Posting this from the current discussion at Computex. Could anyone ELI5 how does DLSS 4 works? I still cant get is over my head how you could all of a sudden have high quality 4K Path tracing and 300fps.
Festro · 7h ago
Normally a GPU creates images on your monitor generated by the video game. If you move in game, your character moves on your screen. There's a near imperceptible amount of delay inbetween your action to move your mouse or hit keys on your keyboard and the creation of that image. The number of images your GPU can make per second is constrained by how powerful it is and how much imagery the game is asking it to create. Reflections, shadows, high detailed textures, all reduce the number of images being created per second.
Typically we want to see at least 30 images per second to feel like motion is happening. 60 if we want it to be 'smooth'. 100+ if we really don't want to see any stuttering.
With DLSS 4 the GPU cheats a bit. Instead of generating the images from the game and your inputs it takes an image, and the image after it, and uses a special process to blend the 2 images into a series of images that visually connect the two. Think of a ball being kicked. In image 1 the ball is on the left of the screen, in image 2 it's on the right. DLSS would add images showing the ball going from 1 to 2.
Instead of needing lots of power to create the image from the game, it's just comparing two images and guessing what should be in any images inbetween. Suddenly you can go from 60 images per second to 300! Because you're adding 4 extra images inbetween the others, all generated by a quicker process than generating them normally.
What you lose, is a bit of power to generate images normally, and you gain an extra unwanted delay in getting the images to your monitor.
And not in ELI5 language:
So you'll go from 60FPS at a 35ms latency to 300FPS at 45ms by adding 4 extra frames from DLSS 4 to every 1 frame generated by the game engine. It'll be very smooth, but with a little lag, and the chance of seeing visual artifacts that the game didn't generate.
Some people are happy with the tradeoff. Others are not.
Typically we want to see at least 30 images per second to feel like motion is happening. 60 if we want it to be 'smooth'. 100+ if we really don't want to see any stuttering.
With DLSS 4 the GPU cheats a bit. Instead of generating the images from the game and your inputs it takes an image, and the image after it, and uses a special process to blend the 2 images into a series of images that visually connect the two. Think of a ball being kicked. In image 1 the ball is on the left of the screen, in image 2 it's on the right. DLSS would add images showing the ball going from 1 to 2.
Instead of needing lots of power to create the image from the game, it's just comparing two images and guessing what should be in any images inbetween. Suddenly you can go from 60 images per second to 300! Because you're adding 4 extra images inbetween the others, all generated by a quicker process than generating them normally.
What you lose, is a bit of power to generate images normally, and you gain an extra unwanted delay in getting the images to your monitor.
And not in ELI5 language:
So you'll go from 60FPS at a 35ms latency to 300FPS at 45ms by adding 4 extra frames from DLSS 4 to every 1 frame generated by the game engine. It'll be very smooth, but with a little lag, and the chance of seeing visual artifacts that the game didn't generate.
Some people are happy with the tradeoff. Others are not.