I’m looking to build a NAS soon (TrueNAS SCALE/Linux) and I am once again confused by the current state of hardware and software of the Intel i225-V/i226 and the Realtek chipsets.
Have all of the Intel 2.5GbE issues been fixed in software on Linux, or is the hardware simply flawed? Are the Killer and Realtek chipsets a better choice at this point?
Not considering other options because I plan to use what the manufacturer puts on the motherboard
Killer is Intel as far as I’m aware. Their low-end consumer gear to not dilute the glorious name of Intel.
I have a board with 1x1G and 1x2.5G, both Realtek chips (the mighty DRAGON 2.5G!). Been working and sending and receiving data for a year now. I’m on a leading-edge kernel, but I’m running the chips ever since 6.0. Would buy board with Realtek again. I mainly use my 10Gbit NIC, but I had the Realtek 2.5G running the show for about a month…just worked. Never payed much attention to the 1G port, it’s running on a separate VLAN in the background. All fine.
Realtek was a problem for FreeBSD in the past (e.g. TrueNAS Core or pfSense), but never for Linux as far as I’m aware. And TrueNAS Scale should be on Debian 12 Kernel…so 6.2ish…fairly new.
There’s a reason why you don’t see Realtek in networking gear except for consumer grade (if you include consumer routers and low-end switches). You’re much better off with Intel, Broadcom etc in general as long as it’s not very old. New “Killer”-products is now rebranded Intel-parts with some additional software components, some wifi stuff is older Qualcomm/Atheros if I recall correctly. Just because it’s “seems to work” doesn’t mean that it’s good.
TrueNAS Scale is on LTS kernels, 6.1.something or so these days - relatively modern.
I don’t think it matters hugely which chipset you go with for home use, with respect to “reliability”. You’re almost as likely to have issues with SATA drivers or power supply instability.
I think for home use it boils down to cost of a restart, you can have meh hardware, reboot once a year because something’s stuck, fine.
You can get lucky with diy hardware and none of a gazillion things that can go wrong goes horribly wrong and you have amazing uptime and reliability >1y (how do you update? dunno)., or you can spend 3x as much on corporate/enterprise hardware… to get that last few nines.
Or … you can do Ceph … which doesn’t really scale down as efficiently or easily (thinking of those beautifully mad people using it on mini PCs with drives in USB enclosures, like those early ZFS videos with a wad a of USB flash drive dongles).
I think if you’re building your own, you should expect once in a while problem every 6 months, and if it goes longer than that, enjoy the nice surprise, and I think realtek/intel/broadcom/marvell are equally ok for that.
I´m not fully sure ¨if¨ intel already fixed all the problems with their nic´s.
but i still read about connection drop issues in regards to both the intel i-225V and the I-226V.
Killer is also using the same chips as well but than with some additional garbage software,
for ¨better¨ gaming experience with less lag which makes zero sense.
Don’t forget the software part (quality of driver)
I’m sure the enterprise market would be interested in better or more consistent latency. This advantage is unique to gaming market? That sounds a lot like Intel NIC business 
Yeah…marketing department at its best.
I assume they just use less powersaving as a default to avoid latencies in changing power states. If you want lower latencies, don’t use Wifi and get SFP+ cards or go Infiniband. That would be some “killer” consumer card.
Am I the only one not caring about 2.5G at all? We had 1G for 20 years or so…and now 2.5x the bandwidth is the saviour? It’s an unsatisfying bridge-technology until everyone gets 10Gbit on-board. Destined to be forgotten by history.
Well yeah i agree that it would be nice if they implement 10Gbe controllers,
to more main stream motherboards with the upcoming generations.