So, sorry, but the visual evidence says otherwise...there IS an IQ improvement between the 5D II and 5D III, in many ways a significant improvement.
Thanks for doing these gifs, it's interesting and I think I can spot the 5d2's banding - but maybe I'm just looking for it.
Oh, its definitely there...I see it clearly. Same crap I have in my 7D, too.
Having said that, *significant* in a non-scientific context is very subjective, as far as I remember the context then was the horrendous price jump to $3500 that caught many people off guard and created higher expectations towards the sensor than Canon currently can (you'd probably say: wants to) deliver.
I'd say "Than Canon could have delivered"...past tense, the 5D III is only a couple months away from it's second birthday. I also believe that Canon would be incapable of producing any higher resolution sensors on their current fabrication process, and I believe in two years time, they could have improved.
What doesn't show up in the gif and what I have to admit I'm guilty of underestimating: The newer ff sensors react *much* better to postprocessing either in nr or sharpening, multiplying the seemingly moderate step up in noise pattern. Esp. with DxO's prime nr it's stunning how iso 6400 looks on the 6d, if only it wouldn't take my laptop 30 minutes to denoise a single picture :-p...
Aye, which is in significant part due to the considerable improvement in banding. Horizontal banding was pretty much eliminated, and vertical banding occurs in these "soft" vertical stripes, rather than the harsh kind that occurred in sensors prior to the 5D III and 1D X. The more random, "natural" appearance of the noise, rather than a patterned, unnatural appearance, greatly helps in it's elimination.
As for DXO, while I admit I may be doing something wrong, I have found their software to be the worst of all the available options for editing RAW files. DXO seems to produce the noisiest results, PARTICULARLY for Canon files (they do much better with Nikon files). Compared to LR, DXO tools result in what I would call about two thirds of a stop WORSE noise performance strait out of camera. Compared to DPP, it is more like a stop worse (I do have to say, as much as I hate DPP's UI, it produces the cleanest noise output for Canon RAW files of ANYTHING, free or for pay...it's really too bad Adobe has't looked into Canon's own RAW demosaicing algorithms.)
I don't know if it is an intentional bias, or just a fundamental lack of interest in properly supporting Canon. I have given DXO's tools several dedicated tries, but in general they are lacking, they seems to be far
slower than Lightroom or DPP, and specifically in Canon's case, the output is just terrible. All things being equal, that isn't surprising. Canon is not a DXO supporter, DXO has never given Canon much time or interest (it is often months or even years before DXO will test certain Canon cameras, whereas they will test Sony and Nikon cameras right out the gate, as soon as they can get their hands on a few copies.)
If you want the cleanest RAW conversions, DPP can't be beat. It's standard deviation of noise is about two thirds to half that of LR, and a full order of magnitude better than anything I've tried from DXO. (The only thing I DON'T like about it is it doesn't deal with aliasing as well as LR...edges come out of LR with this clean, crisp look, whereas you can clearly see stairstepping and in some cases moire a lot more often with DPP.)