Before anyone yells at me, yes I did a search. While there is some good info to be had searching the forums, a lot the 10gb info is outdated and even more of it applies to only 1gb nics. So I thought I would see what everyone is using for a 10gb rj45 network cards for use on x86_64? My entire network is copper so ideally I'd like to stick to rj45 but I could entertain sfp+ if it came with the rj45 transceiver modules.
My requirements:
Reasonably cheap used card from ebay, aliexpress or similar site.
Must work on openwrt using x86_64
Need 2+ rj45 ports. (if it's a fiber nic then I need cheap rj45 adapters too)
Must fit pcie gen 3 full height slot. (if it's a half height nic I then it needs to come with full height adapter)
10G has only recently fixed the crazy heat/power consumption issue. The new 10G can go as low as 1.65w/port.
Realtek
8126 - 5G
8127 - 10G
Confirmed ASPM and linux kernel support
$40/port and lower
Unknown if dual and quad NIC's exist yet; likely 2026 product
Intel
e610 - 10G Massive warning: unknown whether or not this is be as terrible as Intel 2.5G i.e. forever broken ASPM
$120-$140/port ( )
For sale now.
I'm personally waiting for 8127. Kind of a bad time to go 10G with the newer stuff coming out. If you don't care about heat/power consumption/ASPM, get the old stuff.
Thanks. I may not have a choice. My motherboard only has 1 pcie x16 slot. The rest are pcie x1's I guess I have some more searching to do. I need to figure out just how many watts current gen nics consume to see if it's worth waiting for newer energy efficient nics or just get something now. My x86 router consumes a little less then 20w at idle as it stands right now using 2 1gb realtek nics. I built this thing with the idea that I would upgrade later to 10gb. That later is now LoL.
4.8w/port + ASPM will likely be broken i.e. your CPU power consumption will go up at idle. You can be stuck at C2 on some. PEG/x16 slots are attached to CPU lanes = bad ASPM on certain motherboards. Slots connected to chipset lanes is what you want.
Realtek 2.5G 8125 exists today and works great as a stop gap. I expect late 2Q2026 to have $50 dual 8127 nics available.
Thank you all for the replies. I did some cost calculations and because electricity is pretty cheap where I live, It would take about 4 years to pay for a more expensive but lower power draw card compared to the current crop of absolutely dirt cheap but power hungry cards. So I'm going to go with a current model card. The ROI just isn't there for me to hold out for better card. In 4 years I'll probably upgrade my network to something better anyways, like 40gbit or whatever is trendy at that time.
LoL I remember when I first got fast ethernet at 100mbps. I thought it was so fast I couldn't possible ever need anything faster. Now I gripe about gigabit.
It does support 10G on copper, but not multigig? Yes, I know such equipment existed once, but still I still have to ask.... Sounds like something you can use to heat your house.
If you don't care at all about power consumption and plan to go to 40Gb very shortly, go straight to 25Gb or even 100Gb. Your bandwidth requirements are going up 10Gb/year. I don't even know why you are looking at copper.
Then how would you run 2 x 10G with just PCI-E 3.0 x1? Even with single port there is not enough bandwidth. And for the super old Broadcom card you listed it's even worse.
Avoid SFP+ to RJ45 copper as much as you can, only do it when it's unavoidable, you can definitely cook with the heat coming from that module (mine is always 90C)
10G internet is already way beyond "steep" for all residential users. 25G fiber internet will barely be deployed anywhere in 4 years, let alone 40G+. My point is the OP can deploy 25G using the same old, hot enterprise cards being sold on ebay today and skip their already planned 4 year upgrade beyond copper.
Question: your 10/10, is that really "10", aka ~9.4 to 9.5 Gbps maximal throughput, or is this nominal 10, with throughput capped around 8 to 8.2 Gbps? (Which still is a large number, for sure, and nice to have, but I see an somewhat alarming trend in the ISP industry of using "numbers" in a surprising way where 10 means 8 and similar... BEREC for all its flaws recommends that ISPs specify customer visible rates as net payload rates, aka what typical capacity tests report, and not aas gross rates on some ill defined layer.)
Yeah, but that is not how people use their internet access or why they book high capacities... typically, I assume as I have no reliable data, people are not all that interested in transferring huge amounts of data within a months. Rather they want fast transfers instantaneously when they need/want to move data, so they are fine with paying for X Gbps even if their actual duty cycle is rather low...
Ah, thanks. I guess I am with you, even a bit more extrem, I could have switched from 100/40 to 1000/50 for 5 Euro more, put similarly saw no justification for that (especially as it would have meant switching from full dual stack on vdsl2 to ds-lite on docsis, and I am not a docsis fan).
Beyond a certain point (imho around >100 MBit/s downstream), the upstream ratio becomes at least as (if not more) important. Here many ftth offers can score, as they're often 2:1 (e.g. 400/200) or better. It is a game changer if you can just upload 1-2 GB within less than 5 minutes, while xDSL or cable are usually quite weak in the uplink direction.
Well, my 100/40 (actual sync typically around 116.8/46.7) is closeish to 2:1 already...
It is just I rarely upload laarge data nowadays, as I mostly switched to using remote desktop close to my data... (and I am not running off site updates via the internet). But I accept your point and also do not want to claim, that my decision is the only acceptable one, just one that works for me right now, and I really do not like docsis
)
