Doesn't the Color Picker work in 16-bit mode?

I didn't think I implied that I had anything But an 8-bit display. Sorry if that is what came across. 

What I was hoping to do was use images that start as 16, or even 32 bit, and use a dithering method to convert them to 8bit so that banding is not perceptible. But in order to do that at the best quality, you need to start off with good quality data. I was surprised that Photoshop isn't keeping the data as expected. 

In after effects, if I create a gradient that goes from 0-32767 (oddly it won't let me do 0-32768 or it turns black, but that could be a graphics card limitation) then the values are 1 apart. If I switch to 32 bit, the values are .0001 apart, which is annoyingly a smaller value in terms of rounding back to an 8bit image, but still, the values don't jump.

In photoshop, the gradient fill layer adjustment jumps in rather large increments. At 32000 pixels wide, I didn't quite go the full width, it's jumping in values of 4-10 in 16 bit. 32bit only shows 3 decimals, so is only nominally better than 8bit in terms of how many colors it can show. The main advantage of 32bit is it can store values above white and the available layer adjustment are actually not destructive, whereas in AfterEffects, they aren't destructive even if they are in 8bit, which is probably one reason you don't get the same results sometimes.

I have been annoyed Several times at Adobe deciding to destroy data without telling the user. One good example is with PNG, and I think it's back, but haven't confirmed. PNG IS allowed to have RGBA information stored in every pixel. In Photoshop though, that information is there, up until you save it. If you have something like 100,100,100,0 then when you save it, and open it again, you'll be surprised with a new value of 0,0,0,0. AfterEffects doesn't do that though. If a channel has a value, it keeps in on export. 

This rounding, because most people don't need it, is kind of garbage. Put that kind of thing in preferences so that we can uncheck the box that says "use slightly less precise colors for improved performance." That way we could uncheck that box when we need to do something like work on a gradient for a depth map without creating banding. Just because you can't See it as is, doesn't mean you can't see it when the values get clamped later as part of a different process.

Just look into this one repo GitHub - jursonovicst/gradient: 10-BIT DISPLAY TEST SEQUENCE

These are test files for 10 bit displays... I could not check colors with Photoshop 2018 and I had to use gimp. That is not normal. Besides that, don't you think that if Photoshop cannot pick up 16 bit color, than you should write 15 bit (0-32768), not 16 bit (0-32768), right? Or it should use 0,5 to really have 16 bit color. So this is a mistake in Adobe code. A MISTAKE. It will not be a problem for performance. Nvidia geforce 2080 can work 48 (or 48 + 16 alpha) bits without any problems. So does Snapdragon 855 or amd chips...

True 16 bit would seem like a great opportunity now

What output device is capable of outputting full 16-bits?

No digital halftone screen can output more than 256 gray levels because it would take a massive amount of printer resolution to modulate the halftone dots to more gray levels, so that eliminates most commercial print.

Stochastic screening typically used with inkjet printing would have a similar problem where both the horizontal and vertical printer res would have to be high enough to increment the dot spacing to 65K levels. You might argue that splitting the cyan, magenta, and black channels into light and dark value inks can gain some gray levels, but it's still nothing close to 15- or 16-bit.

From a user perspective, I totally agree with that. Can you imagine typing 27356 into the Curves input field, and correcting 27405 to match? It's insane, considering that you probably can't see the difference between 249 and 250 unless you measure them.

I think one frustration with the Color Picker is that while the RGB fields are 8-bit values, the C,M,Y,K,—S,B, & L are not. So if you are trying to spec a CMYK color in .5 percentages (or get its RGB equivalent) it gets difficult. A typical halftone screen can't be more than 8-bit, but it could define more than 100 gray levels and you might detect that difference in a highlight color.

If I try to enter something other than an integer in a CMYK field I get this warning

My 10.4% gets reset to 10, but the other space conversions change, so obviously the color changed even though I'm still seeing 10%C.

It would be better (at least for offset print work) if the CMYK percentage rounding wasn't to the nearest integer but to the nearest number.

InDesign does conversion rounding to 100ths

Suicide  wrote I often use to Color Picker to do two things at once: 1. See the color of the pixel I've picked 2. Compare the colors of different pixels and see if they're actually different.

16 bit depth isn't for appearance. It's for data integrity through repeated editing operations.

Visually separating 8-bit values is difficult enough, and in many cases you can't. Don't forget that 8 bit depth is all you get out of your monitor (at best 10 bit if you have a display system capable of that).

Try this one to give you a little perspective:

Free Online Color Challenge and Hue Test; X-Rite

I managed a perfect score on my second try, a fact I take some pride in because only a few % manage that. The first attempt gave two errors.