Yeah, this is expected.
Path tracers have two ways to sample lighting - they can trace a bounce ray and randomly hit the light source, or they can deliberately sample the light source.…
Can't reproduce this in 3.4, 3.6 or 4.1, so I'll assume that it's fixed for now. If someone can still reproduce this, feel free to reopen.
Can't reproduce this in 3.4, 3.6 or 4.1, so I'll assume that it's fixed for now (3.x introduced many changes to how Persistent Data works, so it would make sense). If someone can still reproduce…
Can't reproduce anymore, probably fixed between 3.4 and 3.6 (my guess would be 39c30f6983009847d010856bee8a193a8ee3e73c, which would make this a duplicate of #89037).
Can't reproduce anymore, appears to have been fixed somewhere between 3.4 and 3.6.
Can't reproduce with OpenImageDenoise, so I'll assume this was specific to the old denoiser.
Depth appears to be fixed in recent versions. Mist is still noisy, but that's inherent to how it's computed - with enough samples and/or denoising, it converges to the expected result.
Since…
Can't reproduce anymore. Based on the stacktrace, this appears to be a duplicate of #77106, which was fixed in 389d5504120fb1a331a13118979a896b7e31ee0f.
The attached file doesn't work in 4.x anymore, but I'm pretty sure that this is just a sampling artifact. In voxels that contain some volume density, Cycles has to perform volume sampling, and…