Originally posted by Des I have a different view about this Mike - which might be a different tolerance for noise. With the K-3, I tried to keep to 3200 or less. I got some low light shots with the K-3 at 6400 ISO that have turned out alright with lots of treatment with DxO DeepPrime AI noise reduction (which is one of the best around), like this:
But if you look closely there is still plenty of noise visible.
Understood, Des. Personal appetite and tolerance for noise is a very individual thing. With an ISO 6400 image from the K-3, I don't expect or need it to hold up to
close scrutiny... No matter how much work I do, I know it's going to look rough at 1:1 reproduction - but I can get pleasing results for images scaled to fill my 17" HD laptop display at normal viewing distance, my 24" QHD monitor when viewed at the same distance one might look at a wall-mounted photo of similar dimensions, or my 42" TV viewed from across the living room... and that's my own personal yardstick for acceptable quality.
Looking at your excellent koala photo, it does indeed seem rather noisy and a bit washed-out at maximum reproduction size - but, respectfully, I wonder if you might be able to squeeze even more out of it. On close examination, I'd say the demosaicing algorithm is contributing to the noisy appearance. I've no experience with DxO software, but the visible maze-style artefacts suggest it may be applying an AMaZE algorithm or something similar. I use plain old RawTherapee most of the time, and for low ISO images I would apply AMaZE demosaicing in most instances, as it's great for reproducing fine detail - but with high ISO files where the priorities are rather different, I find the IGV algorithm much more effective. This, coupled with "aggressive" colour noise reduction - plus, when there are large areas of solid tone, just a little luminance noise reduction (no more than +15) - and no sharpening, typically produces results I'm quite happy with. Not as happy as I'd be with an image at base ISO, of course, but my expectations and tolerance shift to match the situation and end use-case.
My apologies to the OP for drifting off-topic, though hopefully they might find this interesting and relevant...
EDIT: See snapshot below... Not a great example, but I chose it because it has a large expanse of grey that really accentuates colour and luminance noise. The second screen capture shows the cropped image at 1:1 reproduction. This is the level of noise I would typically seek to achieve with an ISO 6400+ image from my K-3 (actually, this photo is at ISO 8,000 - but close enough, and serves its purpose here). Noisy, yes... but no maze artefacts and no washed-out appearance, due to choice of demosaicing algorithm. Scaled to fit a screen and viewed at typical distances, quite useable for my purposes. In this case, since the photo doesn't contain fine detail, a little luminance noise reduction improves things still further...
At 1:1 reproduction, with colour noise-reduction only:
... and with +10 luminance noise-reduction: