Huh, good question. I don't have a clear idea what would trigger the max vs min filter in this case; the pixel derivatives will be pretty crazy. Best bet is to look at a nice mirror ball (zero roughness) before and after and see if the reflection looks worse. If you can't tell a difference, then I guess it's fine.

Seems to be fine here...

https://mrdoob-sandbox.glitch.me/pmrem-halffloat.html
https://mrdoob-sandbox.glitch.me/pmrem-nearest-halffloat.html

This PR noticeably changed the visual result of webgl_pmrem_test:

https://threejs.org/examples/webgl_pmrem_test
https://raw.githack.com/mrdoob/three.js/dev/examples/webgl_pmrem_test.html

Is the new output considered as correct?

It seems other examples are also affected:

https://threejs.org/examples/webgl_loader_gltf
https://raw.githack.com/mrdoob/three.js/dev/examples/webgl_loader_gltf.html

@Mugen87 Both of those examples are positively identical on my macbook. Can you share screenshots and what GPU you're testing on?

Prod:

Dev:

Prod:

Dev:

I also see a difference with a Pixel 4a and latest Chrome.

I can reproduce 👍
The renders are a bit darker now. Working on a fix.

It's also worth noting that #23428 changed the cubeuv configuration so the renders will change sligthly from 137 to 138:

Before	After

As for the change; the pixel values themselves should all be identical, merely arranged differently. So any difference should be scrutinized. I'm still having trouble seeing the differences in those renders; can someone circle something?

I guess @Mugen87 and I have super eyes at this point 😁

https://www.diffchecker.com/image-diff/

As for the change; the pixel values themselves should all be identical, merely arranged differently.

There are less pixels now. The tiles are squashed vertically.
Which means that the resulting accumulation is unlikely to be 100% the same.

No, they only look squashed. The new texture is larger than the old. And not square.

Ahh, I see. I'll have to update the debug code with the new size then.