After months of listening to immersive audio (24/48 Atmos 7.1.4, 24/96 Auro3D 11.1 and discrete 10ch or 12ch DXD) thru HQPe as DSP engine with very long taps of FIR filters for DRC+XO, so far I preferred:
First choice: everything upsampling to 12ch DSD256 using poly-sinc-ext3 (1x) / poly-sinc-ext2 (Nx) + ASDM5ECv2.
The SQ is the highest but the computer’s power consumption is… more than 700w (sustained). It hurts the electricity bill… 
Second choice: 48 or 96 immersive audio contents upsampling to 12ch PCM 24/352.8 (DXD) using poly-sinc-ext3 (1x) / poly-sinc-ext2 (Nx) + NS1.
Output DXD to Merging’s DAC, the data path in the equipment will be the same as DSD (other than DXD/DSD all audio data will enter the FPGA’s DSP) but the power consumption will be very low. This combo has deeper bass and more relax feeling, good for long listening. 