Teardown of the ADOP 10GMS-30M-T 10Gbase-T Adapter

[up-n-atom]

The review(s) and buyers guide on the cheap 10Gbase-T adapters are decent but I wish they would've been a little more thorough by taking them apart to see what truly differentiate them. Perhaps they are all just the same OEM design and software defined, ie. newer firmware? I don't have a reason to purchase each brand, so I purchased the cheapest available to me and took a look inside...





MCU: SiLabs C8051F396-A-GM (C2 programming 2x2 matrix)
PHY: Marvell 88X3310
PMIC: Intel EN5339QI
NOR: GigaDevice GD25Q80
OTHER: AM0Q - U905 (BACK) AK19 - U907 (FRONT)

Based on the hardware and the datasheets (Marvell 88X3310) there is no reason that the ADOP 10GMS-30M-T shouldn't support 2.5/5G, so I'm inclined to believe it's blocked by firmware. Maybe collectively others will share internals of of the other adapters from the buyers guide pending the visible PCB differs.

 

[up-n-atom]

The C2 interface of the SiLabs C8051F396-A-GM are traced to the the 2x2 pad matrix on the top layer along with power and ground, but unfortunately it's locked and cannot be read.

The Marvell 88X3310 firmware is contained on the GigaDevice GD25Q80 and ISADetect.com detected the architecture as armhf.







The firmware is attached. It's version is 0_2_8_0_8850.

It's equivalent to https://github.com/acooks/tn40xx-driver/files/2213763/MV88X3310_phy.h.zip which contains x3310fw_0_2_8_0_8850.hdr but it's not binary equal; There is a minor difference within the header and additional data packed at the end. Newer firmware from the tn40xx is available Problem with .hdr file versions · Issue #3 · acooks/tn40xx-driver

The 16 byte header is followed by a checksum = 0xFFFF - SUM(0..15)

EDIT: Found a schematic with an 88X3310 https://opencores.org/usercontent/doc/1494585946

 

[hmw]

There's definite differences in behavior between Aquantia and Marvell based adapters. The Aquantia ones seem to handle 2.5/5/10G seamlessly. The Marvell based ones struggle when put into a 10G SFP slot and connected to nbase-t - either 2.5G or 5G copper cabling. They do handle 10G and 1G just fine

 

[up-n-atom]

There's definite differences in behavior between Aquantia and Marvell based adapters. The Aquantia ones seem to handle 2.5/5/10G seamlessly. The Marvell based ones struggle when put into a 10G SFP slot and connected to nbase-t - either 2.5G or 5G copper cabling. They do handle 10G and 1G just fine

The Aquantia AQS-107 were definitely in a league of their own as with all their chipsets, it's no wonder Marvell gobbled them up.

MikroTik recommend that their S+RJ10 be run with auto-negotiation and specifically auto duplex MikroTik wired interface compatibility - MikroTik Wiki and that the rev 2 fixes all known issues. Why these were sold with problematic firmware and no update procedure is incomprehensible. I suspect the SiLabs MCU has the ability to do firmware updates beyond its DDMI support, but without access to its firmware it's a premature assumption. The convenience to plug into an I2C master for firmware updates could resurrect these from being paper weights.

 

[nasbdh9]

Some modules based on the Aquantia solution can "translate" 5G/2.5G in the PHY, and then the negotiated rate from the PHY to the MAC of the network card is 10G. In this way, the MAC link of the network card that does not support 10G can support 5G/2.5G.