Discussion Nvidia Blackwell in Q1-2025

[CastleBravo]

Not only that, there are plenty of flaws visible in the B-roll they are intentionally not talking about. Look at the lower right part of the screen (the tree especially) in this shot for instance (timestamped to 4:55):

Ironically in the section they claim "ghosting and tearing" is fixed (yet you can still see ghosting on the character's hand, just improved considerably).

But all in all, both technologies certainly seem to have some merit. The transformer based upscaling does look better, despite its faults (and I have seen Ray Reconstruction in CP2077 at 1440p DLSS Quality and didn't like the blurry painting like resolve at all, thing they are finally showing here) and the 3x - 4x "fake frames" can also be quite useful for 240-300+ Hz monitors.

Not nearly as groundbreaking as Nvidia is hyping IMO, but definitely useful

Don't worry, I'm sure Digital Foundry will detail all the DLSS4 flaws in the DLSS5 showcase.

 

[eek2121]

It might be a good time to start talking about IPC decreases. If 5090 was doing well in raw perf (raster/RT) then they would not have focused so much on new fake frames. 4090 got away with it because it was actually good double normal perf and x4 if you really want fake frames, so that was ok.

33% more SMs and a wider memory bus on 21% more transistors. I think an IPC decrease is possible.

They used Frame generation to make the gains seem larger than they are.

I am willing to bet the 5090 is around 30-40% faster than the 4090. Possibly even more in RT or AI.

You can’t necessarily compare raw transistor counts because the chip design itself has several changes to accommodate for the new memory bus and tech.

Also, don’t knock FG, it is actually pretty solid tech. It helps keep games synced with the monitor’s max refresh. Did NVIDIA do bad by using it in slides? Absolutely. Like for like comparisons are what they should have done. FG itself isn’t bad though…and the fact they crammed it into a 2 slot card is incredible.

If you honestly think these cards are only going to be 5-10% faster than previous gen, I don’t know what to tell you except wait for the reviews.

 

[Heartbreaker]

Well, they seem to be very critical of DLSS3 in this video. I wonder why they didn't show all these flaws with DLSS3 when RTX4k was launched? Funny how that works.

The first old video I checked you can find the reporting these kinds of ghosting artifacts present in the newer version of DLSS 3.5 vs the older one. So it really isn't true to say they only report artifacts in old version after new one comes out.

DF is generally quite critical once you get away from these exclusive first looks.

Here, Alex is almost gleeful in reporting the artifacts in the Newer version. Note it's exactly the same kind of character blurring going on that people claim he didn't report until DLSS 4 came out.

 

[Timmah!]

When are the reviews scheduled for? The end of embargo?

 

[igor_kavinski]

When are the reviews scheduled for? The end of embargo?

Jan 20 or 21st.

AMD better get the 9070 XT in gamers' hands by 15th if they want some sales.

 

[adroc_thurston]

AMD better get the 9070 XT in gamers' hands by 15th if they want some sales.

They wouldn't can N4c if they wanted any sales.

 

[Saylick]

71 fps or 246 fps. MS still same, just bigger number for marketing

X8 multi frames?

I come from the future with an exclusive look at the RTX 8050 12 GB and how it runs CP2077 path tracing:


Founder's Edition MSRP = $1200

 

[techjunkie123]

I come from the future with an exclusive look at the RTX 8050 12 GB and how it runs CP2077 path tracing:
View attachment 114427

Founder's Edition MSRP = $1200

You forgot to make the takeout box more greasy on the outside.

The progression is clear from off to DLSS 4 😅.

 

[eek2121]

Linus has a hands on video, in case you missed it.

 

[Meteor Late]

Actually, according to DF, latency is worse with x3 and x4 than with x2, I though somehow it would be similar to x2 FG.
That is very disappointing, I just don't like FG at all, the star of the show is still Super Resolution to me if it does indeed feature improved image quality consistently across a good range of games.

 

[Meteor Late]

It's also worth noting that the image quality comparisons are with DLSS Performance, so any ghosting artifacts are much amplified compared to DLSS quality.
So it is not 100% accurate to say reviewers are now talking about image issues that they didn't talk about years ago, it's more that reviewers tend to analyze quality using the quality mode, performance of course is much more prone to issues.
Another possibility is that this new transformer model improves image quality decently compared to cnn only when upscaling from much lower resolutions, but doesn't do all that much when upscaling from higher resolutions.

 

[CakeMonster]

With DLSS2 upscaling quality and speed increasing on all hardware, I could see people with enough VRAM reconsider if they really need to upgrade this time. Just dial down to Balance or Performance and maybe you can save the money, or put it toward addressing CPU limitations...

 

[Ranulf]

Speaking of this, this meme is quite well made:

Heh, good points from the comments. Edit to add, PCMR over on reddit has some good memes today.

 

[Tup3x]

I'm actually interested in DLAA (and somewhat DLSS super resolution) improvements. If they can really improve DLAA in motion even further, I'd say that AA has been pretty much solved. It has already been possible to force resolution scale/model (using Nvidia Inspector) but it's nice that they finally add it officially in the UI.

 

[gdansk]

I'm actually interested in DLAA (and somewhat DLSS super resolution) improvements. If they can really improve DLAA in motion even further, I'd say that AA has been pretty much solved. It has already been possible to force resolution scale/model (using Nvidia Inspector) but it's nice that they finally add it officially in the UI.

I find it it to be the best of the fancy Nvidia features that I have tried. I wonder why Nvidia marketing doesn't focus on that instead of fake frame spam.

 

[adroc_thurston]

I wonder why Nvidia marketing doesn't focus on that instead of fake frame spam.

IQ bumps are not something people care about, while more FPS is more FPS.

 

[Heartbreaker]

I find it it to be the best of the fancy Nvidia features that I have tried. I wonder why Nvidia marketing doesn't focus on that instead of fake frame spam.

Scaling only works once, fake frames keep giving and giving.

Fake frames are easy mode for higher FPS. How else could you claim 5070 performs like 4090.

 

[CakeMonster]

I find it it to be the best of the fancy Nvidia features that I have tried. I wonder why Nvidia marketing doesn't focus on that instead of fake frame spam.

They work on the older cards, so little value in putting too much emphasis on them compared to the new stuff. People have strong feelings about this, but at least I prefer upscaling to FG, and I believe that's the majority opinion as well. But to sell to the masses, you have to focus on the new stuff exclusive to 5000 series, not the stuff that will improve your current experience with an older card and make you rethink your upgrade urges...

 

[Meteor Late]

Doesn't this new transformer model requires way more calculations like they said? doesn't this mean performance could be much lower on older cards like RTX 2000 and 3000? it's great that they have improved Super Resolution quality, but is that at the same performance? because I've not heard anything about performance with this new model, or if it's the same.

 

[Heartbreaker]

They work on the older cards, so little value in putting too much emphasis on them compared to the new stuff. People have strong feelings about this, but at least I prefer upscaling to FG, and I believe that's the majority opinion as well. But to sell to the masses, you have to focus on the new stuff exclusive to 5000 series, not the stuff that will improve your current experience with an older card and make you rethink your upgrade urges...

I assume there's an implied silent "Gullible" between "the" and "masses".

FG is of limited utility to start with, and MFG will be of even more limited utility, you have to quite gullible to upgrade for more fake frames.

 

[Jan Olšan]

Fake frame generation are the new E-cores..

That's harsh to E-Cores, those do real work and are under-appreciated IMHO.

 

[SiliconFly]

This is new..

“When we asked [Jensen Huang] how DLSS 4 multi frame generation works and whether it was still interpolating, Jensen boldly proclaimed that DLSS 4 "predicts the future" rather than "interpolating the past." That drastically changes how it works, what it requires in terms of hardware capabilities, and what we can expect in terms of latency.”

 

[gaav87]

Rx 9070xt stealth attack vs rtx 4090 (5% slower)

 

[Fallen Kell]

It seems believable. A real shame if true, though. I still have hope that membw will do something in real games.

Why would you think memory bandwidth would affect games so much? The speed at which textured can be read into the graphics card doesn't do much within a game unless the textures being displayed in the game are changing massively from frame to frame (which is rarely the case). Unless the memory bandwidth gets so high, and that the time it takes to load from main storage drops so low that textures can be read in on the fly within a single frame generation, increasing memory bandwidth will have little performance benefit to a game's graphics.

The size of the overall graphics memory will still have the biggest impact on games, not just from a performance standpoint, but also from it's quality of output (if the game is compiled in such a way that higher resolution textures and 3d models/maps can be swapped into the interface, such as games like Skyrim). The card needs to access those textures and store them in some format where they can be easily read from to apply them onto the 3D surface maps of the game objects. The textures being used in a frame need to all fit within that memory, or else be able to read from some kind of location within the time that a single frame of output is being generated (so for someone who wants 144hz or similar framerates, the whole operation of reading all the textures used within the frame need to be well less than 7ms as the textures still need to be manipulated (i.e. transformed to match relative position of the texture on the object/surface vs the orientation of the viewing "camera", and all the other operations such as lighting and reflections that require those textures to be required before those calculations can be computed still need to occur afterwards within the 7ms timeline).

 

[Saylick]

This is new..

“When we asked [Jensen Huang] how DLSS 4 multi frame generation works and whether it was still interpolating, Jensen boldly proclaimed that DLSS 4 "predicts the future" rather than "interpolating the past." That drastically changes how it works, what it requires in terms of hardware capabilities, and what we can expect in terms of latency.”

And this is why you cannot take whatever Jensen says at face value. Just because he's the CEO of the most valuable company in the world doesn't automatically mean whatever comes out of his mouth is automatically correct.