[RyanC]

I've been down this path a bit, albeit with a quantum instead of the 4848 and can chime in on a few things.

-Driver quality for the Quantum is good, but not RME good. What do I mean? Overall it works pretty well, but some plugins glitch at the lowest buffers and don't one RME (For me, the RME is a HDSPe MADI in an AKITIO node 2). For instance Black Rooster Audio plugins, I could not track through these on the quantum, no problems on the RME. RME does have slightly higher RTL than quantum (.8 ms with ferrofish at 32/44.1).

-Studio One is more stable on windows for me. Maybe this will change with a code refresh for AS native, but as-is, it's just rock Solid on Win10 and has had some issues, particularly with huge sessions with lots of VI's on OSX.

-AMD Zen3 is the current low latency king, and not by a small margin. I compared M1 MBP vs 5950x, single-core using AS native code in Reaper with FF Pro-R. At the 32 sample buffer, the 5950x can do 11 instances of Pro-R to the M1's 6 (on one track/core). I leave my 5950x rig set to the 32 sample buffer for 44.1/48 and 64 for 96 always. Why single core? Because for LL performance, plugins in series create a series process dependency and ZLM in S1 is single core only.

The M1 is off to a good start and either or those is going to blow you away coming from a cheesegrater. Of course with the M1/2 being a radically different architecture, it's possible that some LL performance varies from plugin to plugin more than comparing x86. I did check a couple other FabFilter plugs at 32 samples and 5950x held at slightly less than double. As you increase the buffer sizes, the 5950x holds the lead, but not by as wide of a margin.

That's just sort of my journey there, which sounded similar enough to yours so I thought I would share. I have a multi-user studio that caters to freelancers, so I will always need macs around, PC's are more optional. I'm excited about what Apple is doing too...