So, quick analysis here - nice stuff - I've designed a few devices, and Marvell was one of the vendors under consideration (due to NDA's and what not, can't really say, except that these were vertical carrier provided devices, and no, it's not with Arris or someone similar).
All in all, good clean design from a HW perspective, mostly a Marvell reference board, but this is a good thing - bit spendy compared to some others, but it's a beefy implementation for the most part. Chip count is higher here, as this is more of an enterprise/carrier grade oriented HW design - which explains the higher end-user cost, as the bill of material is going to be higher...
Getting into the HW from a Photo-Analysis - I haven't ripped my WRT1900ac review sample to confirm --- yet...
1) Armada XP, Marvell MV72830 - this is a dual core ARMv7 designed by Marvell - it's functionally equivalent to a Cortex-A9, but it's in-house - similar to what Qualcomm has done with Krait/Snapdragon and Apple with A6-Swift.
2) eSATA and USB2 - I suspect this is hosted on the Armada on the SERDES bus natively - they're not mux'ed, but do device detect based on electrical characteristics - clever...
3) GigE Switch is a Marvell 88E6172 "LinkStreet SOHO" - this is a 7-Port GE Switch w/ 5 integrated PHYs w/ 1GMII, RGMII, and EEE - I suspect this sits on one of the GigE ports on the Armada, the other GigE port I suspect hosts the WAN port for routing purposes - I could be wrong here, but I don't think so - 4 ports for back of the router, plus one upstream from the Armada XP - and the Armada XP has three GigE ports on it's own - so one for WAN, one to go the LinkStreet, and one for debug purposes - pretty standard for Marvell - getting terminal access into the WRT1900ac on the Linksys Firmware would be helpful for interface mapping...
4) USB3 - EtronTech EJ198H or similar, hard to tell as the photos are low Res - I suspect this sits on one of the PCIe lanes as a 1x - the Armada XP can either support one 4x or four 1x lanes
5) WiFi - Marvell Avastar 88W8864's - two of them, configured as single lane PCIe - integrated RF, these are 4*4 on the RF side, with three steams supported for SM - nice they're mounted on a daughterboard, so this opens up opportunities for alternate chipsets at some point
6) Flash Memory - hard to tell on the photos - but I suspect this is a Spansion 1Gb (128MB) NAND flash part, most likely S34ML01G2
7) RAM - Looks like Hynix parts here - likely in a 32 bit bus configuration, but I could be wrong here - these are BGA parts on 16 bits each - so it could also be 16 bit, but I'm ok with 32.
8) Since flash is NAND, the 256MB total memory space is actually not true - part of the flash is mirrored into RAM, as NAND cannot execute in place - considering that the compressed firmware image is approx 22MB, this gives us about 200MB of actual memory to work with - also note, as rMerlin has suggested, that the 128MB of flash - they keep a recovery image, so reduce flash by that amount - which gives us around 100MB or so outside of the firmware images - and then additional space for NVItems (settings), so it starts to get a bit tight for adding additional code...
I'll jump into SW in a bit... to summarize some of the interesting things there.