The M9 sensor has about 1/2 the noise level of the M8. The M8 has a higher saturation count.
Other sources of banding include the SD card used and condition of the battery. I use a slower 4x card. The thought is cards faster than the write speed of the camera cause uneven power draw. A battery on-the-edge can also cause issues. I spent years of my career chasing down sources of noise for early digital sensors. BUT- the compression scheme used for the M9 and M8 is just an unnecessary way of introducing a chaotic process into storing an image. It is a very bad way of doing it, showed Leica does not understand digital imaging.
With the M8 using M8RAW2DNG.
Below: comparison show to compare compressed with uncompressed, manual exposure same settings.
M8 using compressed DNG, ISO 2500
M8 using button dance, M8RAW2DNG, same exposure as above and pushed in Lightroom.
Noise reduction and sharpening turned off.
1950 KMZ J-3, wide-open.
I took the M8 and M Monochrom to the Marine Museum, each with a 1950 KMZ J-3 on them. I used the M Monochrom to meter the shot, set the M8 to the same exposure. The M8 was using the Button Dance. Adjusted the M8 shots using Lightroom in post, all NR and sharpening turned off.
The M Monochrom is set to ISO5000, M8 is set to ISO 160 with manual exposure set to the same as the M Monochrom, 1/90th second and wide-open.
If the M8 had native ISO5000 performance like this in 2006, no one would have disparaged its high-ISO performance. The main reason I bought the M9 was to have uncompressed DNG. When Arvid introduced M8RAW2DNG - it's all I use on the M8.