Nvidia DSR Legacy scaling

What's the difference between Nvidia DSR Legacy Scaling and Nvidia DSR Native scaling? I asked Google and Chat GPT, but nether of them knew what the difference was. Do any of ya'll know?

DL scaling is the new version that uses Tensor cores if your video card has them (Turing, Ampere, Lovelace). DL stands for deep learning. The legacy version is kept for compatibility with older video cards with no Tensor cards. Visually I don't think there is any difference, still using a 13-tap gaussian filter. DL scaling (DLDSR) is just more efficient, costing less of a performance impact.

IMO, the legacy scaling looks better. And I haven't found any meaningful performance difference. I think the DL version probably hasn't been updated for a long time and has been allowed to stagnate and maybe even regress in effectiveness.

I've found the DL versions perform signifcantly faster, and they also add a bit of sharpening as well. Infact in some cases 2.25x DL can look better than 4x pure. The DL versions also have the advantage of not scaling the UI of old games to being insanely small.

DSR is one of the die hard features that make me stick to nvidia no matter what. I play older games more than modern ones and couldn't live without DSR.

Infact, just about to fire up the original Hitman 2: Silent Assassin with 2.25X DL on my 55" TV...

Use DLDSR. Quite a bit faster.