@quad and even as cards they get cheap.

Intel replaced their #i210 / #i219 with #i225 and #Realtek released their #RTL8125

@k4m1 @stman yeah, according to the #RTL8139 #datasheet this is basically a very cheap 10/100M NIC designed #embedded systems and low-end/low-cost desktops, and for a device designed and sold in 2006 it made sense, given back then #Gigabit-#Ethernet and Cat.5 cabling was considered high-end.

  • And unlike contemporary / successor chips by #Intel like the famous #i210 (which is still offered as #i219 but mostly succeeded by the #i225 as a 2,5GBase-T version) is way cheaper, which pre-#RoHS - NICs being sold for like € 10 retail & brand-new....

The few issues known only affect like #Virtualization setups, a market this thing was never designed for (most likely also never tested against).

  • I'd not he surprised if a lot of cheap #ThinClients and other systems used these NICs because of the simplicity of integration, being a cheap 3,3V single-chip (+auxilliary electronics) solution and propably costling less than 10¢ on a reel of 10.000.

It's the reason why to this day we see #Realtek NICs being shipped instead of fanning-out & enabling #SoC-integrated NICs with a #MAC & #PHY instead: Because the auxilliary parts for those are more expensive than just getting a PCI(e lane) somewhere and plonking it down.

  • Maybe there have even been some really cheap, low-end #Routers / #Firewalls aiming at #SoHo customers back in those days, cuz back then 16MBit/s #ADSL2 was considered fast, and Realtek's NICs up until recently only delivered like 60-75% of the max. speed advertised, so by the time someone would notice, that gearvwould've been EoL'd anyway and those who did notice right-away never were the target audience to begin with.

Most modern NICs are more complex and demand more configuration / driver support...

If anyone out there is having issues on a #homelab #Proxmox machines with #Intel NIC's... In my case on a couple of #Lenovo M710q's with #I219-V, an Intel NUC10i3FNB with the same, and a #Protectli VP2420 with 4 * #I225-V. The solution for me so far (and I've not seen any performance impact though I've not _really_ looked hard so don't trust this for production) is popping the below after your 'iface eth0 inet manual' line. In my case of course I'm hanging everything else off a bridge onto that interface (vlans and what have you), or passing the PCI-E device to a VM, so adapt as you need. This came from a site somewhere but I cannot find the source any more.

post-up /usr/bin/logger -p debug -t ifup "Disabling segmentation offload for ${IFACE}" && /sbin/ethtool -K $IFACE tso off gso off && /usr/bin/logger -p debug -t ifup "Disabled offload for ${IFACE}"