was about to comment this very thing. right on the head! edit: and NVIDIA had no choice but to enable RR for older GPUs because they knew if not someone would reverse engineer it
True, but anybody buying a 40 series card for frame generation is an idiot and it seems like it’s not a big enough feature for people to actually want to buy the cards
I don't think so. People are just getting more and more retarded. We have multiple examples where many services are bundled as one under one name and people don't complain that naming is unclear. It is only youtubers fault to instill in viewers this false perception of what DLSS is. That it is one function and not a bundle of functions. Like none of NVidia materials suggest that DLSS3 runs only on 40 series. All their materials speak about FG that works only on 40 series.
Of course, they want people to buy their new stuff, every company does. But you do not want to punish existing customers at the same time Daniel hit it on the head. The new hardware was meant to support this feature, but older hardware was not. To make it work on older cards might mean not taking advantage of hardware designed partially around frame generation.
Nvidia should have named them separately. DLSS = Deep Learning Super Sampling DLAA = Deep Learning Anti-Aliasing DLFG = Deep Learning Frame Generation DLRR = Deep Learning Ray Reconstruction Edit: Just for the context, In my point of view, having more names( and acronyms) for specific functionalities is better than having people confused about DLSS, its assigned version number and its problem with features parity and compatibility between gpu generations, what is/are supported with what hardware generations and what not. Because right now, many people outside of tech scene, actually think that DLSS3 means Frame generation, which is not the case. Correction: super resolution to super sampling
Most of the Nvidia card owners are not tech savvy users, they would go to buy a card and have no idea what to actually ask, they wouldn't know if they need a card with DLRR, DLAA, or maybe if a box only says DLSS, they wouldn't buy it because there's no DLFG on it, and so on. They need to brand it in easy to remember names, and DLSS3 was the choice (it wasn't a bad choice if you aim for casual gamers), the problem is that important information falls through the cracks, and then everybody suffers from it.
@@lohn2559 I think having more acronyms would’ve been better than having people confused about Frame generation and thought that it is what Nvidia touted as “DLSS3”, when it actually is not. At the very least, those acronyms can be easily associated, compatibility wise, with GPU generations, what is compatible with what and what not.
@@moldyshishkabob Considering they've already been willing to drop the SS and use the DL with DLAA, I think this would be acceptable. But it should be DLFG. Deep Learning Frame Generation. "Fast Frames" sounds kinda stupid when it's standing alongside a bunch of other DL algorithms that have pretty technical names. Only real problem with this is that it'll break the single "DLSS on/off" into like 5 separate toggles but w/e
It’s rarely a good look to rebrand after the fact, I get it, but Nvidia would do well to consider renaming their feature set, or at least decouple all non-upscaling features from the DLSS title
Even if it’s not a good long, it’s still worth it in the long run to not muddy things up. Unless it’s required to used DLSS, Frame generation needs its own name.
Could've easily just said DLSS-R that's available to all, and thats settled. Instead they drew a grey line between 3 and .5. It is deliberate marketing.
But why rebrand it if it fools people into thinking they _need_ 40 series to continue to get updated features? They aren't concerned with supporting old cards nearly as much as selling new ones.
Another way of looking at it, is the underlying model architectures. DL Super Resolution is the main upscaling & denoising model, while the Frame Generation is an attachment that interpolates additional frames and feeds it back into the denoiser. In 3.5 DL Ray Reconstruction, the denoisers are being updated to better infer ray traced outputs, hence this is a true update to the main upscaler-denoiser itself, hence it can run on all RTX GPUs.
People still confusing about DLSS... this is simple way to know the version. DLSS 1 - Original Super Resolution.. debut on 20 series GPU. (DLSS) DLSS 2 - Upgrade version of Super Resolution ... debut on 30 series GPU. (DLSS) DLSS 3 - Introduction of Frame Generation.. debut & works only on 40 Series GPU (DLSS +FG).. Super Resolution works on All RTX GPU. (DLSS) DLSS 3.5 - Introduction of Ray Reconstruction.. works on All RTX GPU (DLSS +RR) Frame Generation still lock only on 40 Series GPU. (DLSS +FG+RR)
thank u this make sense, so basically u get better version of new stuff but if u want to fully utilize and jump up fps up 40-50 u need new hardware which makes sense, u need better engine for better performance but old engine could still benifit from it but obviously not as much as new ones Great
This isn't a mistake that they want fixed. They are intentionally making the naming convention confusing in the hopes consumers will be misled into making a purchase, thinking that card will do more than it can. E.g. that a 30 series card will have frame gen via DLSS 3.5.
If it was intentionally misleading, wouldn't it make more sense leading people into buying a 40 series card because they think they need one to use anything above DLSS 2.x?
Just wanna say I recently came across your channel, and I think you make the videos interesting and the way you convey everything Is very easy to understand great work!
That official slide is really really terrible and hard to read for end user, (but to be fair, it was made to showcase what each DLSS version does, not which one is available to each card) If one were to be made for people checking which features is supported on their card, this is how it should be. Version - 40 series - 30 series - 20 series DLSS2 - yes - yes - yes DLSS3 - yes with FG - yes - yes DLSS3.5 - yes with FG and RR - yes with RR - yes with RR
The older nvidia 20/30 series won't be locked out of frame gen entirely, they're supposed to work with amd's fsr3 version. It won't be the same as frame gen native to dlss3 but will at least be an alternative for the cards lacking the hardware for baked in frame gen.
It really proves that nvidia was only locked out 20/30 series out of frame generation, because otherwise theese cards will be just fine with additional speed boost for free, and play all games with nice speed (even with 20 series will be useable with new titles ray tracing in HD) - so the players will not buy the 40 series. But i am courious what will happen after fsr3. I hope after that this they have no reason to lock this feature anymore and make available for all cards. (reminds me when they finally made raytracing avaliable for non RTX cards too)
It is possible that the Frame Generation has issues on the previous versions, could be it's simply on the wrong side of the chip to route it efficiently, lag related issues... dunno. Regardless Nvidia should release the software lock they have on it and let us fiddle with it. *I don't care* if it looks worse, I just want to play with it. It doesn't harm Nvidia if it looks worse on unsupported hardware, it's not supported, just unlock it, same as when they unlocked DXR on Pascal and Turing lite. The reason they won't do that is entirely about selling the 4000 series. That's all. If performance was bad enough on the old hardware then allowing Frame Generation would only help sell the newer cards.
I think with DLSS 2 the name/brand DLSS got so much recognition they decided to just roll with it. So now we have DLSS 3.5 which is a package containing several technologies, one of which is only available on RTX4000 and future cards. Kinda like with AMD how HyperRX is a group of several technologies.
The video is excellently ilustrated. Loving your content daniel you can cleary see how your a teacher because you explain so much in so much detail thats easy to understand. thanks :) keep up the great content
Frame generation can work on 20 and 30 series cards. But not as well because of some acceleration features. This is why nvidia is so toxic. They want to privide only the best, not give the option. Frame gen could add 50% less frames but work on older hardware and be fully functionnal on 40 series. But no, you have to pay to get the full feature. PC means options, nvidia chose the MAC treatment.
with slower cards, such as 20 or 30 series there will be a lot more of graphical glitches because of slow rendering of those generated frames and if consumers will see that mess - it will hurt technology, and sells of 40 gen, so, nvidia obviously removed it from previous gen cards.
Well, it's bad business to bring out technology that can make old cards better. Like imagine if DLSS keeps on getting better and better and you'd be able to use your 3080 for games in 10 years and still get 60 fps at like 1440p lol
@@kosmosyche taking a simple time in your life to actually look at the diagram of Ada lovelace doesn't hurt your brain to show the difference between Ada lovelace and Ampere and previous gens
That's clear now. Thx. Just a word on super resolution: we can say that there are spatial super resolution AND temporal super resolution. Therefore DLSS is not wrong. With DLSS 3 you've got spatial and temporal super resolution knowing that temporal one is only accessible to RTX 40xx.
There is nothing that is specific to "AI" cores that you can't run on compute pipelines. This applies to all of DLSS. Nvidia is playing with people but AMD is still the one getting blamed for everything...
Exactly maybe the actual architecture of the chip makes it less efficient vs the 4000 series but it’s definitely not impossible…that said who would upgrade to a 1600$ card if their 3090ti could do the same and get 60+ fps on the newest games? That’s why they’re freaking out over AMDs announcement who can do it without AI cores 😂😂😂 Nvidia=🤡💩🤡
Dedicated hardware will always run better, more efficient and will end with better results than running on all purpose hardware. AMD is doing the latter simply because they have no choice, - That is the difference.
Yeah but it does free up CUDA cores to do whatever while the Tensor cores take it to do the last steps before it leaves the GPU to go to the video out chip. AMD has the entire process on GPU cores as they don't have 2 types of incompatible cores on their GPUs.
Love the videos, keep it up. This video wasn't particularly interesting to me since I already understood the differences between each dlss version. I think this video could be very useful to anyone confused by it though.
Thank you for the explanation. Been a fan of the channel since about February right after making my first build in January. I opted for a 3070 and Ryzen 7 5800X and built around that, coming up to about 1.2k. Just for the 40 series to come out a few months later. Overall still satisfied with the build. Just wish I did a bit more research first.
Kinda crazy you had to even make this video but I guess not everyone is a builder/ into computers that is a gamer. 🤷♂️🤷♂️. Glad ya made it for people that were confused, lol.
Yeah it is kind of odd that they're basically using DLSS as a brand name for all related technologies. Also, if people find this stuff confusing, just wait until they hear about Intel CPUs.
Think of it like this: DLSS is a software, that evolves regardless (sort of) of HW. DLSSv1, DLSSv2, DLSSv3, DLSSv3.5.... in DLSSv3 Frame Generation was added, which can only be exploited with proper HW, in this case 4000 series, but it is still an iteration of the previous DLSS version only newer, doing the same things (improved) + the new features. In DLSSv3.5 Ray Reconstruction was added, which can again be exploited by HW that supports it, this time all RTX GPUs. Thats it.
Nvidia and being customer friendly would also be shocking. The more people don't know the more power Nvidia have. The reason they didn't do it the way Daniel and most people would do it, is to keep the confusion going. They simply want people to fear to miss out. It is actually psykological manipulation they are trying to perform.
*Yeah I was confused at first, I thought 3 meant frame generation at first. I wonder if there is any performance uplift for 20 series cards from DLSR 2 to 3 & 3.5 (No RT), would love to see some side by side comparisons.*
@@halcyonacoustic7366nah, not really, the biggest difference was moving from 1.0 to v2, the difference between 2 and 3 is minor, but I'm sure there's more going on behind the scenes with the AI but on the surface they look pretty similar in terms of visual quality and performance, the biggest thing for DLSS 3 is that it has the frame gen feature which is clearly Nvidia's poster child for DLSS 3, explains why they've managed to confuse a lot of people getting them to think DLSS 3 on a whole only works with the RTX 40 series... meanwhile DLSS 3.5 only just adds a better de-noiser for Ray tracing
@@halcyonacoustic7366nope, DLSS2 is Super Sampling, aka image upscaler, DLSS3 is Frame Interpolation, nothing to do with image quality. Stop thinking 3 is an evolution of 2. 3 is the lastest v2 + FrameGen 3.5 is the lastest v2 + latest v3 + RR From my experience 2.5.4(Death Stranding Director's Cut version which fixed ghosting/trail) is actually even better(the SS part) than the 3.1.1(Spider-Man Resmater, it re-introduced the ghosting for some reason) version i tried, i don't know about later versions, i swapped the DLL versions from the original DS(2.3), DS Director's Cut(2.5) and Spider-Man(3.1), and the 2.5 version was the best. DLSS doesn't update 1 -> 2 -> 3 -> 3.5, it's more like 1 -> 1.1 -> 1.2 -> 1.2.1 -> 2 -> 2.1 -> 2.1.1 -> 2.1.2 -> 2.2 -> 2.3 -> 3 -> 3.1 -> 3.5, it's incremental until they rebrand when adding a new tech. In my naming example, the jump from DLSS DLL v2.3 to DLSS DLL v3, has nothing to do with Super Sampling, it's just adding FrameGen, the same with the jump from v3.1 to v3.5, they just added Ray Reconstruction, the section of the software that does Frame Gen and Super Sampling is the same as the version prior to it, the jumps that aren't incremental are about adding new tech, the incremental jumps are the ones making it look/run better.
DLSS 3 does mean frame gen. the only people who say it doesn't are pedantic freaks like you and daniel. nvidia themselves still refer to dlss super resolution as dlss 2.
Fun Fact - FSR3 should work on RX580 and 1060. The RX590 is listed but that probably means in 1080p, 900p, say 1080p with 83% res scale. You would need to turn off upscalling - just the FG. This info is on AMD FSR3 page. You know where this is going - 600p on handheld should be ok for FSR3 - wouldn't look that great but on the smaller screen it doesn't matter that much.
@@michahojwa8132The other commenter has it correct. Polaris isn't good enough at async compute (which is what AMDs frame gen relies on) performance. You'd likely end up with a performance regression due to the increased latency for each frame
@@2intheampm512 AMD says FG works with rx590. Most of difference from rx580 can be overclocked. So that's, why im sure it can work. Also, rx580 is comparable to ps4pro. Maybe the old guy gets the update
Hi Daniel ☺👋. Firstly, love your content 🥳✊. Secondly, I agree and feel your RTX Feature table is exactly how I thought Nvidia should’ve done it as for the average Joe, Nvidia chart explanation confuses all but the more tech pc enthusiast user😇👍. Job well done …. Now we have to Nvidia, to once again, admit they goofed and change their RTX Feature table for yours 😅😂🤣😅😂🤣
I talked about the naming confusion a lot when frame generation was first announced. I couldn't believe how many people were arguing with me that it was completely clear. Also driving the point home...the file you're looking at is the frame generation DLL (that's what the G at the end means,) not the DLSS upscaling DLL.
it's the same DLL. edit: wait, it isn't, does he have a game installed that has FrameGen but not DLSS? Games doesn't usually have 2 DLSS DLL versions, they're all packaged into one.
@@MaxIronsThird DLSS is nvngx_dlss.dll, frame generation is nvngx_dlssg.dll. Every game that supports DLSS uspcaling will have the first DLL, and every game that supports frame gen will have the second. I'm not surprised that Daniel got them confused when there's just a single letter difference in the names.
FSR runs on "any" cards, been Nvidia, Intel or AMD, runs on apple silicon and even runs on "any" consoles. Tell me why developers should expend resources and time do work on DLSS tech?
The problem is people confusing DLSS feature set version with specific features. The slide from Nvidia is fine, it describes both the features included in each DLSS version, and which cards can use those features. This is purely user error
Well, thank you for the video, I hope it helps some viewer. In my case, I make that reasoning: all RTX GPU supports DLSS 3.5 - only RTX 4000 and up support all it features; below that, it's everything besides FG (DLSS 3.0 was confusing though). Again, it's like Android/iOS: your smartphone may be compatible with the latest version, but not with all its features...
It is an nvidia strategy to force those who have 2000 and 3000 series cards to buy the 4000 series and the next 5000 to benefit from frame generation. It's just that they didn't think that maybe they won't choose nvidia anymore, I think that amd with the next series will recover this difference on raytracing.
I have rtx 3060ti and i dont see myself bying anything new for next couple of years . This card will play everything at atleast 60 fps , and i never cared for ultra settings and super high resolution im fine with 1920x1080 .
@@malyman316 Getting 60+~ fps with maximum settings (with RT toggled off if included) on 1080p is a guarantee on RTX 3060TI... Provided the game is not poorly optimized, of course.
Labelling all of it DLSS, is like calling a kitchen or bathroom “the living room” just because in a particular house (composed of kitchen living room and bathroom), the living room came/appears first and is big.
Incredibly helpful explanation, thank you ! Nvidia obviously wants to keep people confused so they sell as much 4000 series cards as possible, they can't do it in another way because they're catastrophic 🤦♂ What a world we live in..
I was really hoping Nvidia had unlocked DLSS 3 for older RTX cards now... The video title is misleadig. DLSS3 is not supported by RTX 2000 and 3000. 3.5 is supported partially. But who cares if FSR 3 does frame generation as well, right? An open standard will always win.
You forgot about DLAA (unless I missed it). It works in the same tech as the super resolution but is used differently. In my opinion DLAA is the unsung hero of these features
DLAA is really just DLSS where the render frame size is the same as the display frame size. DLSS has always worked that way, it was just not recommended to be used outside the recommended source scale rez factors (1.5, 1.72, 2.0). AA is always a part of basic DLSS dataflow, scaling enabled or not.
DLSS is a versioned "package" with features. New features come with each new version, and it's the individual features that have hardware dependencies, not the package version overall.
AMD is in the same stupid boat. They claimed you can get the RX 590 to run FSR3 without frame generation. But how is it FSR3 if it doesn't have frame generation? Isn't it just FSR2.2 then?
Spot on. Really dumb naming approach. DLSS and frame generation should be disentangled. I don't think most people would be that bothered with ray reconstruction as a distinct feature. It's great but is very much an under the hood feature/uograde that only serves to confuse by using it mainstream marketing materials.
For me dlss 2 means upscaling. And it doesn’t matter which version. dlss 3 is frame gen. Dlss3.x running on rtx 2000 series is really a dlss2 version 3. It can get confusing for some ppl.
With all due respect, people who think of it like that are confusing themselves. Even if you run the latest version without frame generation or ray reconstruction enabled, you're still using DLSS 3.5, not v2. You're better off thinking of it in terms of features, as Daniel just showed, instead of in terms of version.
I just hope FSR 3.0 working on 20 and 30 series Nvidia cards pushes Nvidia to also allow DLSS 3.0 on those cards and to keep their pride do their best to outperform FSR 3.0 on those cards.
DLSS framegen won't come to 20 and 30 mainly because the DLSS Frame Gen is based on the fact that optical flow accelerator, 20 and 30 do have OF but it's not fast enough to generate frames and if you try you will get even more latency than 40 series get, they might try to copy amds HYPR-X thing which enables FSR Framegen in any dx11/12 game and leave to 40 series only just like HYPR-X is only on RDNA 3
I like how everyone with an nvidia card is excited not because amd is making frame gen but because nvidia might unlock their own solution as a reaction honestly tho that's what I hope too
@@unclexbox85 Except someone hacked the driver and demonstrated that it does in fact, at the basic level, work on at least 30 series. It had wonky artefacts and the performance gain was only 80% of what 40 series sees. But don't worry. Nvidia will eat a bullet befor walking anything back. They have chosen not to enable features, that the silicon was perfectly capable of, on certain SKUs SO MANY times in the past and never ever budged. This is the reason why they locked the open source Linux driver out via some VBIOS crypto BS. They got pretty fed up with the open source folks just enabling everything the silicon could do.
I don't understand why didn't name it dlss 2.5. Still a little confusing but it's better to say all rtx gpus support dlss 2 and 2.5, and 4000 only supports dlss 3. That would have made it so much more clear.
How much performance difference is there between DLSS2 and DLSS 3 on 20 and 30 series? Anyway to test that? It'll be hard to tell imo because I guess we don't know if it's just general perfornance improvements or newer DLSS updates. I'm not a super technical guy so would appreciate if someone could help me explain this better.
Just because the dll doesn't crash on a 20 or 30 card doesn't mean that all of the functionality is implemented or that the new code-paths are actually being used. It's a black box. Heck, maybe the .dll detects the old card and sends the work off to some common system dll that came with the driver. Am I missing information that truly shows that the rendered frames are going through updated algorithms?
Currently getting around 50-70 fps on almost maxed out settings on my RTX 3080 rig running at 1440p. How much of a boost can I expect without the frame generation tech?
Got it, so at the very least my 3060 will run DLSS better with every update, as in, better upscaling and better FPS with this exact same card, with that, I'll give the DLSS 3 naming a pass :) can't wait for DLSS 4.0
It would less confusing if they just made frame generation it’s own separate thing and the ray reconstruction should have been called either DLSS 2.5 or 3
Thanks for the clarity. I was wondering if this DLSS update just upgrades all supported games with Ray construction or does that only support new games announced. Would be very cool if this retroactively upgraded all DLSS with ray reconstruction
Is it possible to use FSR 3 frame gen with DLSS 3.5 without the NVIDIA specific frame gen? For example running 20 series cards FSR3 + DLSS 3.5( The super sampling and ray reconstruction ) any info on that?
Thank you so much for clarifying this. In theory can DLSS generate more frames for the Leman, because running your game at a lower internal resolution gets you more frames out of it once it's scaled up?
its simple , DLSS version: x.x is supported by all rtx cards , DLSS version : x.x + Frame generation is only for rtx 40 series ( in this case x.x is ≥ 3 as frame gen was released with dlss 3 ), so DLSS no metter what version is supported in all rtx cards , but you need an rtx 40 series card for that + frame generation feature. note : we may see another features in the upcoming dlss versions but it may or may not be exclusive to some rtx serie of cards just like FG, so get used to it.
So to get this straight Even If I have an Nvidia 2060, I should always download the latest DLSS and replace it in my game files to get better performance? So the new DLSS 3.5.0 would be better than DLSS 2.5.1 even for my 2060?
i think a better name for "Frame generation" is Frame Tweening Animators use this to make animations smoother at lower animation fps especially in stick figure animations or simple ones
I hope the video helps some people understand this more clearly. I think you were already communicating well on the topic. Honestly, while I think the branding could be clearer, I haven’t ever been confused by Nvidia’s presentation of the features, only frustrated by the way some people talk about them on social media. When Nvidia talked about frame generation, they used the words frame generation specifically, even in their early slides. It was only on social media that some people started using the inaccurate shorthand DLSS 3 to specifically refer to frame generation. To use a geometric example, if I was using vector drawing software and they added a new pentagon primitive drawing option so that I could quickly add pentagons, I would talk about that as “the pentagon drawing tool” not the “updated version X polygon tool” even if it were part of the new polygonal tools in that version. It’s not the category that the tools are grouped under that makes the difference - it’s the lack of specificity and precision in the way that people talk about them.
Been trying to explain this to some confused people as well, that DLSS 3 works with the older cards, it's just the frame gen feature that's restricted, turning off frame gen doesn't just send you back to DLSS 2
It means along with the 3000 series, I will probably be skipping the 4000 series, as well as AMD’s current offerings. I scored when I got my 2070s ftw3 before the pandemic and before the GPU embargo. I gamed my arse off through the pandemic and now I gliding right through 2023 in 1440p.
When you game 1440p on a 4K tv at a distance you can hardly tell the difference. I lounge back on my lazy boy and experiment all the time between the resolutions and come to the same conclusion on 99% of games which is it doesn’t matter a lot. I do 1440p performance dlss and it looks amazing. I’m sitting back about 10ft though. If it was a monitor and I was resting my nose on the screen that might be a different story. For now my 3070 gets 60fps on cyberpunk with ultra RT so screw it haha
@@TheWancer it’s the rumors are true and it’s just a continuation of Ada Lovelace architecture than yeah lol As long as I can play demanding games like cyberspunk at max settings 1440p there’s no reason to give nvidia my hard earned dough 🖕🏽
I downloaded the dlss mod for starfield and ive got significant performance improvements, especially since the nvidia driver update. And im on a laptop with a 2070 max-q. I cant wait for an official dlss update from bethesda for the game, should work better.
Confusion of the highest orda. I honestly think nVidia themselves are confused. Confucius say, "If you want to confuse your customers, confuse yourself first".
"Super resolution" is the tech that enables games to run at higher resolution on lower resolution displays by downscaling. "Super sampling" or "Super scaling" is the tech that enables low resolution to run at higher resolution quality by upscaling.
That's a good point; DLSS = Deep Learning SUPER SAMPLING (as a form of anti-aliasing) DLSS 3 has nothing to do with super sampling, it shouldn't be called DLSS.
NVIDIA expects us to believe that the tiny 4060 die has more flow accelerators than the massive 3090ti die. That's why the 4060 supports frame generation whilst the 3090ti does not. The 3090ti... a card 2-3x as strong as the shitty 4060.... The reality is, frame generation is software locked to sell RTX 40 series cards. That's the only reason it doesn't work on 20 and 30 series cards.
Yes, that's how computer architectures work. New ones have capabilities that older ones don't. The 3090 also doesn't have AV1 capability, yet it still has a really good media encoder.
@@ryanspencer6778 The die is 160mm for the 4060. The die for the 3090 is over 600mm. You understand the raw numbers here. Even with a newer node, and all the number fudging in the world, the 4060 doesn't have more of anything that the 3090. Maybe that's why the performance of the 4060 is so terrible, they wasted most of the die on flow accelerators. I don't believe it though. 82 RT cores vs 24. Every single metric just makes the 4060 look pathetic in comparison. Which it is. Why would it magically have an tonne extra of one magical component? Think about it. It doesn't make sense.
FG is 4xxx only. Rest works on every RTX GPU. How's this confusing? I understand what you're saying and where you're coming from and i agree that the naming scheme is a bit of a mess, however, perhaps ppl who actually get confused by this should get a brain instead of a GPU?
The naming confusion isn't a bug. It's a feature. They want as many people to buy 4000 series cards as possible to "avoid confusion".
This man here absolutely gets it.
was about to comment this very thing. right on the head!
edit: and NVIDIA had no choice but to enable RR for older GPUs because they knew if not someone would reverse engineer it
True, but anybody buying a 40 series card for frame generation is an idiot and it seems like it’s not a big enough feature for people to actually want to buy the cards
I don't think so. People are just getting more and more retarded. We have multiple examples where many services are bundled as one under one name and people don't complain that naming is unclear. It is only youtubers fault to instill in viewers this false perception of what DLSS is. That it is one function and not a bundle of functions. Like none of NVidia materials suggest that DLSS3 runs only on 40 series. All their materials speak about FG that works only on 40 series.
Of course, they want people to buy their new stuff, every company does. But you do not want to punish existing customers at the same time Daniel hit it on the head. The new hardware was meant to support this feature, but older hardware was not. To make it work on older cards might mean not taking advantage of hardware designed partially around frame generation.
Nvidia should have named them separately.
DLSS = Deep Learning Super Sampling
DLAA = Deep Learning Anti-Aliasing
DLFG = Deep Learning Frame Generation
DLRR = Deep Learning Ray Reconstruction
Edit: Just for the context, In my point of view, having more names( and acronyms) for specific functionalities is better than having people confused about DLSS, its assigned version number and its problem with features parity and compatibility between gpu generations, what is/are supported with what hardware generations and what not. Because right now, many people outside of tech scene, actually think that DLSS3 means Frame generation, which is not the case.
Correction: super resolution to super sampling
Most of the Nvidia card owners are not tech savvy users, they would go to buy a card and have no idea what to actually ask, they wouldn't know if they need a card with DLRR, DLAA, or maybe if a box only says DLSS, they wouldn't buy it because there's no DLFG on it, and so on.
They need to brand it in easy to remember names, and DLSS3 was the choice (it wasn't a bad choice if you aim for casual gamers), the problem is that important information falls through the cracks, and then everybody suffers from it.
Yeah people would love more acronyms you’re right
*DLSS = Deep learning Super Sampling
In the very least separate DLFG and dlss so it's more clear what features are locked to 40 series and not
@@lohn2559 I think having more acronyms would’ve been better than having people confused about Frame generation and thought that it is what Nvidia touted as “DLSS3”, when it actually is not. At the very least, those acronyms can be easily associated, compatibility wise, with GPU generations, what is compatible with what and what not.
Nvidia should just call frame generation something like DLFF (Fast Frames) and not have it tangled with DLSS.
100%
If they want to keep the same branding for the suite of features, the "DL" prefix should be plenty.
DLSS, DLFF, DLRR...
@@moldyshishkabob DILF
@@moldyshishkabob Considering they've already been willing to drop the SS and use the DL with DLAA, I think this would be acceptable. But it should be DLFG. Deep Learning Frame Generation. "Fast Frames" sounds kinda stupid when it's standing alongside a bunch of other DL algorithms that have pretty technical names.
Only real problem with this is that it'll break the single "DLSS on/off" into like 5 separate toggles but w/e
@@Tosslehoffe That's very fair, I think DLFG sounds good.
DLFF = DL Fake Frames
It’s rarely a good look to rebrand after the fact, I get it, but Nvidia would do well to consider renaming their feature set, or at least decouple all non-upscaling features from the DLSS title
They already decoupled DLAA from DLSS in terms of branding. I see no reason why we couldn't get a DLSS, DLAA, DLFG, and DLRR toggle.
Even if it’s not a good long, it’s still worth it in the long run to not muddy things up. Unless it’s required to used DLSS, Frame generation needs its own name.
@@Tosslehoffe I actually think this is a good and much clearer way.
Could've easily just said DLSS-R that's available to all, and thats settled. Instead they drew a grey line between 3 and .5. It is deliberate marketing.
But why rebrand it if it fools people into thinking they _need_ 40 series to continue to get updated features? They aren't concerned with supporting old cards nearly as much as selling new ones.
Another way of looking at it, is the underlying model architectures. DL Super Resolution is the main upscaling & denoising model, while the Frame Generation is an attachment that interpolates additional frames and feeds it back into the denoiser. In 3.5 DL Ray Reconstruction, the denoisers are being updated to better infer ray traced outputs, hence this is a true update to the main upscaler-denoiser itself, hence it can run on all RTX GPUs.
People still confusing about DLSS... this is simple way to know the version.
DLSS 1 - Original Super Resolution.. debut on 20 series GPU. (DLSS)
DLSS 2 - Upgrade version of Super Resolution ... debut on 30 series GPU. (DLSS)
DLSS 3 - Introduction of Frame Generation.. debut & works only on 40 Series GPU (DLSS +FG).. Super Resolution works on All RTX GPU. (DLSS)
DLSS 3.5 - Introduction of Ray Reconstruction.. works on All RTX GPU (DLSS +RR) Frame Generation still lock only on 40 Series GPU. (DLSS +FG+RR)
thank u this make sense, so basically u get better version of new stuff but if u want to fully utilize and jump up fps up 40-50 u need new hardware which makes sense, u need better engine for better performance but old engine could still benifit from it but obviously not as much as new ones Great
This isn't a mistake that they want fixed. They are intentionally making the naming convention confusing in the hopes consumers will be misled into making a purchase, thinking that card will do more than it can.
E.g. that a 30 series card will have frame gen via DLSS 3.5.
..but its not confusing lol
If it was intentionally misleading, wouldn't it make more sense leading people into buying a 40 series card because they think they need one to use anything above DLSS 2.x?
Just wanna say I recently came across your channel, and I think you make the videos interesting and the way you convey everything Is very easy to understand great work!
I believe he's a teacher, so that makes sense 😅
@@乙口尺口-u3m This is true, but I've had some teachers that did not explain things well. But yes him being a Teacher definitely helps haha
@@乙口尺口-u3m hes a math teacher yes :D
ok
@@乙口尺口-u3myup, a maths teacher, i think he works in a elementary school but i'm not sure
That official slide is really really terrible and hard to read for end user, (but to be fair, it was made to showcase what each DLSS version does, not which one is available to each card)
If one were to be made for people checking which features is supported on their card, this is how it should be.
Version - 40 series - 30 series - 20 series
DLSS2 - yes - yes - yes
DLSS3 - yes with FG - yes - yes
DLSS3.5 - yes with FG and RR - yes with RR - yes with RR
The best one!
It literally states, "frame generation 40 series gpus". If that is confusing you're just really dumb
It is wrong If DLSS3=FG then how it is applicable to 20, 30 series???
Ok but how are you going to explain that people without rtx 4000 cards will be running DLSS 2 and DLSS 3.5 but not DLSS 3
Ooohhh! Thats clears up some of the confusion I had when some mod authors were talking about their mod being DLSS 3 working on all RTX cards.
The really complicated part apparently is frame gen is NOT dlss, they are two features that can be used separately or together, or not at all
The older nvidia 20/30 series won't be locked out of frame gen entirely, they're supposed to work with amd's fsr3 version. It won't be the same as frame gen native to dlss3 but will at least be an alternative for the cards lacking the hardware for baked in frame gen.
It really proves that nvidia was only locked out 20/30 series out of frame generation, because otherwise theese cards will be just fine with additional speed boost for free, and play all games with nice speed (even with 20 series will be useable with new titles ray tracing in HD) - so the players will not buy the 40 series. But i am courious what will happen after fsr3. I hope after that this they have no reason to lock this feature anymore and make available for all cards. (reminds me when they finally made raytracing avaliable for non RTX cards too)
I have always appreated your videos since day one! you THE MAN
It is possible that the Frame Generation has issues on the previous versions, could be it's simply on the wrong side of the chip to route it efficiently, lag related issues... dunno.
Regardless Nvidia should release the software lock they have on it and let us fiddle with it. *I don't care* if it looks worse, I just want to play with it. It doesn't harm Nvidia if it looks worse on unsupported hardware, it's not supported, just unlock it, same as when they unlocked DXR on Pascal and Turing lite.
The reason they won't do that is entirely about selling the 4000 series. That's all. If performance was bad enough on the old hardware then allowing Frame Generation would only help sell the newer cards.
I think with DLSS 2 the name/brand DLSS got so much recognition they decided to just roll with it. So now we have DLSS 3.5 which is a package containing several technologies, one of which is only available on RTX4000 and future cards.
Kinda like with AMD how HyperRX is a group of several technologies.
this is definitely the case
I don't see what is confusing. Frame generation is only for 40 series graphics cards, everything else is for every RTX series. Thats all.
The confusing part is NVIDIA conflated it with DLSS 3, so people think older cards can do Frame Generation too.
The video is excellently ilustrated. Loving your content daniel you can cleary see how your a teacher because you explain so much in so much detail thats easy to understand. thanks :) keep up the great content
Frame generation can work on 20 and 30 series cards. But not as well because of some acceleration features. This is why nvidia is so toxic. They want to privide only the best, not give the option. Frame gen could add 50% less frames but work on older hardware and be fully functionnal on 40 series. But no, you have to pay to get the full feature.
PC means options, nvidia chose the MAC treatment.
with slower cards, such as 20 or 30 series there will be a lot more of graphical glitches because of slow rendering of those generated frames and if consumers will see that mess - it will hurt technology, and sells of 40 gen, so, nvidia obviously removed it from previous gen cards.
@@PITERPENN yeahright, 4060 is so fast compared to 3080 isn't it?
Well, it's bad business to bring out technology that can make old cards better. Like imagine if DLSS keeps on getting better and better and you'd be able to use your 3080 for games in 10 years and still get 60 fps at like 1440p lol
Where you get this info?
@@kosmosyche taking a simple time in your life to actually look at the diagram of Ada lovelace doesn't hurt your brain to show the difference between Ada lovelace and Ampere and previous gens
That's clear now. Thx.
Just a word on super resolution: we can say that there are spatial super resolution AND temporal super resolution. Therefore DLSS is not wrong. With DLSS 3 you've got spatial and temporal super resolution knowing that temporal one is only accessible to RTX 40xx.
There is nothing that is specific to "AI" cores that you can't run on compute pipelines. This applies to all of DLSS. Nvidia is playing with people but AMD is still the one getting blamed for everything...
Exactly maybe the actual architecture of the chip makes it less efficient vs the 4000 series but it’s definitely not impossible…that said who would upgrade to a 1600$ card if their 3090ti could do the same and get 60+ fps on the newest games? That’s why they’re freaking out over AMDs announcement who can do it without AI cores 😂😂😂
Nvidia=🤡💩🤡
Yeah and you can run ray tracing on software. The point is that it would not be fast enough to have a benefit.
Dedicated hardware will always run better, more efficient and will end with better results than running on all purpose hardware. AMD is doing the latter simply because they have no choice, - That is the difference.
Yeah but it does free up CUDA cores to do whatever while the Tensor cores take it to do the last steps before it leaves the GPU to go to the video out chip. AMD has the entire process on GPU cores as they don't have 2 types of incompatible cores on their GPUs.
Obviously there is nothing specific to graphics either which means you should go ahead and run everything on your CPU right?
I'm glad I'm not the only person confused by that graph! lol
Love the videos, keep it up. This video wasn't particularly interesting to me since I already understood the differences between each dlss version. I think this video could be very useful to anyone confused by it though.
Thank you for the explanation. Been a fan of the channel since about February right after making my first build in January. I opted for a 3070 and Ryzen 7 5800X and built around that, coming up to about 1.2k. Just for the 40 series to come out a few months later. Overall still satisfied with the build. Just wish I did a bit more research first.
I have the same build. Just game at 1080p or 1440p with dlss.
@@maxcornejo20 dude I ended up selling my 3070 for 250. Bought a 4070 for 550. Paid 300 for dlss3 and 4 more gb vram lol
appreciate the info, it did clear things up for me :)
Kinda crazy you had to even make this video but I guess not everyone is a builder/ into computers that is a gamer. 🤷♂️🤷♂️. Glad ya made it for people that were confused, lol.
Yeah it is kind of odd that they're basically using DLSS as a brand name for all related technologies.
Also, if people find this stuff confusing, just wait until they hear about Intel CPUs.
An excellent explanation, Mr Owen
I was thinking about this,
Thankyou for making this. This is really a marketing moves by nvidia, now ppl knows
Daniel makes a good demonstration
Thank you this made this all make so much more sense!
Think of it like this: DLSS is a software, that evolves regardless (sort of) of HW. DLSSv1, DLSSv2, DLSSv3, DLSSv3.5.... in DLSSv3 Frame Generation was added, which can only be exploited with proper HW, in this case 4000 series, but it is still an iteration of the previous DLSS version only newer, doing the same things (improved) + the new features. In DLSSv3.5 Ray Reconstruction was added, which can again be exploited by HW that supports it, this time all RTX GPUs. Thats it.
I don't get what's difficult to understand here. Frame Generation is 40xx series only everything else is available on all 20/30/40 cards.
I like watching David float around the screen.
Nvidia and being customer friendly would also be shocking. The more people don't know the more power Nvidia have. The reason they didn't do it the way Daniel and most people would do it, is to keep the confusion going. They simply want people to fear to miss out. It is actually psykological manipulation they are trying to perform.
Solid video. Cleared up a ton.
*Yeah I was confused at first, I thought 3 meant frame generation at first. I wonder if there is any performance uplift for 20 series cards from DLSR 2 to 3 & 3.5 (No RT), would love to see some side by side comparisons.*
I would guess from 2 to 3 is major image quality improvements and 3 to 3.5 is minor ones.
@@halcyonacoustic7366nah, not really, the biggest difference was moving from 1.0 to v2, the difference between 2 and 3 is minor, but I'm sure there's more going on behind the scenes with the AI but on the surface they look pretty similar in terms of visual quality and performance, the biggest thing for DLSS 3 is that it has the frame gen feature which is clearly Nvidia's poster child for DLSS 3, explains why they've managed to confuse a lot of people getting them to think DLSS 3 on a whole only works with the RTX 40 series... meanwhile DLSS 3.5 only just adds a better de-noiser for Ray tracing
@@halcyonacoustic7366nope, DLSS2 is Super Sampling, aka image upscaler, DLSS3 is Frame Interpolation, nothing to do with image quality.
Stop thinking 3 is an evolution of 2.
3 is the lastest v2 + FrameGen
3.5 is the lastest v2 + latest v3 + RR
From my experience 2.5.4(Death Stranding Director's Cut version which fixed ghosting/trail) is actually even better(the SS part) than the 3.1.1(Spider-Man Resmater, it re-introduced the ghosting for some reason) version i tried, i don't know about later versions, i swapped the DLL versions from the original DS(2.3), DS Director's Cut(2.5) and Spider-Man(3.1), and the 2.5 version was the best.
DLSS doesn't update 1 -> 2 -> 3 -> 3.5, it's more like 1 -> 1.1 -> 1.2 -> 1.2.1 -> 2 -> 2.1 -> 2.1.1 -> 2.1.2 -> 2.2 -> 2.3 -> 3 -> 3.1 -> 3.5, it's incremental until they rebrand when adding a new tech.
In my naming example, the jump from DLSS DLL v2.3 to DLSS DLL v3, has nothing to do with Super Sampling, it's just adding FrameGen, the same with the jump from v3.1 to v3.5, they just added Ray Reconstruction, the section of the software that does Frame Gen and Super Sampling is the same as the version prior to it, the jumps that aren't incremental are about adding new tech, the incremental jumps are the ones making it look/run better.
DLSS 3 does mean frame gen. the only people who say it doesn't are pedantic freaks like you and daniel. nvidia themselves still refer to dlss super resolution as dlss 2.
Why is every company doing the USB version thing?
Thanks, was looking exactly for this
Fun Fact - FSR3 should work on RX580 and 1060. The RX590 is listed but that probably means in 1080p, 900p, say 1080p with 83% res scale. You would need to turn off upscalling - just the FG. This info is on AMD FSR3 page. You know where this is going - 600p on handheld should be ok for FSR3 - wouldn't look that great but on the smaller screen it doesn't matter that much.
>Fun Fact - FSR3 should work on RX580 and 1060.
FSR3 works only as in upscaling on these, FSR FG only works staring from RDNA1 and Turing
@@unclexbox85 It's other way around, on rx580 upscaling+fg will work too slow, use just fg and it'll work.
@@michahojwa8132The other commenter has it correct. Polaris isn't good enough at async compute (which is what AMDs frame gen relies on) performance. You'd likely end up with a performance regression due to the increased latency for each frame
@@2intheampm512 AMD says FG works with rx590. Most of difference from rx580 can be overclocked. So that's, why im sure it can work. Also, rx580 is comparable to ps4pro. Maybe the old guy gets the update
Subbed for the high quality work you do.
Hi Daniel ☺👋. Firstly, love your content 🥳✊. Secondly, I agree and feel your RTX Feature table is exactly how I thought Nvidia should’ve done it as for the average Joe, Nvidia chart explanation confuses all but the more tech pc enthusiast user😇👍. Job well done …. Now we have to Nvidia, to once again, admit they goofed and change their RTX Feature table for yours 😅😂🤣😅😂🤣
I talked about the naming confusion a lot when frame generation was first announced. I couldn't believe how many people were arguing with me that it was completely clear.
Also driving the point home...the file you're looking at is the frame generation DLL (that's what the G at the end means,) not the DLSS upscaling DLL.
it's the same DLL.
edit: wait, it isn't, does he have a game installed that has FrameGen but not DLSS?
Games doesn't usually have 2 DLSS DLL versions, they're all packaged into one.
@@MaxIronsThird DLSS is nvngx_dlss.dll, frame generation is nvngx_dlssg.dll. Every game that supports DLSS uspcaling will have the first DLL, and every game that supports frame gen will have the second. I'm not surprised that Daniel got them confused when there's just a single letter difference in the names.
FSR runs on "any" cards, been Nvidia, Intel or AMD, runs on apple silicon and even runs on "any" consoles. Tell me why developers should expend resources and time do work on DLSS tech?
Because fsr sucks so bad
The problem is people confusing DLSS feature set version with specific features. The slide from Nvidia is fine, it describes both the features included in each DLSS version, and which cards can use those features. This is purely user error
This was helpful. Thank you.
I really like the t-shirt. Very slick and stylish.
Thank you, I wanted to upgrade my old gpu to an rtx series one and now I know what everything means
Well, thank you for the video, I hope it helps some viewer.
In my case, I make that reasoning: all RTX GPU supports DLSS 3.5 - only RTX 4000 and up support all it features; below that, it's everything besides FG (DLSS 3.0 was confusing though).
Again, it's like Android/iOS: your smartphone may be compatible with the latest version, but not with all its features...
dlss so simple!
It is an nvidia strategy to force those who have 2000 and 3000 series cards to buy the 4000 series and the next 5000 to benefit from frame generation. It's just that they didn't think that maybe they won't choose nvidia anymore, I think that amd with the next series will recover this difference on raytracing.
I have rtx 3060ti and i dont see myself bying anything new for next couple of years . This card will play everything at atleast 60 fps , and i never cared for ultra settings and super high resolution im fine with 1920x1080 .
@@malyman316 Getting 60+~ fps with maximum settings (with RT toggled off if included) on 1080p is a guarantee on RTX 3060TI... Provided the game is not poorly optimized, of course.
Excuse me, what isn't obvious about naming a featureset and describing its backwards compatibility with the number 3.5???
Thank you! Much more clear!
Thank God you explained it I was lost
Labelling all of it DLSS, is like calling a kitchen or bathroom “the living room” just because in a particular house (composed of kitchen living room and bathroom), the living room came/appears first and is big.
Incredibly helpful explanation, thank you ! Nvidia obviously wants to keep people confused so they sell as much 4000 series cards as possible, they can't do it in another way because they're catastrophic 🤦♂ What a world we live in..
Seriously?? lol
@@focusmicro Wdym seriously
I was really hoping Nvidia had unlocked DLSS 3 for older RTX cards now...
The video title is misleadig. DLSS3 is not supported by RTX 2000 and 3000.
3.5 is supported partially.
But who cares if FSR 3 does frame generation as well, right?
An open standard will always win.
Thank you for explaining in clear understandable non-jibberish
You forgot about DLAA (unless I missed it). It works in the same tech as the super resolution but is used differently. In my opinion DLAA is the unsung hero of these features
DLAA is really just DLSS where the render frame size is the same as the display frame size. DLSS has always worked that way, it was just not recommended to be used outside the recommended source scale rez factors (1.5, 1.72, 2.0). AA is always a part of basic DLSS dataflow, scaling enabled or not.
@@normanblack4005 Exactly. DLAA is just TAA with hardware acceleration.
Could be worse, could be the people in charge of naming USB standards.
DLSS is a versioned "package" with features. New features come with each new version, and it's the individual features that have hardware dependencies, not the package version overall.
AMD is in the same stupid boat. They claimed you can get the RX 590 to run FSR3 without frame generation. But how is it FSR3 if it doesn't have frame generation? Isn't it just FSR2.2 then?
Nvidia is worse than AMD
Great explanation
Thank you, appreciate it.
Spot on. Really dumb naming approach. DLSS and frame generation should be disentangled. I don't think most people would be that bothered with ray reconstruction as a distinct feature. It's great but is very much an under the hood feature/uograde that only serves to confuse by using it mainstream marketing materials.
For me dlss 2 means upscaling. And it doesn’t matter which version. dlss 3 is frame gen. Dlss3.x running on rtx 2000 series is really a dlss2 version 3. It can get confusing for some ppl.
With all due respect, people who think of it like that are confusing themselves. Even if you run the latest version without frame generation or ray reconstruction enabled, you're still using DLSS 3.5, not v2. You're better off thinking of it in terms of features, as Daniel just showed, instead of in terms of version.
the problem is the mix togheter FG with DLSS 3 . there is no separate implentation . dlss3 woould perfectly work on 3000 series and below
I just hope FSR 3.0 working on 20 and 30 series Nvidia cards pushes Nvidia to also allow DLSS 3.0 on those cards and to keep their pride do their best to outperform FSR 3.0 on those cards.
DLSS framegen won't come to 20 and 30 mainly because the DLSS Frame Gen is based on the fact that optical flow accelerator, 20 and 30 do have OF but it's not fast enough to generate frames and if you try you will get even more latency than 40 series get, they might try to copy amds HYPR-X thing which enables FSR Framegen in any dx11/12 game and leave to 40 series only just like HYPR-X is only on RDNA 3
I like how everyone with an nvidia card is excited not because amd is making frame gen but because nvidia might unlock their own solution as a reaction honestly tho that's what I hope too
It's way too early to say this, we don't even know if FSR or the driver level FG from AMD even work well.
@@unclexbox85 Except someone hacked the driver and demonstrated that it does in fact, at the basic level, work on at least 30 series. It had wonky artefacts and the performance gain was only 80% of what 40 series sees. But don't worry. Nvidia will eat a bullet befor walking anything back. They have chosen not to enable features, that the silicon was perfectly capable of, on certain SKUs SO MANY times in the past and never ever budged. This is the reason why they locked the open source Linux driver out via some VBIOS crypto BS. They got pretty fed up with the open source folks just enabling everything the silicon could do.
I don't understand why didn't name it dlss 2.5. Still a little confusing but it's better to say all rtx gpus support dlss 2 and 2.5, and 4000 only supports dlss 3. That would have made it so much more clear.
Learn to live with it. DLSS 4.0 coming innnnnnnnn!!!
How much performance difference is there between DLSS2 and DLSS 3 on 20 and 30 series? Anyway to test that? It'll be hard to tell imo because I guess we don't know if it's just general perfornance improvements or newer DLSS updates. I'm not a super technical guy so would appreciate if someone could help me explain this better.
Move the RTX labels outside the table on the left and fill in the rest with ticks and crosses.
Just because the dll doesn't crash on a 20 or 30 card doesn't mean that all of the functionality is implemented or that the new code-paths are actually being used. It's a black box. Heck, maybe the .dll detects the old card and sends the work off to some common system dll that came with the driver. Am I missing information that truly shows that the rendered frames are going through updated algorithms?
Currently getting around 50-70 fps on almost maxed out settings on my RTX 3080 rig running at 1440p. How much of a boost can I expect without the frame generation tech?
Got it, so at the very least my 3060 will run DLSS better with every update, as in, better upscaling and better FPS with this exact same card, with that, I'll give the DLSS 3 naming a pass :) can't wait for DLSS 4.0
It would less confusing if they just made frame generation it’s own separate thing and the ray reconstruction should have been called either DLSS 2.5 or 3
"DLSS 3 is not exclusive to rtx 40-series gpus!" is my new "Michael was never in witness protection!"
Thanks for the clarity. I was wondering if this DLSS update just upgrades all supported games with Ray construction or does that only support new games announced. Would be very cool if this retroactively upgraded all DLSS with ray reconstruction
Ray reconstruction requires developer integration. Just like frame generation.
Why do RTX 2000 cards need Ray Reconstruction? Without things like FG or AMDs FMF they can't even hold 30 Fps on RT ultra in cp2077 for example.
Is it possible to use FSR 3 frame gen with DLSS 3.5 without the NVIDIA specific frame gen? For example running 20 series cards FSR3 + DLSS 3.5( The super sampling and ray reconstruction ) any info on that?
Thank you so much for clarifying this. In theory can DLSS generate more frames for the Leman, because running your game at a lower internal resolution gets you more frames out of it once it's scaled up?
its simple , DLSS version: x.x is supported by all rtx cards , DLSS version : x.x + Frame generation is only for rtx 40 series ( in this case x.x is ≥ 3 as frame gen was released with dlss 3 ),
so DLSS no metter what version is supported in all rtx cards , but you need an rtx 40 series card for that + frame generation feature.
note : we may see another features in the upcoming dlss versions but it may or may not be exclusive to some rtx serie of cards just like FG, so get used to it.
i thought chart was pretty easy to read
So to get this straight
Even If I have an Nvidia 2060, I should always download the latest DLSS and replace it in my game files to get better performance?
So the new DLSS 3.5.0 would be better than DLSS 2.5.1 even for my 2060?
i think a better name for "Frame generation" is Frame Tweening Animators use this to make animations smoother at lower animation fps especially in stick figure animations or simple ones
I hope the video helps some people understand this more clearly. I think you were already communicating well on the topic.
Honestly, while I think the branding could be clearer, I haven’t ever been confused by Nvidia’s presentation of the features, only frustrated by the way some people talk about them on social media. When Nvidia talked about frame generation, they used the words frame generation specifically, even in their early slides. It was only on social media that some people started using the inaccurate shorthand DLSS 3 to specifically refer to frame generation. To use a geometric example, if I was using vector drawing software and they added a new pentagon primitive drawing option so that I could quickly add pentagons, I would talk about that as “the pentagon drawing tool” not the “updated version X polygon tool” even if it were part of the new polygonal tools in that version. It’s not the category that the tools are grouped under that makes the difference - it’s the lack of specificity and precision in the way that people talk about them.
Been trying to explain this to some confused people as well, that DLSS 3 works with the older cards, it's just the frame gen feature that's restricted, turning off frame gen doesn't just send you back to DLSS 2
yeah actually it does mean EXACTLY THAT and daniel is dumb as fuck
It means along with the 3000 series, I will probably be skipping the 4000 series, as well as AMD’s current offerings.
I scored when I got my 2070s ftw3 before the pandemic and before the GPU embargo. I gamed my arse off through the pandemic and now I gliding right through 2023 in 1440p.
When you game 1440p on a 4K tv at a distance you can hardly tell the difference. I lounge back on my lazy boy and experiment all the time between the resolutions and come to the same conclusion on 99% of games which is it doesn’t matter a lot. I do 1440p performance dlss and it looks amazing. I’m sitting back about 10ft though. If it was a monitor and I was resting my nose on the screen that might be a different story. For now my 3070 gets 60fps on cyberpunk with ultra RT so screw it haha
@@mikehatten5738 right might get the 5070 maybe even skip that lol
@@TheWancer it’s the rumors are true and it’s just a continuation of Ada Lovelace architecture than yeah lol
As long as I can play demanding games like cyberspunk at max settings 1440p there’s no reason to give nvidia my hard earned dough 🖕🏽
Aint no way your gliding through 2023 with a 2070 at 1440
They should've named it DLGT - deep learning graphic technologies or something like that.
Absolutely!!
I downloaded the dlss mod for starfield and ive got significant performance improvements, especially since the nvidia driver update. And im on a laptop with a 2070 max-q. I cant wait for an official dlss update from bethesda for the game, should work better.
Confusion of the highest orda. I honestly think nVidia themselves are confused. Confucius say, "If you want to confuse your customers, confuse yourself first".
In my opinion, DLSS should have been called DLU (Deep Learning Upscaler). There's no super-sampling happening here, it's actually the exact opposite.
"Super resolution" is the tech that enables games to run at higher resolution on lower resolution displays by downscaling. "Super sampling" or "Super scaling" is the tech that enables low resolution to run at higher resolution quality by upscaling.
Super sampling is the opposite of what you're describing. It's running crazy internal resolutions and then downscaling them to fit native.
That's a good point; DLSS = Deep Learning SUPER SAMPLING (as a form of anti-aliasing)
DLSS 3 has nothing to do with super sampling, it shouldn't be called DLSS.
Let's be thankful they're not lying and hamstringing the drivers to prevent new features from working on old cards. Not going to complain.
Lets face it, Is designed to be confusing to make people think they need the newest cards for it to work...
Actually there's no confusion here. Everything is clear.
NVIDIA expects us to believe that the tiny 4060 die has more flow accelerators than the massive 3090ti die.
That's why the 4060 supports frame generation whilst the 3090ti does not.
The 3090ti... a card 2-3x as strong as the shitty 4060....
The reality is, frame generation is software locked to sell RTX 40 series cards.
That's the only reason it doesn't work on 20 and 30 series cards.
There are hacks for DLSS3 on rtx30 gpus that prove what you're saying isn't true.
Yes, that's how computer architectures work. New ones have capabilities that older ones don't.
The 3090 also doesn't have AV1 capability, yet it still has a really good media encoder.
@@Wobbothe3rdI haven't seen it. I haven't looked for it though so Ill have to take your word for it.
@@ryanspencer6778 The die is 160mm for the 4060. The die for the 3090 is over 600mm.
You understand the raw numbers here. Even with a newer node, and all the number fudging in the world, the 4060 doesn't have more of anything that the 3090.
Maybe that's why the performance of the 4060 is so terrible, they wasted most of the die on flow accelerators. I don't believe it though.
82 RT cores vs 24. Every single metric just makes the 4060 look pathetic in comparison. Which it is. Why would it magically have an tonne extra of one magical component?
Think about it. It doesn't make sense.
FG is 4xxx only. Rest works on every RTX GPU. How's this confusing? I understand what you're saying and where you're coming from and i agree that the naming scheme is a bit of a mess, however, perhaps ppl who actually get confused by this should get a brain instead of a GPU?
I’m still confused can the mobile 4060 run dlss 3.5