The internal architecture of the single-channel Burr-Brown/TI PCM-63 and PCM-1704 uses a pair of offset R2R ladders (called Co-Linear) to eliminate the MSB switching transition (from 1000... to 0111...) that plagued earlier R2R converters.
In a an e-mail exchange I had a few months ago with Kingwa of Audio-GD, the Master 7 uses 8x 1704's, paralleling two of them in the plus direction, and paralleling two of them in the negative direction (using digital phase-splitting ahead of the converters), per channel, adding up to 8 in all. The 1704's operate in push-pull mode, driving the differential input of the ACSS current-mode analog stage.
The Reference 10.32 simplifies things a bit, retaining the push-pull architecture, but only using a single 1704 in each direction, adding up to 4 in all. The Reference 10.32 also has a volume control and a headphone-capable linestage, and retains the balanced-drive ACSS current-mode analog circuits of the Master 7.
The NOS-1704 and SA-2 use 4x 1704's, but they are paralleled instead of operating in push-pull mode. Paralleling would seem to offer no obvious advantage, except that the NOS-1704 and SA-2 use resistive I/V conversion with passive lowpass filtering, and the doubling of current offers a 6 dB improvement in noise figure from the following diamond-differential analog stage. Although the input and output of the diamond-differential stage is single-ended, the internal operation is fully balanced and in Class A (unlike opamps, which commonly operate in Class AB to keep cool).
Based on photos in other forums, the NOS-1704 is an SA-1 with the upsampler circuit deleted, and a selection between two different S/PDIF interface chips, one good for 96/24, and the other for 192/24.
According to Kingwa, both the ACSS and diamond-differential stages have a very high slew rate to avoid slewing the transitions between digital steps, a known problem with opamps in other DACs. (I've measured a 20 MHz comb spectra coming out of a Burr-Brown PCM-63, and the newer converters are not slower.)
Although the Master 7, Reference 10.32, SA-2, and NOS-1704 aren't the cheapest DACs on the market, the Burr-Brown/TI PCM-1704 converter is priced at US$74 each, and they are mono devices. In comparison, the ESS Sabre 9018 is around $17 each in 100x quantities, and they are 8-channel devices, paralleling 4x together for stereo operation. (Based on conversations with ESS engineers, the 9018 has best performance in current mode with a virtual load of less than 10 ohms; voltage mode is only for low-cost applications like home theater.)
Nobody seems to know if the PCM-1704 is out of production or not; the spec page warns against new designs, and the pricing has substantially increased in the last few years. Rumor has it that the military and aerospace customers demand a well-characterized R2R converter with a stopped-clock capability, and the PCM-1704 is one of the very few that still offers that, so TI is still making them in limited quantities. (Perhaps TI is selling existing stock. Or maybe all three, starting and stopping the line as customers demand.)
It's a good question what the sonic differences between the ACSS (Master 7, Reference 10.32) and diamond-differential (SA-2, NOS-1704) circuits are. My power amplifiers are fully balanced and transformer-coupled throughout, so technically they would be topologically closer to the diamond-differential circuit.
