DLSS, DLSS 2 and DLSS 3 are the same thing, they are all the same basic Deep Learning Super Sampling. It doesn't generate frames, as such, it looks at a series of frames in the buffer, and interjects certain frames into the sequence, with or without lighting affects as needed. The only real difference between the 3 is the Algorithm used, it's capacity and decision making process.
Gen3 is advanced enough that it's like your brain at 20yrs old vs Gen1 or Gen2 which is your brain at 5yrs old and 10yrs old, which is why they aren't backwards compatible. At 5 you understand blue+red=purple, at 20 you understand blue+red = any shade of purple from violet to maroon to lilac to mauve depending on the levels and types of blue and red mixed.
Your FPS goes up because instead of the 2 actual frames being reproduced, the gpu is putting out 4 frames, but realistically it's only the same 2 frames, just output twice. This can introduce blurriness if there isn't time for the full frame to be interjected at the full detail levels and only the wire frame is doubled.
The difference between DLSS and DLSS 3 is that 3 is far better at choosing the frames to be reproduced and far better at choosing what parts of the frame can be reproduced without adding to the blurriness etc.