Something doesn't add up here. TPU found the 4090 to generate 40 fps at 4K with RT enabled in CP77. Now the 5090 can only do 30 fps. Hmm....

Yay? I guess?

To put it in real terms: 230 fps with 50 ms latency means that effectively, even if you were constantly using an input device to move, for roughly every 12 frames displayed that input has not actually been processed as part of the graphics frame.

Would be interesting to see how frame gen, when in a scenario where the graphics load is so high the real FPS would be say around 10fps, would handle cut-scene changes...

3 months ago, a 5000 series card was almost going to be an auto-buy upgrade for me.

The more I hear about all this frame generation AI bullshit, the less I like it. The 9070XT is sounding better and better, especially if the price is right.

(I've also thought DLSS on any game made it look a horrible fuzzy blurry washed out mess)

Nater3 months ago, a 5000 series card was almost going to be an auto-buy upgrade for me.

The more I hear about all this frame generation AI bullshit, the less I like it. The 9070XT is sounding better and better, especially if the price is right.

(I've also thought DLSS on any game made it look a horrible fuzzy blurry washed out mess)

Never auto-buy anything. Brand loyalty only makes you poorer and a few unethical, jackasses richer.

Nater3 months ago, a 5000 series card was almost going to be an auto-buy upgrade for me.

The more I hear about all this frame generation AI bullshit, the less I like it. The 9070XT is sounding better and better, especially if the price is right.

(I've also thought DLSS on any game made it look a horrible fuzzy blurry washed out mess)

Not to spoil that plan, but AMD will be bringing their own version of AI/Tensor/DLSS processed upscaling and are already pushing out some frame gen stuff of their own, as have Intel

VinceroNot to spoil that plan, but AMD will be bringing their own version of AI/Tensor/DLSS processed upscaling and are already pushing out some frame gen stuff of their own, as have Intel

That's why you ignore all that, wait for TPU pure rasterization tests and then decide which card to buy. We just have to hope that GPU manufacturers and game developers don't enforce mandatory upscaling and frame generation with no means to disable.

VinceroNot to spoil that plan, but AMD will be bringing their own version of AI/Tensor/DLSS processed upscaling and are already pushing out some frame gen stuff of their own, as have Intel

The HW accelerated AI specifically is what will (hopefully) make FSR no longer trash vs DLSS 3. OFC, DLSS 4 is out by then...

RTX 4090>RTX 5090 in native performance?

dgianstefaniThe HW accelerated AI specifically is what will make FSR no longer trash.

I wouldn't have said FSR was 'trash'... it worked better than using lower rendering at say 1080p or 720p on a 1440p monitor.
It's actually more annoying that the resolution display scaling on the GPU isn't as clever / good as FSR in that respect.

Anyway... let's see what they bring

As far as I heard the raw performance difference is only around 25% - 30% faster from what I was reading in the jump from the 4090 - 5090. Alot of the push this generation is better DLSS so not too surprising. The big question is going to be how DLSS performs across the board in Latency and image quality because most high end gamers are not going to use it if it introduces any latency that affects the game.

"RTX 5090 barely managed around 70 FPS in the game"

Uhm at 4K DLSS Quality, you're basically getting native picture quality that is running full path tracing at 70fps, without the added latency from frame gen. That's rather incredible.

VinceroI wouldn't have said FSR was 'trash'... it worked better than using lower rendering at say 1080p or 720p on a 1440p monitor.
It's actually more annoying that the resolution display scaling on the GPU isn't as clever / good as FSR in that respect.

Anyway... let's see what they bring

FSR is trash, to this day. In fact, on my handheld, if I have any other options like TAA or XeSS, I'll choose those. It's incredible how bad FSR is.

colossusrageblackUhm at 4K DLSS Quality, you're basically getting native picture quality that is running full path tracing at 70fps, without the added latency from frame gen. That's rather incredible.

Yeah but for 20 hundred plus whatever scalping tax it doesn't feel as amusing.

I'm more interested in a DLSS4 P VS DLSS3 Q comparison, both sans FG, in terms of image quality and stability. At a pedestrian resolution of 1440p preferrably. If the former isn't significantly worse than the latter we have really good news for everyone.

Ignoring the AI stuff, the pros and cons of it and the features that are 50 series only, I did my own analysis based on the core specs and came up with the results bellow, which are my expectations...

5070 against the 4070 Super
(My expectation is the 5070 will be approx 5% slower than the 4070 Super)

5070 Ti against the 4070 Ti Super
(My expectation is the 5070 Ti will be approx 6% quicker than the 4070 Ti Super)

5080 against the 4080 Super
(My expectation is the 5080 will be approx 10% quicker than the 4080 Super)

5090 against the 4090
(My expectation is the 5090 will be approx 27% quicker than the 4090)

"DLSS 4 with Multi-Frame Generation has allowed for obscene jumps in performance"

I just hate that claim/sentence, its not performance, its doubling frames from an actual rendered frame, its frame interpolation, its adobe flash tweening, its nothing new or special and its not....what I would call "performance"

Now I am not hating on it, im all for providing the user with a better experience and if a frame in between makes it feel smoother, go for it.
but stop hyping it up like that and stop pretending that frame generated fps is the same as "real" fps because it isnt, again, we could push 20 frames in there if we wanted, "wow look at that, 2000 fps, crazy"...yeah.....great.

ZoneDymo"DLSS 4 with Multi-Frame Generation has allowed for obscene jumps in performance"

I just hate that claim/sentence, its not performance, its doubling frames from an actual rendered frame, its frame interpolation, its adobe flash tweening, its nothing new or special and its not....what I would call "performance"

Now I am not hating on it, im all providing the user with a better experience and if a frame in between makes it feel smoother, go for it.
but stop hyping it up like that and stop pretending that frame generated fps is the same as "real" fps because it isnt, again, we could push 20 frames in there if we wanted, "wow look at that, 2000 fps, crazy"...yeah.....great.

I agree Tv's have been doing this for years, now they spin it as performance. Its like Nvidia is Gaslighting everyone about this technology.

dir_dI agree Tv's have been doing this for years, now they spin it as performance. Its like Nvidia is Gaslighting everyone about this technology.

Jensen walked on to stage with a shield... Turned out the media and crowd were too passive to make him need to use it...

It’s just a reminder of how much taxing the PathTracing is. In games that sport it only playable by 4090/5090 + upscaling unless it’s 1080p.
I find upscaling to be fine in quality mode for 1440/2160p final resolution.
Even FSR looks nice when game developers take some time to implement it right.
For example FSR3 sucks big time on CP2077 and FSR2.1 looks much better with minimal shimmering. Someone got lazy there and it’s not AMD.

Anyway…
RT is here to stay, no doubt.
We (I) can live with “simple” RT with or without upscaling. DLSS upscaler is fine already outside of 1080p native/final resolution and FSR is catching up. Even FSR3.1 quality mode looks great and very close to DLSS3.5 when comparing them in 4K as long as it’s done right.

dia6oloIgnoring the AI stuff, the pros and cons of it and the features that are 50 series only, I did my own analysis based on the core specs and came up with the results bellow, which are my expectations...

5070 against the 4070 Super
(My expectation is the 5070 will be approx 5% slower than the 4070 Super)

5070 Ti against the 4070 Ti Super
(My expectation is the 5070 Ti will be approx 6% quicker than the 4070 Ti Super)

5080 against the 4080 Super
(My expectation is the 5080 will be approx 10% quicker than the 4080 Super)

5090 against the 4090
(My expectation is the 5090 will be approx 27% quicker than the 4090)

Those estimates could be right but only on raster native. Entering RT and upscaling it’s a bit better for the latest. Leaving out the FG/MFG which is not quite there yet.
RayTracing outside of PathTracing can become the new standard really quickly with the latest hardware from all GPU manufacturers.

Shows again still that professional workloads is driving the gaming market. There is no need for raw performance in professional/prosumer space, but for AI/ML.
They just worked around it and found a way to “satisfy” gamers too with frame generation, or so they’re thinking.
Introducing PathTracing is pushing/forcing this notion further and human psychology is susceptible to the latest and greatest must have(s). They even starting to call path tracing as “full raytracing” so the “simplest” form of it feel like incomplete or broken.

Fascinating how some can manipulate the masses.

All I care is the 32GB frame buffer. This is sweet for Flux and stable diffusion.

remekraYay? I guess?

cinematic performance

Bee9All I care is the 32GB frame buffer. This is sweet for Flux and stable diffusion.

Another nVidia “smart move” to push people like you to the $2000-3000+ price point by keeping all other GPUs at bare minimum frame buffer of 16GB and below.
They could have made 5080 with 20 or 24GB for users like you even if we accept that games don’t need that frame buffer. IMO should have been at 20GB.
nVidia: Here we have hardware for everyone on various price points.
But no… you want this? Get your hand elbow deep in your pocket.

I mean if someone can’t accept that there are doing this on purpose then green is all you see.

It's nice to see an AI prosumer GPU with 32GB of vRAM running games.

Bee9All I care is the 32GB frame buffer. This is sweet for Flux and stable diffusion.

k why do you care about that?

Quick study of what the numbers are:

Latency (ms)	FPS	DLSS Super Res	DLSS Frame Gen	Ray & Pay Tracing
45	+260	Performance	4x	On
25	+400	Performance	4x	Off
28	+90	Performance	Off	On
52	+200	Quality	4x	On
46	+160	Quality	3x	On
43	+100	Quality	2x	On
35	+60	Quality	Off	On
70	+30	Off	Off	On

If those numbers are accurate, it seems like DLSS upscaling will be much more performant on Blackwell. And it's the new transformer model.

I'm actually looking forward to DLSS comparisons between similar 40 and 50 series cards.