None of this is "one size fits all." The results/performance can vary depending on what is being used as a "server." With my Mojo DejaVu server (and with the Antipodes DX3 I still own), I have the option to output an ethernet cable from the server to a separate Roon endpoint, or renderer, such as the Bricasti M5, Metrum Ambre, Sonnet Hermes, or similar. In some cases, removing the renderer from the server unit can be sonically beneficial, but not always.
I also have the option with both of the above servers to output USB directly from the server into my DAC. In other words, both of those servers can also function as the renderer. In the case of the DejaVu, the USB output board is the excellent JCAT USB Card XE, which results in the USB output performing at least sonically equal to using a separate renderer (when using the renderers and DACs I have tried in my system). I suppose, I could try inserting the ENO between the server and the renderer to evaluate whether that makes a sonic difference.