Originally Posted by WannaBeOCer
DLSS 2.0 definitely isn't blurry, there are some scenes where native looks slightly better and there are some scenes where DLSS provides more detail. This technology will be on consoles and AMD has been working on a similar neural network using Microsoft's DirectML since the launch of the Radeon VII. I'm just curious what AMD will call it, maybe MLSS?
When I was playing Deliver Us the Moon I noticed DLSS provided more detail than native. For example, text on a monitor was legible while native the text wasn't.
There is a bit of misunderstanding here.
DirectML is an API that is suppose to allow you to tap into machine learning capability of the GPU through directx 12, basically on the fly.
DLSS is not doing machine learning on the fly. It is a library through the drivers that has been "pre-learned" at nvidia labs and gets a certain algorithm that tells the GPU how to render the game and the textures it gets. Those algorithms and library is using the tensor cores to just do the calculations on the frames, not to actually do machine learning on the fly.
So using directML will not actually do the same thing, unless AMD also open a lab to calculate frames like AMD. They might just use directML to tap into the calculations ability of their GPU to calculate rendering through pre-known algorithms, though since AMD hardware doesn't have specific cores for compute, but they are using the same type of cores, it might hurt overall performance to shift some cores into compute over rendering.
But we will have to wait and see.