 1st generated frame. It just uses a different time point for its interpolation. That said, technically those generated frames can be generated at the same time but are presented to the viewer at their appropriate time.
Yes I know the generated frames are made from the two real frames. It did look like they became progressively flawed though as you made your way to the next real frame. Maybe it was just coincidence and will have to look at more examples in the future.
The 4090ti will cost 2k on the 30th 😅
Lol pretty much
dlss is at a point where the image is so soft those generated frames that introduce extreme image flicker jumps straight to the eye. i was really annoyed of that, and dont like how df foundry didnt critize it.
when i use fluid motion frames2 on amd for test purpose and fps from 60 to 120 the difference is not like if i render 120fps.
I noticed this in the DF video and I think MFG looks much worse than FG did in the previews. I'm sure Nvidia will be making MFG way better, but wether it wil be good at the 50-series release or if it will take until the launch of the 60-series before that remains to be seen.
Honestly, most of the new tech Nvidia showed us at CES, while cool and probably a big deal in the long run, are things that won't matter much for the next 3-5 years and people buying a Blackwell GPU need to be aware of that.
Some yeah, like Nvidia ACE was pretty cool. The new DLSS Transformer model should benefit us right away although at higher computational cost.
I find distracting any small square flickering and even AA not beigh smooth bothers me.
Concealing 3 frames is asking a LOT ! No amount of trained AI can cover that.
But yeah - you can say it works good -enough
We paying to get used by nvidia now?
I don't know when, why or how, but I have this feeling Nvidia are going to release a new version of G-Sync (AI-Sync maybe?) in the next few years that will be designed to work directly with Frame Gen to help with the pacing of the generated frames, which in turn would result in a cleaner final 'fake' frame.
I think we are going to see more and more software solutions and features going forward.
@@TerraWare Agreed
They switched from hardware optical flow accelerator to ai optical flow. So the optical flow accelerator isn't excuse now. So im curious if they'll unlock fg for older gpus, at least for rtx 3000. There's interview with Bryan Catanzaro from Nvidia on Digital Foundry channel, and Alex asked him about it. The response isn't clear, but he said "we'll see what we're able to squeeze out of older hardware". I hope they'll finally unlock fg for older gpus, at least for rtx 3000.
The reason why his response wasn't "clear" was due to the fact that he was there mainly to promote 50 series cards. Kinda defeat the purpose of giving a solid confirmation that FG will come to RTX 20 and 30 series. Although, I am more curious to know if MFG is gonna be available for RTX 40 series (maybe 3x option will do).
I understand since 90% of NVIDIA users have 10, 20 and maybe low end 30 series everyone is worried if FG will be available for those models. But as a 40 series user I'm worried more about the quality of the new FG x2, since they changed the model from optical flow to something which is unxplored
@ashralph9903 If 30 series can run fg x2, then 40 series should run mfg x3. So if they will give fg x2 for 30 or even 20 series, then people will ask for at least mfg x3 for 40 series. But I still hope they will give fg x2 for at least 30 series.
@mike_miller Well. Probably they wouldn't switch from hardware optical flow accelerator to ai optical flow if ai model would be worse than hardware solution. They said that ai optical flow is faster than hardware optical flow, and can provide more accurate informations, or something like this if i remember correctly. So i guess fg x2 will have no issues.
reposting this video in my own channel posts
Sure. Thanks
If you’re nitpicking this much then you’re not even trying to enjoy the game.. and this is where I just disregard. You’re also seeing it slowed down compared to full speed.
I don't quite get what switching "from optical flow to AI" means... So basically there was a good reliable tech, hardware supported, and now they just switch to some unstable, unreliable AI model for this? Honestly, I use FG x2 currently and in most games i don't notice any artefacts unless I want to stare at the screen and seek for those artifacts. But with new FG it seems like they now have artifacts that are really hard not to notice... If i'm using FG x2, will i get the same quality that we have now with optical flow?
Good point. It looks like the whole model has been revamped and doesnt use optical flow. They say its improved and I wouldn't worry too much about it not using optical flow. FSR 3 essentially uses a software method and works really well.
In this video we don't really know what the base framerate was. They said it was 4K with path tracing and DLSS upscaling. If the base framerate was 40 FPS for example it could result in more artifacting with the problem areas.
Still looks better than lossless scaling 😊
I need to spend more time with LS. I've only tried the 1X frame gen in Kingdom Come and it worked pretty good with a base of 85FPS. That said I'd expect DLSS 4 to look better absolutely.
@TerraWare I'm messing with it now on cyberpunk and I got it to work but still way off to much artifacts. The trick is to limit ur framerate to lowest so it's not jumping up and down.use g sync and vrr.
Won't your 5000 series gpu be a data center to perfect the new iterations of DLSS MF?
Well said!!! No need to comment..
Im so confused! My pc has a 1650 super. So I'm looking to upgrade this year. At first i was going to go with the 7800xt. But then i decided to go with either a 7900 xtx, or a 4080 super. I refuse to pay the insane price of a 4090 or 5090! Now, with all this dlss vs fsr stuff, i don't know what to do 🙄
If it were I, I'd wait for the 5080 at this point since its supposed to be similar price to the 4080 Super and while not much faster, the DLSS package is too good to miss out on.
This is what I would do. I love AMD, I have a 7900XTX and am excited for FSR 4 it looks good but DLSS is in way too many games and its gotten even better. I will have a video up later today comparing the old CNN with Transformer DLSS model. Its really good.
@TerraWare 🫡
Nobody plays games frame by frame. When you let the game go it looked great.
@Darkthorn55 For sure. This is more of a closer look and theres no way to really truly show FG on YT the way it would look on your screen. Its kind of like VR
we still need actual analysis not analysis of other analysis lol.
These examples are maybe obviously visible on slowed down videos or when showing super low framerate content. This will be 100% different scenario when the user has a 240Hz monitor and the native fps runs for example average of 80 fps. When they play the content at 240 fps level with MFG, the difference between frames are so insanely tiny that any small detail differences are not an issue.
Users won't use the MFG if it causes major screen tearing type of major visual issues. If all the frames have smooth frametimes, no tearing semi low latency, and good native quality, then the MFG experience would be great. Some might think that some of these small game artifacts are a big issue. In reality, users won't focus these between insanely fast frames, but they do see massive artifacts that pop up all the time. I just hope there are no massive game breaking artifacts/issues.
The issues are more in your face slowed down absolutely but I was able to notice some of the problems even at regular speed but there are many factors, the base framerate, your tolerance and if something stands out or not. Problem with making content about FG is trying to capture footage and display it in a format people can see because of the amount of frame rate you are dealing with, capture limitations and RUclips's 60FPS max format.
@@TerraWare I 100% agree that it's hard or just impossible to make 60 fps YT videos to show the MFG. These video examples that Nvidia uses are all low native fps content that uses MFG, because it's impossible to show the real life difference when the game runs, for example, native 80 fps at 3x multi frame gen 240fps. This is something people have to test. Even if someone had that 240 fps video, they couldn't show the viewers how it feels in real life.
Because 99.9% of viewers can't ever even test MFG, they can easily just say negative stuff about it. Basic internet behavior that always goes the same way. It's about the same thing as reading news headlines and making opinion based by that.
Fake Frames for 2,000. NVidia is a meme.
Lossless Scaling is 5 bucks lol. I do want to do a video in the future comparing the two hopefully.
@@TerraWare I mean obviously lossless scaling gonna lose lol.
@@PhotoJohn80 bruh 2k is surely too much but don’t act like “fake frames” are the only thing you’re paying for 😭. You’re still buying a graphics card that can do all that previous ones can do and even more, so stop with this bullshit
@@AWeGPUs5090 MSRP is 25% higher and it uses 25% more power to provide just about +30% performance-wise, compared to 4090.
Fake frames or not, 5090 is probably the least exiting gaming GPU on the market. Almost no improvement in terms of price/performance ratio. This is even worse than just a "meh" product.
@ I absolutely agree. And I also thing that the 90s high end RTX cards are not really a product that actual gamers should buy. Of course it’s not a value product, but I think that enthusiasts would still buy it even if it costed 3k $ and more. Regardless of its flaws, you still can’t really say that you’re paying 2k just for fake frames. I think that in general this “fake frames” topic that people are complaining so much about is just not worth talking so much about. Amongst all the new techs that Nvidia as announced MFG is the most insignificant and least exciting, still people that are too lazy to do some research are only talking about this. What about DLSS 4 (only upscaler) amazing quality and improved performance (less VRAM used) and its backward compatibility with 30 and 20 series? What about neural shaders, faces and materials with textures’ compression (even lesser VRAM used when developers implement them) ? What about Reflex 2? What about G-Assist? What about Nvidia Ace? Are we gonna sleep on the fact that now videogame bosses have the possibility to learn from the player’s behavior and modifying their movesets by themselves instead of having scripted moveset and input reading? AMD can never compete with this immense suite of techs. And I’m not an Nvidia fanboy, I just think that rn Nvidia is offering so much more for the future of gaming compared to its competitors (this is kinda a fact and not an opinion). But people are only talking about fake frames, that is the stupidest thing to talk about rn
honestly there is extremely small chance to see those issue when actualy gaming you never move that slow i am super curious hopefully will get my hands on 5090 next month
Cencey watch dlss 3.5/4 multi frame
I will pay 500€ for the real rendered frames and for the fake "generated" frames another 1500€ of value in trumpcoin. Deal?
Transformer is unusable for me, the bushes shimmer heavily in the Badlands with it on.
Yeah. I have a video that should be available in about an hour or less doing a deep dive on the DLSS 4 transformer model and I do point out this issue and why it happens. It seems to be more unstable than CNN when not using RT/PT in the game. They'll probably fix it but I know exactly what you mean.
Ok. So it's crap and nVidia fanboys try to justify this crap in comments, because it's nVidia crap. And DF is sponsored by nVidia totally so they're as objective as convicted in their own trials.
trash tech, lossless scaling is the best btw
Only looks good in slow motion. 😂
It can be quite a challenge to analyze FG as technology and make a video about it. With multi frame generation it's even harder because of the amount of frames and only being able to capture at 120 FPS and YT only playing at 60 FPS. It's more of an academic look at it than what it would look like while playing if that makes sense.
Ngl all of this is cool but yall wasting yall time now as reviewers has done unboxing and testing as we speak. Benchmarks will be out in a few days so all this kinda seeing from a video, guessing will come to a end
The full truth will be out then that’s when we really can compare when the fps actual numbers are up and when they have proper experience with frame gen and dlss 4 and the 5090 all together
Sure but the quality of the generated frame is quite important and not many reviewers spend as much time diving into the frame by frame part of it except for a few maybe. Ultimately this was just something I was curious about which is the motivation for most my vids.
Wow that looks awful, is the future of gaming this blurry mess?
Ill keep my 4080 super and the enhanced normal frame generation. This is dogshi7
Sorry to say Nvidia, your multi frame cash crab makes games unplayable
Yeah. Tiny flickering on certain lights and writing showed on super slowed down gameplay. Absolutely unplayable. That’s definitely a con I’m willing to take to achieve 3x fps
amd announced 9070xt march release. this will make me buy the 5090 instead. sad that i wanted amd to win
What with a name like RTX Zombie lol.
@@Steelninja77 i play left 4 dead 2. zombie
Those aren't even competing products but depending on how they price the 9070XT it could be a pretty good product in the market. FSR 4 looks pretty good from the limited amount we have seen so far.