[MoreSEMI]

It is not possible to calibrate it away without knowing what the central wavelength of the LED is. That would require a spectrometer and if you manage to build one on chip per pixel which is currently not possible/practical.

I don’t understand what you mean with the GPU. It is has no information about the exact color of the LED.

[jmole]

This is equivalent to saying that you can't build a color-accurate display at all because you don't know the central wavelength. Not only is this inaccurate (LEDs are binned for exactly this purpose), but brightness variations are by far the greater contributor to display inaccuracy.

The shift in wavelength is primarily determined by temperature and current, and they work in opposite directions so sort of cancel each other out. And in any case, we're talking about well-characterized shifts on the order of a few nm over the operating range. The eye's cones are broadband, so you're not going to notice wavelength shifts, especially compared to the brightness variations over the same range.

This is a big deal for white LEDs because you have no control of the resulting color temperature (the phosphor emission and blue component wholly determine the output), but for an RGB structure, you have pixel-level control over each component.

[MoreSEMI]

I am sure your right about the eyes. But I would posit that most people wouldn’t know uncalibrated from calibrated anyhow. So seems like a moot point.

As to binning LEDs that works because it is constant. You can calibrate it once and done. But if you change the brightness by changing the current, it means your calibration is out of wack. Perhaps you can make a calibration at multiple current settings, but that seems inconvenient when using PWM will achieve the same thing.