Having found improvement from 4 paralleled chips I was curious whether further doublings of chips would lead to even more improvements. Going from 4 to 8 chips necessitated lowering the noise of the I/V stage in order to get some semblance of a level playing field for comparison. Already I was using paralleled stages as the opamp I used wasn't particularly low noise itself (AD744) - this opamp was chosen because I was wanting to experiment to see if I could improve the HF cleanliness and AD744 is one of the very few opamps which allows bypassing its classAB output stage (OPS). I really needed a bipolar input, undegenerated input stage opamp with a bypassable OPS to get the number of paralleled stages down to a manageable size. The opamp also needed to be cheaply available on Taobao. After a fairly brief search, I landed on AD829 - a part I'd never come across before but which fitted the bill perfectly. It also has a considerable extra advantage in that it is compensated for a minimum gain of 26dB - almost unheard of in opamps. Even 'decompensated' opamps usually have under 20dB minimum gains. This 26dB figure turns out to be a perfect fit for the I/V stage - its also under half the price of AD744 so all round an excellent solution here. A single AD829 with its 2nV/rtHz noise figure is a good match for 4 DAC chips, to go to 8 needs 4 paralleled for the same noise level as I want to keep the output compliance voltage constant meaning halving the LC filter's impedance. It took some layout tweaking to get the AD829 stable whereas the AD744 gave no issues whatsoever, eventually I got the AD829 array working and sounding great, slightly better (by which I mean more engaging) with 8 chips than 4.
So then what about 16 chips?