Out of curiosity I created four files with Renoise.
Each of them contained a generated perfect sinewave, and the differences were in dithering (all were exported in 16 bits).
Then I analyzed them with a spectrum analyzer.
Sonalksis
No dither
Sonnox
So it would seem that although Renoise’s dither removes some of the distortion, it doesn’t remove all. Does this mean it’s insufficient?
the buggy dither makes Renoise sound more analogue though 
I’m sorry, but can you please go into more detail?
How exactly did you create the test waveforms? What was their sampling rate, bit-depth, etc.? Were they generated with an external tool such as SoundForge, or rendered from a VSTi?
What are the graphs supposed to be demonstrating? You do not really say how we should interpret any of them.
You say that one is undithered - does that mean it’s the original (totally unprocessed and unconverted) test signal that should theoretically show zero distortion, or do you mean it’s a converted signal (from ???-bit to 16-bit?) with dithering disabled that would theoretically show the worst distortion? (It appears to be more ‘noisy’ so I’m guessing it’s the worst, but you did not specify)
Please be more specific. 
I’m not sure why you think these graphs make dither buggy? Can you explain?
It looks like the noise floor has increased by ~15dB compared to no dither, which I would expect, and is in line with the others. I’d be more worried about the Sonalksis graph, what’s happening at the high end?
Without more information it’s hard to make a judgment though? I wonder how they actually sounded, that’s what matters after all.
Ditto.
If anything, it looks to me like Renoise has the most uniform output and is spread out most evenly across the spectrum?
Sonnox looks to be a close 2nd place, though it’s curving up very slightly towards the higher end.
Sonalksis has some obvious aliasing going on, which explains the big hump right around nyquist (sample rate / 2).
First of all, I’m not trying to attack anyone with this. I’m just curious and would like to know why the commercial dithers are clearly differently designed (I’ll change the wording of the original post in a minute).
I used polyiblit vsti to generate a sinewave. Sample rate was 44100khz, the source bit depth was the vst’s internal bit depth, usually 32 or 64.
The graphs show how the quantization distortion is turned into noise by dither (which is the purpose of dither, see wikipedia or similar for details).
The commercial dithers seem to drown all distortion into the noise floor but Renoise’s doesn’t. This could be a thought-out compromise, but then again
Sonnox (Paul Frindle) and Sonalksis designers don’t seem to agree.
Sorry, but this is wrong. The non-uniformity is called noise-shaping and it’s deliberately done to move the noise to
frequencies where our ears aren’t sensitive. Renoise’s output has clearly distinct tones (distortion products) about
12db above the noise floor.
No worries!
I’m familiar enough with dithering to understand its purpose, but I admit that I don’t know much about the more subtle details and different techniques behind it. I’d read a little bit about noise shaping before, but I didn’t realise it was actually the preferred method for this. Very interesting to learn! Anyway, since you didn’t say what to look for in the graphs, my natural thought was to look for the typical sort of aliasing distortion or something like that, so of course the Sonalkis graph stood out in that regard with all the extra high frequency activity.
Renoises dither is indeed the most simple version of a dither one can imagine. I personally am not a fan of dithering, but this is something which everyone has to decide by its own.
Lets try to add some more advanced dithering modes in upcoming Renoise releases…