Siberian Patient

You can clearly see the banding in the JPEG picture whereas I can barely see it in the PNG one, or maybe what I see is due to my shitty laptop TN display.

 

PNG gradient: http://postimg.org/image/wb5o7976z/

JPEG 75%: http://postimg.org/image/mrxc0lj1b/

JPEG 100%: http://postimg.org/image/u5ouxfsvp/

 

Well, there is still minor banding on PNG gradient at my "8 bit" laptop's not calibrated panel, but it have consistent and uniform decrease of brightness from left to right, JPEG even at 100% is inferior and have additional narrow vertical bands, 75% is even worse. Maybe my panel use some light dithering, maybe it's normal for PNG, maybe it so becuase it not calibrated and use manufacturer's software preset.

In any case, external recorder with 8 or 10 bit lossy 'uncompress' would give the best possible image.

Sadly, but with reasonably priced semi-pro recorders only 60i or 30p are possible.

Also, the banding that you see in the first place with an 8bit codec on the GH4 is most likely due to the H.264 compression, for instance if you look at those three pictures I have included, one is in PNG 8bit/channel, no loss of information, the other one is JPEG 8bit/channel, lossy. 

I will completely agree with you here, banding/blocking is not 8 bit color representation issue, it's camera's onboard H264 encoder's issue.

Most lightweight cameras nowadays have it in some form, with exception of those who record to DNxHD, ProRes or another sort of proprietary high bitrate codec.

Simple test: connect camera's HDMI to monitor/capture card/external recorder and compare picture taken from HDMI with picture recorded by camera to SD-card, the last will be with some or another type banding/blocking artifacts, depending on footage type.

Another test: check Vimeo for Sony RX10 footage captured to Atomos Ninja - there is no any visible banding as compared to RX10 internal encoder's picture, the last gets muds and bandings sometimes.

Keeping in mind 2 things: 1st - many 10 bit and true 8 bit (not 6bit with FRC) panels on the road to consumers, & 2nd -  free of banding/blocking H264 encoding requires  computing power 4 cores/8 streams x86 CPU @ 3.5-4 GHz for 30fps (I mean x264 here - more viable option to sustaining portfolio in good form through coming years, without banding&added noise will be capturing to external HDMI encoders with little compression, instead of using camera's onboard encoder.

Looks like there is still no enough megaflops-per-watt requiered for free of banding H264 hardware encoding in lightweight cameras.

As for personal use - adding dithering noise to footage with banding may be the solution.

But, if turning to dithering in audio reproduction - there are many types of dithering noise and those different types are recognizable by listener.

Eye is more sensitive/complex/information dense 'device' than ear, so the question is what type of dithering is the best for videos taken from onboard encoders.