Good bang for buck non-Wifi hardware?

Yup the mini pcs are better than raspberry options. They have multiple USB 3.0 ports.
M.2 to ethernet adapters are sold in aliexpress like https://s.click.aliexpress.com/e/_oBpZQJG idk if driver is available. On these thread it says it was successful Does OpenWrtx86 support Realtek 2.5Gbps Nics? (RTL8125B) with r8169 driver.
Usb to ethernet adapter drivers are available.

While I share your gut feeling - probably because everyone heard about the sh*tshow SD cards could be in the Raspberry Pis - there is little technical difference. eMMC is the soldered flavour of MMC, on which the SD card protocol is based.

The controllers are more robust than the controllers on SD cards.

And quality is better/higher standards.

1 Like

AFAIK the eMMC has no driver on official OpenWrt so even you flash the R4S version on it, you cannot use eMMC.

When you use SquashFS, with a decent quality SD card, it's almost no wearing risk on the card, at least my R4S has been working perfectly for more than a year, and now adding a Raspberry Pi as WiFi extension to a small corner with an ancient 2GB SD card it's also working very well.

2 Likes

Same with my R4S approaching 2 years now as home gateway. OpenWrt on SquashFS writes very little to the SD card.

2 Likes

x86 mini PCs are all the rage these days but for an ARM option I can recommend the NanoPi R4S (very reliable in my home network for almost two years now) or perhaps a build around a Raspberry Pi 4 1GB ($35 for the Pi, $15 for a USB dongle, then whatever power supply/PoE hat and case you like).

I am just really turned off by x86 in the networking space for some reason. Just seems overkill, moderately expensive and inefficient compared to ARM devices and native routers/firewalls. It's nice they can be expanded and updated with better NIC's at any time, but that's where it ends for me. At the moment I've had my eyes on used Edgerouter 4's for about $80 or so.

ER-4 looks quite ok at this price, but....I doubt it can give you decent Wireguard performance.

I had one, and while I liked it a lot, Octeon isn't as well supported as one would like (both upstream and in OpenWrt). I forgot which company initially brought it to market - Marvell bought them, I believe - but they didn't bother with upstream. So not a safe bet, despite its being powerful hardware. Better get some regular better supported ARM64 stuff then.

Everyone talks about slow wireguard performace. Isn't wiregaurd performance mostly reliant on the speed of the endpoint remote (decryption) hardware?

@Borromini is the Octeon setup not fully supported?

You'll find this thread to be instructive:

x86_64 has one quite major aspect going for it, it just works, all the time, with every kernel - as it's what's literally tested by basically any kernel developer, server admin and linux user on the planet, that shows in stability and reliability. At the same time it just has the raw performance to do things without cheating (hardware flow-offloading or other mechanisms), even my 11 year old baytrail-d Atom j1900 can do 1 GBit/s wirespeed while being half asleep (yes, sqm/cake is a different topic, for that you have to step up your game a tad) - try that with almost any target (yes, cavium/ octeon qualifies (but is exotic at best), PowerPC/QorIQ isn't bad (but is exotic at best), rockchip and the RPi4+ do, mvebu does to a limited extent; mt7621a only with hardware flow-offloading).

Compared to modern routers, x86_64 doesn't necessarily need to be more power hungry (there are devices out there in the 4.5 watts to 6 watts idle range; however your old gaming rig will need (much) more than that), nor really expensive (as in used ones well below 50 bucks or brandnew ones -delivered- ranging from 130-230 bucks meeting these parameters (up to 2.5GBASE-T; 10GBASE-T costs ~80-90 bucks a piece for the ethernet cards alone, but hey, x86_64 can do it nevertheless).

2 Likes

Not necessarily. It depends what x86 you are talking about. The Intel N100 x86 is a 6W TDP CPU with future proofing performance to spare. Compare that to the much slower rk3399 ARM CPU at 7W TDP in the NanoPi R4S and x86 is not so inefficient compared to ARM devices after all.

3 Likes

Just as a small aside, the TDP does not matter that much, as it only tells you for what heat dissipation the cooling has be designed - it does not tell you much or anything about the idle power requirements of the whole system at the power socket. An Intel raptor lake system with a 65 watts TDP CPU may very well get along with ~15 watts idle power consumption, on the right mainboard and coupled with the right components.

But, yes, with well selected components you can get away with 3.8-4.5 watts at the power plug (Futro S920) or ~6 watts for an ASRock Q1900DC-ITX, around 6-7 watts for an alderlake-n/ n100 or ~15-16 watts for a desktop alderlake/ raptorlake system - and these values can compete quite well with modern ARM systems.

1 Like

Take Intel TDP with a grain of salt. They have 105w TDP CPU's that hit 300w under an AVX workload.

Sure, the rest of the hardware matters a lot, but I'm of course thinking of comparable hardware capabilities other than the CPU. And most all current CPUs will idle much lower than than their peak power that must be dissipated when running full out, granted to different degrees.

A 65W TDP raptor lake CPU is an orange to a 7W TDP rk3399 (or 6W TDP n100) apple. Nonetheless, the ratios of TDP to idle power are not that different (4.3 and 3.5) comparing the raptor lake system idling along at ~15W, to a rk3399 NanoPi R4S idling along at ~2W. Perhaps just coincidence.

my dell wyze 5070 consumes only 10w isnt that energy efficient. comeon.

My MSI cubi3 Silent consumes 6-7 watts (15W TDP cores, I believe). It runs proxmox with few containers, openwrt in a VM.

No, not when my Odroid C4 doing similar tasks consumes 1w