WiFi 6 (ax) adapter

probably bears down to

mvebu-pcie soc:pcie: PCI host bridge to bus 0000:00
00:01.0 PCI bridge: Marvell Technology Group Ltd. Device 6820 (rev 04)
00:02.0 PCI bridge: Marvell Technology Group Ltd. Device 6820 (rev 04)
00:03.0 PCI bridge: Marvell Technology Group Ltd. Device 6820 (rev 04)

bus version 1.1 | 2.0 | 3.0 | 4.0 | 5.0 and number of lanes 1 | 4 | 8 | 16 and coding (ballast)


Theoretical gross bandwidth, practical net bandwidth potentially loss since transfer protocol takes its bite.
PCIe version bandwidth per link x1 x4 x8 x16 Coding (balast in %)
1 2.5 GT/s 2.5 Gbit/s 250 Mbyte/s 1 Gbyte/s 2 Gbyte/s 4 Gbyte/s 8b10b / 20%
2 5 GT/s 5 Gbit/s 500 Mbyte/s 2 Gbyte/s 4 Gbyte/s 8 Gbyte/s 8b10b / 20%
3 8 GT/s 10 Gbit/s 0.9846 Gbyte/s 3.938 Gbyte/s 7.877 Gbyte/s 15.754 GByte/s 128b/130b / <2%
4 16 GT/s 20 Gbit/s 1.969 GByte/s 7.877 Gbyte/s 15.754 Gbyte/s 31.508 GByte/s 128b/130b / <2%
5 32 GT/s 3.9 Gbyte/s 15.8 Gbyte/s 31.5 Gbyte/s 63 Gbyte/s 128b/130b / <2%

Not sure but seems that MV6820 provides PCIe 3 on the PHY Datapath Width Mode. Unclear the number of lanes, probably not 16 since that usually applies to GPUs.


The next bottleneck are then the 1GbE ethernet ports, upstream or downstream where the bandwidth gets throttled.

1 Like

Well, this one features a reasonably 4 antenna count (not pertinent to the thread - just saying)

@Pepe could you possibly give a hint on how fast each mPCIe-slot is connected to CPU?

connecting a UART console it shows (output depending on the hardware installed)

Detected Device ID 6820
board SerDes lanes topology details:

lane # speed type
0 6 SATA0
1 5 USB3 HOST0
2 5 PCIe1
3 5 USB3 HOST1
4 5 PCIe2
5 0 SGMII2

If that is any indication to go by it would seem to be single lane (x1) for each PCIe slot.

If it is indeed PCIe3 x1 for the slot with the Wlan card it would mean max. 0.9846 Gbyte/s (less protocol overhead) and thus render 5GHz ax superfluous in terms of expanding bandwidth on that spectrum.


deriving from

ahci-mvebu f10a8000.sata: AHCI 0001.0000 32 slots 2 ports 6 Gbps

it would seem that the speed tags listed in the above topology table are indicative of the PCIe link’s bandwidth in Gbps.


SGMII2 = Serial Gigabit MediaI Independent Interface and 2 likely indicating 2.5GBASE-T (over legacy Cat5e/6 cabling) according to IEEE 802.3bz
Thus the speed tag 0.

Further reading on IEEE 802.3bz for the interested

1 Like

If @Pepe or someone from Turris Team will read this they could also tell if it maybe would be possible to make mPCIe ax card to future-proof Omnia and Mox devices. I know that it’s very hard to do it and number of chips/cards must be high but it can be “solved” using Indiegogo. I would personally prefer at least two form factors/versions - half one for NB (1x1 2x2 or whatever could be packed in such size and reasonable price) and full for Omnia (3x3?)

1 Like

Don’t know about the M but the TO would require a new board and other hardware components if

is meant for w/lan bandwidth in excess of 1GByte/s.

But maybe I miscomprehended the meaning of future proof?

https://forum.openwrt.org/t/mt7621-hw-flow-offload-and-4x4-4-n-ac-wifi-router/40674


A bit off-topic perhaps - would reckon (roughly and rather speculative) that a soho cpe featuring:
  • 1 x SPF+ (single of combined with)
  • 1 x 10GbE WAN
  • 1 x switch with 5 x 10GbE
  • 3 x 3 6Ghz ax
  • 3 x 3 5GHz ax
  • 2 x 2 2.4GHz ax
  • SoC quad-core @ 2GHz
  • 2GB DDR3 (preferably though DDR4)
  • 8GB hard drive (preferably not eMMC)

could break even as of today’s pricing basket, after

  • r&d
  • design
  • engineering
  • procurement
  • manufacturing
  • quality assurance (& testing)
  • certification
  • shipping
  • insurance
  • warranty
  • marketing
  • manpower (software maintenance, repairs, support)

, at the equivalent of ~ USD 900K. Projecting that on serial production output of 1,000 pieces makes it ~ USD 900 a piece, that is non-profit.

1 Like
which may still be a good while

https://forum.openwrt.org/t/mt7621-hw-flow-offload-and-4x4-4-n-ac-wifi-router/40674/5

802.11ax does improve the situation again (even if you only have 802.11ac clients), but currently none of these devices/ targets are supported by OpenWrt (there seems to be some initial work on ipq807x and ath11k, but I wouldn’t expect that to become ready within the next couple of months).

1 Like

So after FTTH, now even DOCSIS https://www.marketscreener.com/SERCOMM-CORPORATION-6495319/news/Sercomm-Announced-the-World-s-First-Verified-for-Interoperability-DOCSIS-3-1-Device-with-2-5Gbps-E-29146482/ passed the 1 gig mark, and with that over 100 million customers that need faster routers.
Would be nice to have something from you guys, because your stuff is awesome and i miss it.

Sry to write this, but did you read what @anon50890781 wrote above? On Turris Omnia there is speedwise no option for WiFi6 (but maybe in far future with USB-WiFi6?). On MOX it might in theory be an option (modules are interconnected with 2,5GB/s - but that is shared ressources and MOX lacks 2,5GBE upstream connectivity), but the only available WiFi 6-card on the market by now is crippled 2×2 and necessary drivers are not available (but you may try and compile the Intel sourcecode yourself, I linked it somewhere above).

1 Like

That device is foremost a cable modem with added router capability, whilst TO provides router products without modem capability.

That aside the mentioned device does not look like to feature any Wlan capabilities, let alone WiFi6

i gave up on mox and omnia, they won’t handle the bandwidth i think. I look for a Omnia v2 with WAN 2,5 gig and atleast one LAN 2,5 gig port

That is only a modem no Router

So, why mention a cable modem without WiFi in this thread about WiFi 6? The TO is on par on the upstream WAN bandwidth with what that cable modem is capable of.

1 Like

Because you need a faster WAN port to feed the wifi, the TO does only have gigabit and the modem can provide 2,5 gig, how is that on par?
I just wanted to point out that the complete eco system is shifting beyond gigabit and that where are no real solutions beside using the router of the ISP or building something your own with x86 and Intel X550-T2 as router. I’m not a fan of both.

TO provides 2.5G in the SFP cage.


Perhaps you are lucky if such broadband capacity is provided at your location. Even in Europe there are a lot of places that not even get 100M, let alone less developed places.

But I am digressing from the WiFi6 topic.

Maybe you could collaborate with TO to get a new crowd funding campaign going for a new router device.

1 Like

Yes but it is shared and not dedicated afaik? or am i wrong at that? also going TP -> media converter to SFP -> omnia is not really a nice solution i would say.


Vodafone bought in many countries the DOCSIS networks from liberty global and they will provide for 2/3 till the end of the year gigabit, which is in reality 1150 mbps. And they currently work at a new box which does have 2,5 gbps LAN ports and they are looking to launch it mid next year with 2 gbps offers bundled

that are 8 million customers in germany alone and many more in other vodafone countries.
So i don’t see how that is beeing lucky, looks more like mainstream to me with now the big players hopping on the <1 gbps train.


that would be nice.

That is WAN on CPU port, shared how and amongst what?


For DSL one needs a modem, for cable one needs a modem.


Compared to roughly 41.5 million households in Germany that would seem hardly like mainstream, but that is perhaps a matter of perspective.

That aside the ISP with such a large user base has different ways to subside the cost for CPE, let alone buying power when procuring CPE (components) from manufacturers - nothing TO could compete with.

How many out of the cited potential 8 million households do you reckon will spring in excess of 300 bucks for a single piece of CPE?


That is cable and requires a modem and them being bulky by design.

You are mixing up TO and MOX. MOX shares a 2,5GB/s on the interconnections but NOT on any mPCIe (where it is necessary to plug a WiFi card) and NOT on WAN- or LAN-ports.

I thought it had been clarified somewhere else that this port is only in theory 2,5GBE but connection to CPU is limited to 1GBE. But there’s anyhow no real world scenario that includes the SFP that can help on this topic.

I think one must accept that TO will not go beyond 1GBE connectivity. True 1GBE yes - even with WiFi(5) -, but not beyond.
A potential option is the MOX - so this would be an option for an access point - but that would require new modules with 2,5GB/s mPCIe and 2,5GBE WAN to be developed…

1 Like

Compex has now WIFI 6 cards. They are Qualcomm reference design, M.2 interface and very wide.
Probably OpenWRT needs to add support for them too, but probably they will support AP mode.

1 Like

Link to current discussion on OpenWrt forum: https://forum.openwrt.org/t/802-11ax-wifi-ap-mpci-e-cards/63577

Edit: it seems those cards will not fit into TO natively

  • they have m.2 connections
  • they are ca. double width compared to standard mPCIe-cards

But it might fit, when using a mPCIe-to-m.2 adapter (because then the card is a little bit above the other slots), but then no other cards will fit into the neighbour slots… but even if that worked out, the data rates of 802.11ax (4,9GBit on 5/6GHz and 1,1GBit on 2,4GHz) can never be reached, as the mPCIe-slots of TO deliver 1GBit only, right?
What I am also curious about, is if the needed power could be satisfied…
I will try the 2,4GHz card in TO slot #3 with the above mentioned m.2-adapter, as soon as driver support is available for ath11k.