DLSS4 Transformer Model vs CNN Performance in Cyberpunk 2077 – RTX 4090



Read more about Cyberpunk 2077➜ https://cyberpunk2077.mgn.tv

In this video we’ll be testing the updated DLSS4 in Cyberpunk 2077 with the RTX 4090 and 7800X3D. We will be testing the performance differences between the old CNN model compared to the new Transformer Model.

I’m really not that good with making image quality comparison tests so we’ll only focus on performance differences here. We’ll be testing at 4K with DLSS Quality, Balanced, and Performance using both Convolution Neural Network and Transformer models of DLSS, as well as with frame generation enabled.

We will then also repeat the tests at 1440p.

System Specs:

Ryzen 7 7800X3D
32GB DDR5 6000 MT/s CL30 with tuned sub-timings
MSI Liquid X GeForce RTX 4090
Gigabyte Eagle B650 Motherboard
2TB Gen4 NVMe Storage Drive
Corsair HX850I power supply
Windows 11 23H2
ReBAR enabled
Memory Integrity disabled

Timestamps:

0:00 – Cyberpunk gets DLSS 4
0:39 – 4K High, DLSS Quality, CNN
1:06 – 4K High, DLSS Quality, Transformer Model
1:34 – 4K High, DLSS Balanced, CNN
1:48 – 4K High, DLSS Balanced, Transformer Model
2:07 – 4K High, DLSS Performance, CNN
2:28 – 4K High, DLSS Performance, Transformer Model
3:18 – 4K High, DLSS Performance, CNN, FG On
3:33 – 4K High, DLSS Performance, Transformer Model, FG On
4:27 – 1440p High, DLSS Quality, CNN
4:41 – 1440p High, DLSS Quality, Transformer Model
4:51 – 1440p High, DLSS Balanced, CNN
5:05 – 1440p High, DLSS Balanced, Transformer Model
5:20 – 1440p High, DLSS Performance, CNN
5:30 – 1440p High, DLSS Performance, Transformer Model
5:50 – 1440p High, DLSS Performance, CNN, FG On
6:00 – 1440p High, DLSS Performance, Transformer Model, FG On

Join this channel to support a creator:
https://www.youtube.com/channel/UCgjV_NEAcVqZODDRZ0iyqCA/join

source

23 thoughts on “DLSS4 Transformer Model vs CNN Performance in Cyberpunk 2077 – RTX 4090”

  1. This is just a short video showing the differences in performance between the new Transformer Model and the old Convolution Neural Network DLSS modes in Cyberpunk 2077. As I mentioned I'll leave the image quality comparisons to the professionals as I'm really not good at making those kinds of videos.

    Reply
  2. Same on 4080 Performance slightly faster with CNN, which is disappointing. It looks better.

    But where was the 10% performance increase meant to come from? DLSS? FG? I'm not seeing it, probably just marking BS.

    Reply
  3. I've noticed it's blurry in RDR2, and the performance penalty gets worse the further the generation you go. It's not much, but it's still something when every FPS counts.

    Reply
  4. with path tracing + ray reconstruction, the transformer model is a lot better than the old CNN one. before, especially on lower DLSS modes, the game had this kind of "painted" look to it – edge of objects moved around a lot and were not very defined. all of it gets fixed with transformer models. there is a performance penalty for me at least but it's worth it

    Reply
  5. The Transformer model in Quality setting really impressed me. 1440p completely indistinguishable from native, even during movement and on distant detail. I hope people can find a way to rip the DLL so we can use it in other games that won't get patched.

    Reply
  6. Why aren't you enabling ray reconstruction? This technology was developed to solve the blur and ghosting issues that occur when ray reconstruction is enabled. It definitely needed to be tested in your video. Without ray reconstruction, it's very difficult to understand the difference between DLSS 3 and 4.

    Reply
  7. Good stuff, thanks for the video. I have to assume that x2 FG will also get better/smoother for 4xxx users as drivers mature for nvidia's push for the 5xxx series. The new upscale model being supported on older cards is a huge W.

    Reply
  8. So now its twice as good as native cuz according to nvidia fanbots dlss3 was already better than native? 😨😨😨🤣

    720p = better 4k aka 5070=4090 aka the more you buy the more you save era has begun 😂😂😂.

    Reply

Leave a Comment