I'm curious, what exactly is DLSS aiming for? Is it to improve image quality or performance? How does it work in real-time? Can you explain it a bit technically? As a beginner, I'm a bit confused.
How does DLSS technology work?
👁️ 9 views💬 3 replies❤️ 0 likes
3 Replies
What's really interesting about DLSS is that it aims to improve both performance and quality at the same time. It doesn't just boost FPS; it also makes the rendered image sharper and more detailed. But what I'm curious about is how much of the GPU's resources these processes consume. For example, while the Tensor Cores in an RTX 3060 can handle all of DLSS's calculations, would an older GTX 1660 suffer a significant performance hit even if it could run DLSS? It's a detail NVIDIA doesn't really discuss in their official sources.
I'm also confused about how many different versions of DLSS there are. There's DLSS 2.0, 3.0, and the new DLSS G... Do they all work on the same principle, or are there differences between them? I'm particularly interested in how the tracking changes when switching to DLSS G. When applying DLSS to our games, which model do you trust more? Or do they each have their own advantages?
DLSS, just like FSR (FidelityFX Super Resolution), reduces the rendered resolution and then upscales it back to the original resolution with AI assistance to boost performance. However, DLSS leverages NVIDIA's dedicated Tensor Cores and supercomputer-trained datasets, delivering much sharper and cleaner results compared to FSR.
Essentially, it renders at a lower resolution (e.g., 1440p) while maintaining the aspect ratio for a 4K display, then uses DLSS’s AI model—trained on vast datasets—to "predict" and smooth out the frame, enhancing quality. So, it balances performance gains with better image quality.
DLSS actually aims for both, but I first noticed it saving performance. Last year, I was struggling to hit 60 FPS at 4K on Cyberpunk 2077 with ultra settings. I enabled DLSS 3, and it instantly jumped to 100 FPS. When I compared it frame by frame, I thought there was a loss in image quality, but then I learned that NVIDIA cleverly upscales the data from 1440p to 4K, and I was amazed.
The real-time part feels like magic. NVIDIA trains DLSS by analyzing millions of images on their supercomputer to figure out how to improve it. Then, in-game, the GPU renders at a lower resolution, but DLSS doesn’t just upscale that frame—it smartly adds pixels without adding latency. I saw the difference myself: when I compared four dark corridor shots in the same game, DLSS 2 Quality mode looked almost as sharp as native 4K.