The $100 price tag is much less important to me than long term production runs. I hope they take a page from the pcengines playbook, and arrange to build this with minimal changes over a 5-10 year period.
For instance, I can imagine them holding all specs constant except that they ship a new wifi radio every 2-3 years.
This would allow other open source projects to target it and be rock solid. For instance, my pc engines openbsd router is something like 8 years old and has never crashed.
I’d happily pay a $50-100 premium to get a brand new board that’s fast enough, but that had all the bugs worked out the last few years.
It looks like at least one of the radios is on the same SoC as the CPU, Ethernet, etc., so swapping out just the radio every few years isn't quite feasible. Such highly-integrated WiFi SoCs are pretty standard for this price range, and getting a processor with enough PCIe and Ethernet to use only external MAC+PHY solutions is a lot more expensive.
The MT7976C they're including is a 3x3 WiFi 6 chip, and you can buy them as a standalone module for ~ $25. I think the idea is that it's providing the access point radio.
That's the part I'd suggest swapping out every few years. That might mean people are stuck with older bluetooth or a low performance WiFi radio or whatever on the SoC, but that won't impact performance much. (In the worst case, they could just disable the SoC radios to prevent RF interference with the "real" radio.)
I think this is intended to offer simultaneous dual-band (2.4GHz and 5GHz) using two separate radios, as all decent APs for over a decade have done. The SoC provides one of the two radios, and the second is connected over PCIe. Whether your 2.4GHz or 5GHz radio more badly needs an upgrade in a few years can be hard to predict.
I'm reasonably sure this line of chips is designed to be the only active radio in an access point:
> Filogic 830 packs a wide variety of features into a compact, ultra-low power 12nm SoC, allowing customers to design differentiated solutions for routers, access points and mesh systems. The SoC integrates four Arm Cortex-A53 processors operating at up to 2GHz per core for up to +18,000 DMIPs processing power, dual 4x4 Wi-Fi 6/6E for up to 6Gbps connectivity, two 2.5 Gigabit Ethernet interfaces and a host of peripheral interfaces. Filogic 830’s built-in hardware acceleration engines for Wi-Fi offloading and networking enable faster and more reliable connectivity. In addition, the chipset also supports MediaTek FastPath™ technology for low latency applications such as gaming and AR/VR.
My reading of that is that the ARMs are there for software defined networking tasks, and not to run the router OS. I could be wrong. However, the OpenWRT spec lists this chip as a network adapter, not as the main SoC.
I think the "Filogic 820" branding may actually apply to the combination of the MT7981B SoC and the MT7976C NIC. https://wikidevi.wi-cat.ru/MediaTek seems to have actual detailed specs for these chips. It looks like the MT7981B is what has the ARM Cortex-A53 cores to run Linux, Ethernet, PCIe, and a 2x3:2 radio. The MT7976C is just a NIC with no application processor cores, but it has both a 2.4GHz and 5/6GHz radio. A NIC with two separate radios seems to be a relatively recent development, probably spurred by increased demand for tri-band routers after the opening of the 6GHz band.
For instance, I can imagine them holding all specs constant except that they ship a new wifi radio every 2-3 years.
This would allow other open source projects to target it and be rock solid. For instance, my pc engines openbsd router is something like 8 years old and has never crashed.
I’d happily pay a $50-100 premium to get a brand new board that’s fast enough, but that had all the bugs worked out the last few years.