Yes, I feel there is enough evidence now that ports are certainly possible if there is the financial incentive to go through the process. Way back when the discussion was in its early days, I fully believe if I had argued that Doom, Wolfenstein 2, Hellblade and the Witcher 3 would come to the Switch, you would have pushed back on that statement. Disagree?
Yes, I 100% disagree.
What mattered for the port discussion were the scarcity of AAA 3rd party franchises that move tens of millions across a generation, and those were effectively scarce.
The huge gap in performance resulted in the Switch getting no Call of Duty, no Battlefield, Battlefront, no recent Final Fantasy, no Overwatch, no Far Cry, no recent Assassin's Creed, no Destiny, no The Division, no GTA, no Red Dead Redemption, etc.
The list goes on and on.
AAA 3rd party games are super scarce on the Switch,
despite the financial incentive to port games to it. Devs need to take away too much of their 9th gen games to make it work on the Switch. Fast paced idTech 5 games need to run at 30 FPS with very low resolution and Witcher 3 needs to run at sub-Vita resolution with what looks like gouraud shading at places to be barely playable.
And now we're 1 year away from another generation of consoles that will again be a giant leap from the 2013 models, and the gap between the switch and these new consoles will be so big you won't get anything but the eternally cyclical 3rd party sports games on it.
Found a good article that breaks down FP16 into real world utilization and the challenges that implementing it can bring on. Overall, a roughly 10% boost in performance seems rather modest compared to the doubling of the peak floating point performance used in marketing documentation, but that is still a nice perk just the same. I would think that Nvidia's development tools have utilized half precision shader support with similar results. Neat, just not as significant as I would have hoped.
Some factors to take into account:
1 - On the PC, devs use FP16 only in cases where quality isn't affected when compared to using FP32. This is obviously not a limiting aspect for the Switch, as devs should be a lot more willing to sacrifice quality to make a game playable. FP16 pixel shaders are a lot more prevalent on mobile games and the Switch should be no different.
2 - Vega 64 is known to have an excess of shader throughput, so using that GPU to evaluate FP16 gains isn't ideal. The best GPU to do that comparison should be a smaller one like the embedded Vega 8/11 vs a similarly sized one without RPM like the nvidia MX150 or MX130.
So does it make a "big" difference? Probably yes, a lot more than the 10% you saw for the Vega 64. But the Switch needed a GPU some 3x faster to make a difference in the ports, and FP16 certainly doesn't provide that.