To celebrate the launch of NVIDIA DLSS 3.5 in Cyberpunk 2077: Phantom Liberty, Digital Foundry hosted a roundtable video chat with Bryan Catanzaro (Vice President of Applied Deep Learning Research at NVIDIA) and Jakub Knapik (Vice President of Art and Global Art Director at CD PROJEKT RED), among other guests.
When discussing the new technology, NVIDIA’s Bryan Catanzaro stated that not only is DLSS 3.5 more beautiful than native rendering, but in a way, its frames are more real when coupled with path tracing than native rendering with the traditional rasterized approach.
You know, actually, I think DLSS 3.5 makes Cyberpunk 2077 even more beautiful than native rendering. That’s my belief. The reason for that is, again, because the AI is able to make smarter decisions about how to render the scene than what we knew how to do without AI. I think that’s going to continue to develop.
Cyberpunk 2077 frames using DLSS (including Frame Generation) are much ‘real-er’ than traditional graphics frames. If you think about all of the graphics tricks, all the occlusions and shadows and fake reflections, screen-space effects…Raster in general is just a bag of fakeness, right? So, we get to throw that out and start doing path tracing and might actually get real shadows and real reflections.
The only way we can do that is by synthesizing a lot of pixels with AI. It’d be far too computationally intensive to do [path tracing] rendering without tricks. So, we’re changing what kind of tricks we’re using and I think, at the end of the day, we’re getting more real pixels with DLSS 3.5 than without.
CDPR’s Jakub Knapik agreed with him, calling rasterization ‘a bunch of hacks’ stacked on top of each other.
It is weird to say, but I agree with you. It’s a very interesting perspective where you’re saying, what’s the tradeoff here? On one hand, you have a rasterizing approach which is, as Bryan said, a bunch of hacks. You have a silo of rendering layers that have no balancing between each other, you’re just stacking them on top of each other to generate frames. Every single layer is a hack, an approximation of reality: screen space reflections, all that stuff, and you create pixels like that as opposed to having a way more accurate definition of reality with path tracing, where you generate something in the middle [DLSS].
In the past, you could say that you were trading off some quality for performance by using DLSS, but with DLSS 3.5, it’s really undisputably a better looking image than without.
The discussion on ‘fake frames’ first emerged with DLSS Frame Generation since it generates one frame independently of the rendering pipeline and interposes the ‘simulated frame’ after every ‘real frame’. Still, the performance boost enabled by the generated frames is indisputably big, especially in CPU-bound games where the regular upscaler can only do so much. It’s no coincidence that AMD is following suit with FSR 3, due to debut soon in Forspoken and Immortals of Aveum.
As for DLSS 3.5 specifically, the new Ray Reconstruction feature provides big quality benefits when upscaling and ray tracing are both active. You can read about that in my Cyberpunk 2077: Phantom Liberty test. For the full game review, head to this article.
