DLSS 4 finally better than native?

[Stoly]

When DLSS was first announced it promised IQ every bit as good as native. Far from it, there were numerous issues like ghosting, smear and blurriness.
Then DLSS 2 came out and it was a great improvement delivering IQ that in some instances almost matched native. Still the same issues while reduced, persisted.

DLSS3 is just pretty much DLSS2+framegen so IQ was the same.

Finally, DLSS4 brings big IQ enhancements by using a transformer model vs neural model used prior. Take a look at this article from Techspot

People claimed that DLSS 2 Balanced mode was still superior to FSR Quality, now it seems DLSS4 Performance rivals DLSS 3 quality and DLSS4 quality is better than native.

Best thing is you can use DLSS upscaling with the RTX20 and up.

You may want to check it out

 

[LazyGamer]

I'd say that in some situations the answer will be yes - if only due to increased performance with no perceptible loss in image quality.

But that it will remain situation and end-user dependent.

 

[Peter_Brosdahl]

I've mostly enjoyed nearly every iteration of DLSS so far. The first version blew but it didn't take long for updates to improve upon it.

That being said I've mostly been against MFG (multiframegen) but it too has come a long ways and I have been tinkering with it for more recent games (Indiana Jones, CP2077, Hogwarts, Dragon's Age, and now SW Outlaws). I normally crank everything to max (sans post crap) and when possible like to use DLAA which further improves IQ. However, doing all this usually requires dropping to 1440p and sometimes 1080p with my 4090 but I have to give credit to the displays I have because things look incredible but MFG is often needed to keep FPS looking smooth.

As far as the transformer model goes, yep, I'm a fan. I still don't know if I'll be able to get, or will want, a RTX 5090 but thanks to DLSS both my 4090 and 4080 SUPER are doing great as is the mobile RTX 4080 (aka desktop 4070) in my laptop.

 

[Brent_Justice]

DLSS 4 Upscaling Transformer Model is definitely better than the CNN model, it's a nice improvement. I wouldn't say it's "perfect" though.

I lean towards DLAA for gaming not DLSS Upscaling for the 'best' image quality. However, if I HAD to use DLSS Upscaling to improve performance due to me wanting to have features like Path Tracing or Ray Tracing enabled, then I would most certainly prefer to use DLSS 4 Upscaling Transformer Model for Upscaling, it's the 'best' option in that regard.

People also need to be very careful when speaking about DLSS/4, it needs to be specified if you are talking about the Upscaling component of DLSS or the Frame Generation component. Both of those technologies do two very different things, with two very different goals and outputs in mind. I dislike it when people speak on DLSS/4 and don't mention if they are referencing the Upscaling component or Frame Gen component, but instead wrap everything up into just the DLSS name. It happens all over the place, and it is getting confusing and annoying. Obviously , NVIDIA's intent to confuse people on the branding and wrap up the two DIFFERENT technologies into one brand name, but we need to be VERY careful about specifying which part of DLSS is being discussed and compared and talked about in terms of image quality and performance. This is a big pet peeve of mine.