Since it is stated to run on the tensor cores of the GPU, it is limited to RTX 20 series and upwards.
Generally speaking, the tensor cores are not used for anything else besides doing tensor operations and are dedicated hardware to speed up training and evaluating of neural networks. So they are not used in unreals rendering pipeline. [HR][/HR]
MSAA is used to remove aliasing of an image, so for example UE would render the image at 1080p, and MSAA would only remove aliasing, but again output 1080p.
DLSS is super resolution. That means UE still renders at 1080p, but the image you get as result after the neural network will be higher resolution (4k) and without aliasing since the reference images to train the network were at 16k.
Therefore it seems to be better to compare performance between 4k with MSAA and 1080p with DLSS, since both will then output 4k. Of course there is also the quality of the image you get when using DLSS compared to directly render 4k with MSAA…