Been testing the DLSS Transformer vs. CNN in Cyberpunk on a mid-range mobile 3070 140W gaming laptop I have, and very impressed with the ViT model when it comes to temporal stability and how clean the image looks on the 16 inch display. Basically 1600p performance mode with the transformer model + ray-reconstruction at ray tracing medium settings looks as good as, if not better, than 1600p Quality mode with the CNN model, in my opinion.
There are minor artifacts, but ray-reconstruction cleaned up most of the noticeable ones (originally tried it without and had a quite a bit of noise/shimmering.) I am able to get a solid 60fps with these settings, so the minor (few fps) performance hit using the transformer model isn't noticeable. Before I wouldn't bother doing any ray-tracing on this laptop and would just do a mix of high/ultra at 1600p DLSS quality.