$125 - (Dell Labeled) Aquantia AQS-107 SFP+ to RJ45 NBase-T Transceivers

tinfever

New Member
Nov 3, 2018
8
4
3
These are the AQS-107 transceivers that may be the only kind on the market that can reliably let you connect NBase-T connections (like 2.5G from your ISP) to the SFP+ port on a device that only supports 10G link negotiation (like many of the Brocade switches).

These are hard to find in general, currently out of stock everywhere, and usually run $200+.

I can't vouch for these working or say if the Dell branding might cause issues, but these hit on one of my eBay saved searches.

Currently $125 with 5 in-stock.

Relevant threads:
 
  • Like
Reactions: mach3.2

int0x2e

Member
Dec 9, 2015
94
69
18
43
3 ideas for someone who may be doing this to connect to a 2.5gbps GPON ONT -
1. These may be much better, but there was a review on the STH website that compared NBASE-T SFPs that claimed some worked fine at 2.5.
2. If you're like me, and your GPON ONT came in a Sfp form factor, it won't work reliably at anything other than 1gbps on most devices, but there are people saying that the Broadcom 57810S supports 2.5gbps mode correctly. A lower cost device based on the same chipset is the Dell N20KJ.
3. Patrick did a review of an aliexpress unmanaged switch that supports 2.5-base-t copper and has SFP ports. That may be another interesting solution.
 

Cruzader

Active Member
Jan 1, 2021
210
226
43
1. These may be much better, but there was a review on the STH website that compared NBASE-T SFPs that claimed some worked fine at 2.5.
There are cheaper that can do 2.5 in general, but not that will actualy do a 2.5 link in older switches that do not look at 2.5 "as a thing".

The AQS-107 will show as 10gbe on a old unit like the popular brocades but 2.5gbe on the other end and function as 2.5gbe
its the only model i know of that will do a 2.5gbe link in sfp+ slots that does not have 2.5/5 in its spec, it converts on its own chip something that is not a common feature.
 
Last edited:

Dave Corder

Active Member
Dec 21, 2015
205
113
43
40
Going by plain Dell part number (with lost leading zero) there are a few even cheaper ($45):
That's a steal! Where was this when I had to buy one for my Xfinity connection?
 

Cruzader

Active Member
Jan 1, 2021
210
226
43

xinpig

New Member
Feb 17, 2021
7
3
3
There are cheaper that can do 2.5 in general, but not that will actualy do a 2.5 link in older switches that do not look at 2.5 "as a thing".

The AQS-107 will show as 10gbe on a old unit like the popular brocades but 2.5gbe on the other end and function as 2.5gbe
its the only model i know of that will do a 2.5gbe link in sfp+ slots that does not have 2.5/5 in its spec, it converts on its own chip something that is not a common feature.
So I could hook a 2.5gig connection from my motherboard on one end, and then put the transceiver into an icx 6450 and not have any issues? Been trying to find a way to make use of 2.5gig on my PC without spending a ton of money.
 

tozmo

Active Member
Feb 1, 2017
123
86
28
73
So I could hook a 2.5gig connection from my motherboard on one end, and then put the transceiver into an icx 6450 and not have any issues? Been trying to find a way to make use of 2.5gig on my PC without spending a ton of money.
At $125 per transceiver, I'm not sure about the value of this.
 
  • Like
Reactions: abq

Cruzader

Active Member
Jan 1, 2021
210
226
43
So I could hook a 2.5gig connection from my motherboard on one end, and then put the transceiver into an icx 6450 and not have any issues? Been trying to find a way to make use of 2.5gig on my PC without spending a ton of money.
Ive used in 6610 so would assume it works just as well in the 6450.

At $125 per transceiver, I'm not sure about the value of this.
The value is not putting 300-400$ into a small 4x 2.5gbe managed with sfp+ just to gain a few 2.5gbe ports.
Something that is more common than you would probably expect.

To not have that added consumption/space/failure point.

Tho its mostly a 60-80$ item not 125$.
 

ano

Active Member
Nov 7, 2022
121
51
28
with 10/40g pricing the way it is, the 2.5G seems kinda weird that its gaining so much traction, seeing as 10G-baseT allready uses very little power.

only real usecase we have for stuff like this is with limited IO options and 10G-baseT allready onboard.
 

tinfever

New Member
Nov 3, 2018
8
4
3
Going by plain Dell part number (with lost leading zero) there are a few even cheaper ($45):
Wow. Well played. I'd even searched for the Dell PN and found no listings, but I was using the full PN with the 0.

The seller is probably wonder why they suddenly sold all their units within 3 hours after a month of listing haha

It stings a little to miss that deal!

with 10/40g pricing the way it is, the 2.5G seems kinda weird that its gaining so much traction, seeing as 10G-baseT allready uses very little power.

only real usecase we have for stuff like this is with limited IO options and 10G-baseT allready onboard.
I wish everything was just 10G but when your Google Fiber ISP hand off only negotiates at 2.5G and you want to have dual routers/firewalls for redundancy so the connection has to go in to switches that don't speak 2.5G, gotta make some lemonade from the lemons.

So I could hook a 2.5gig connection from my motherboard on one end, and then put the transceiver into an icx 6450 and not have any issues? Been trying to find a way to make use of 2.5gig on my PC without spending a ton of money.
The sad thing is it would probably be cheaper for you to get a ConnectX-3 NIC and run it at 10G to the switch, than it would be to adapt the switch to 2.5G using one of these transceivers. All assuming you have the free PCIe slot/lanes.
 
  • Like
Reactions: noduck

xinpig

New Member
Feb 17, 2021
7
3
3
The sad thing is it would probably be cheaper for you to get a ConnectX-3 NIC and run it at 10G to the switch, than it would be to adapt the switch to 2.5G using one of these transceivers. All assuming you have the free PCIe slot/lanes.
My Dark Hero has an open pci 4.0x4 slot and I have a ConnectX-2 with 2x 10gb ports lying around somewhere.
 
Last edited:

dag

Member
Apr 23, 2020
15
26
13
I don’t want to piss on everyone’s parade, and I realize it’s an old-ish post, but a word of advice: the Dell flavor only officially supports 10GbE. Having said that, it also unofficially supports 5GbE, but nothing below that—so no support for 2.5GbE. I have both the Aquantia and Dell branded SFPs, and while the Dell will establish a link at the PHY level at all speeds, including 2.5GbE, not a single packet gets thru at anything <5GbE. Conversely, the Aquantia flavor is as happy as can be at all speeds.

They’re identical, except the Dell version I have shows revision AQR107*A* on the chip (I have 2 Dell SFPs, same for both):
B240C63F-3810-462E-9E5D-A82A9FB6D2AC.jpeg

Since I can see a link at the PHY level on the Dell at 1GbE and 2.5GbE, it’s either a crippled version of the 107 purposely sold to Dell so it doesn’t cannibalize its own switching business, or it might just be a revision of the chip, and the connectivity is crippled at the firmware level. Hoping it was the latter, I’ve tried to identify the chips on the board in hope one of them was an EEPROM I could mess with. See pic for back side (the purple/white gunk is from the stock thermal pad).
DFDAC33C-3FF6-4F50-AA0A-905AC756D4DA.jpeg

The hope was to dump the data from the Aquantia flavor, and flash the Dell flavor with it. Easier said than done. The “2165” chip is an ADP2165, it’s a DC-DC regulator, not what we’re looking for. Out of the other 3, the most likely candidate is the one on the top right, labeled “4H120”. Here’s a close-up:
AC6FC545-15AE-4879-8470-5059DC270F43.jpeg

The other 2 are unlikely to be EEPROMs, their markings aren’t helpful so I’m not 100% sure what they actually do.

I’ve tried, but I just can’t identify the “4H120” chip. I plan to desolder it and mess with it with flashrom, however I’m a little worried I might destroy a perfectly good SFP module, the chip is so tiny I know I will have a hard time when it’s time to put it back (it looks like a TSSOP-8 to me).

Other than that, both Dell and Aquantia flavors may be old, but they still work vastly better than any other SFPs when it comes to anything < 10GbE. Aquantia was really ahead of its time, Marvell killed all the good stuff when they acquired them and discontinued the product line while pushing their Marvell Alaska chip instead, which is in 99% of the SFPs sold these days. Alaska is utter rubbish at 2.5 and 5GbE, it just can’t handle the traffic well, and it requires the host to handle the intermediate speeds natively as well. Conversely, Aquantia is able to fool your 10GbE-only switch into thinking the link is at 10GbE—the AQR107 constantly introduces PAUSE frames to “tame” its host, and while some purists will consider it a bit crude, in real life, it works really well.
 
Last edited:

Dave Corder

Active Member
Dec 21, 2015
205
113
43
40
I don’t want to piss on everyone’s parade, and I realize it’s an old-ish post, but a word of advice: the Dell flavor only officially supports 10GbE. Having said that, it unofficially supports 5GbE, and supports nothing below that—so no support for 2.5GbE. I have both the Aquantia and Dell branded SFPs, and while the Dell will establish a link at the PHY level at all speeds, including 2.5GbE, not a single packet gets thru at anything <5GbE. Conversely, the Aquantia flavor is as happy as can be at all speeds.
Ooh, interesting! Looking forward to learning more about what you find out.
 

tinfever

New Member
Nov 3, 2018
8
4
3
I’ve tried, but I just can’t identify the “4H120” chip. I plan to desolder it and mess with it with flashrom, however I’m a little worried I might destroy a perfectly good SFP module, the chip is so tiny I know I will have a hard time when it’s time to put it back (it looks like a TSSOP-8 to me).
Thanks for chiming in! Even if it's not what everyone would want to hear, it's a heck of a lot better than not knowing!

Not sure if this would be any help but doing some internet sleuthing I found a schematic for a FPGA daughter board that also uses the AQR107 chip (attached for posterity). It shows the AQR107 connects to an SPI flash chip. I'd expect at least parts of this daughter board schematic to be part of the reference implementation for the AQR107 chip, so there might be some sections in common with these SFP+ modules.

Also, that 4H120 chip looks more like DFN/QFN to me in your photos. The schematic references a AT45DB041E flash chip, which comes in an "8-pad UDFN" package that looks pretty similar to your photos to me. No telling if it's the exact same make or model of chip though.
 

Attachments

dag

Member
Apr 23, 2020
15
26
13
Also, that 4H120 chip looks more like DFN/QFN to me in your photos. The schematic references a AT45DB041E flash chip, which comes in an "8-pad UDFN" package that looks pretty similar to your photos to me. No telling if it's the exact same make or model of chip though.
Nice find, thank you! The Adesto EEPROM in the datasheet you unearthed is a clone of Atmel’s AT45DB041D, it should be supported natively by both flashrom and buspirate via some minor tweaks. Main caveat: I just can’t find anything that links “4H120” to an AT45DB. But the good news is now we know the EEPROM in the SFP uses an SPI interface, not I2C, and it’s likely to be 4Mb (though it could be more/less).

The IR3895 in the daughter board you uncovered matches the ADP2165 in the SFP (though there are some difference e.g. input voltage range). The same goes for the 50MHz oscillators. My point is, overall, there is a high degree of similarities between the SFP and the datasheet, which reinforces my assumption the EEPROM is indeed the 4H120 chip, which should be able to be read like an AT45DB, however the pinout is unknown.
 

Dave Corder

Active Member
Dec 21, 2015
205
113
43
40
I noticed last night that both a Dell-branded version and a "real" AQS-107 show zero SFP information in my MikroTik CSS610 switch. Haven't tried them in a NIC yet to see what happens there or what ethtool -m shows for the SFP EEPROM.
 

dag

Member
Apr 23, 2020
15
26
13
I noticed last night that both a Dell-branded version and a "real" AQS-107 show zero SFP information in my MikroTik CSS610 switch. Haven't tried them in a NIC yet to see what happens there or what ethtool -m shows for the SFP EEPROM.
Same here, in some devices I get nothing, and in others I get a bunch of gobbledygook.
 

dag

Member
Apr 23, 2020
15
26
13
So… I managed to get a hold of an old AMX765. It looked so much like the initial revision of the Aquantia SFPs that I had to open it to check. Lo and behold, here is what was lurking inside:
E84B200F-995E-4957-BE4D-EB97CCC613A3.jpeg
Interestingly, the Aquantia chip also shows revision A. The 1713 marking most likely means 13th week of 2017, so it’s an older chip. And, just like the Dell, the SFP won’t let packets go thru if the PHY is synced <5GbE. Long story short, it leads me to believe the A version is not merely a later revision of the chip, it’s more likely a special version of the chip Aquantia built for OEMs that didn’t want to allow people to add 2.5GbE support to their 10GbE-only switches (a pretty sh*tty, yet not surprising move). In other words, my hopes of using the firmware from the vanilla Aquantia SFP are pretty much dashed at this point.

Another interesting tidbit, the EEPROM is a SOIC8 chip (bottom left):
38718AA8-2B9E-4D52-89A1-A7CA80847386.jpeg

Hoping it’s an Atmel clone, I should be able to check without desoldering it.

The mystery deepens…
 
  • Like
Reactions: Aluminat