Intel I226-V vs I225-V NIC Differences: Power and Errata

The Intel I226-V vs I225-V NIC differences come down to four things Intel documents and one it does not. Documented: the I226 draws less power in every link state, adds 802.3az Energy Efficient Ethernet, holds 2.5G on longer cables, and ships without the inter-packet-gap erratum that made first-run I225 silicon fall back to 1 Gbps or worse. Not on any spec page: the I226 carries its own PCIe ASPM L1.2 erratum, which is why a 2022 controller still generates link-drop threads in 2026. Everything else, including the PCIe interface, queue count, package and FreeBSD driver, is identical. Buy the I226-V for a new appliance; do not rip a working I225-V out to get one.

Spec table: what Intel actually changed

Figures are from Intel ARK for the I226-V and the I225-V, and from the I225/I226 datasheet, revision 2.6. Datasheet power rows are Intel’s reference-card measurements at 3.3 V, 3 m cable, 25 °C.

	I225-V	I226-V
Launch (ARK)	Q4 2019	Q2 2022
TDP (ARK)	1.95 W	1.3 W
2.5G active (datasheet)	1,214 mW	910 mW
2.5G idle (datasheet)	885 mW	762 mW, or 600 mW with EEE
Cable disconnected	250 mW	17 mW
Host interface	PCIe 3.1 spec, Gen2 x1 at 5.0 GT/s	Same
Speeds	10/100/1000/2500BASE-T	Same
802.3az EEE	No	Yes
Queues	4 Tx / 4 Rx, 5 MSI-X vectors	Same
Jumbo frames	9.5 KB with TSN off	Same
PXE	Legacy and EFI	EFI only
Package	7 x 7 mm VQFN-56	Same
Operating temperature (-V SKU)	0 to 70 °C	0 to 70 °C
Expected discontinuance (ARK)	1H 2027	1H 2033
FreeBSD driver	igc(4)	igc(4)

Two rows need a note. Both parts are one PCIe Gen2 lane at 5.0 GT/s, enough for a 2.5 Gbps line rate with headroom; the “3.1” is the spec revision the controller complies with, not its link speed. And TDP is ARK’s headline number; the datasheet’s worst case for the I225 at 100 m and 55 °C is 2,200 mW. Across four ports, ARK’s figures work out to 7.8 W versus 5.2 W, which ServeTheHome summarises as “1.3W TDP (x4 = ~5W)” for the I226 unit.

Where the I226-V wins

Power. The datasheet lists “dramatically reduced power consumption in all silicon states (20-50% reduction)” as the first I226 highlight. Cable disconnected is the extreme case, 250 mW down to 17 mW, and it matters on a four-port firewall where two ports usually sit empty. The 2.5G idle saving, 885 mW to 762 mW, is what an always-on WAN port sees.

EEE, on the I226 only. The datasheet puts it in one sentence: “I225 doesn’t support EEE, but I226 does support EEE.” The catch is that igc(4) ships with hw.igc.eee_setting defaulting to 1, meaning disabled, so an OPNsense or pfSense box never sees the 600 mW idle figure unless you enable EEE, and enabling EEE is the first thing the link-drop checklist tells you to undo. Treat it as a Linux desktop feature, not a firewall feature.

Cable tolerance. Intel lists “improved cable BER performance, especially on longer cables” as an I226 highlight, and the I226-IT holds 2.5G across the full -40 to 85 °C industrial range where the I225-IT stopped at 70 °C. For a -V part the BER point is the relevant one: 2.5GBASE-T over marginal Cat 5e is exactly where I225 links renegotiate down to 1 Gbps.

The inter-packet-gap erratum is gone. Intel’s I225 specification update is precise: first-version I225 silicon needs a minimum receive IPG of 8 bytes, while IEEE 802.3 requires the MAC to tolerate 5 bytes at 2.5 Gbps. Against a link partner that sends tight gaps, throughput “can be as low as 1-10Mbps”. The workarounds were forcing 1 Gbps, or driver 1.0.0.30 with NVM 1.38, which detect the drops and step the link down automatically. Intel fixed it in I225 v2 silicon, which Puget Systems dates to Q3 2020 and appliance sellers label “B3”. Every I226 postdates the fix.

Support horizon. ARK lists expected discontinuance of 1H 2027 for the I225-V and 1H 2033 for the I226-V. If the appliance has to run six years, that row settles it.

Where the I226-V does not win

Throughput. Same MAC, same PCIe lane, same queues. ServeTheHome’s I226-V firewall review landed “within a 1% test variation from the i225 V4 version” on the same CPU. Nobody should buy an I226 for speed.

ASPM L1.2. This is the I226’s own erratum and the reason “i226-V link drops” is a search term. The Linux igc commit that works around it says the part advertises PCIe L1.2 support but its exit latency from that state is longer than the packet buffer can absorb under load, so inbound traffic stalls. The FreeBSD port of the same fix adds the detail firewall builders need: “Disabling ASPM system-wide (BIOS or OS ASPM policy) does not fix it. The L1.2 enable bit must be cleared directly in the device’s own PCIe L1 PM extended capability.” A BIOS toggle is not a guaranteed cure; a driver new enough to carry the workaround is. That change cites FreeBSD Bugzilla 279245, “igc(4) I226 (and I225) TX hangups”, so the I225 is not immune to the symptom, only to this cause.

Errata shared by both. A July 2024 igc patch cites the “Ethernet Controller I225/I226 Specification Update Rev 2.1”, erratum 9, “TSN: Packet Transmission Might Cross the Qbv Window”, and shrinks the I226 MAC retry buffer from 192 to 88 bytes. A February 2026 patch proposes PCIe link recovery for I225 and I226 after ASPM L0s transitions return 0xFFFFFFFF on register reads; the PCI maintainer’s reply was that it looks like a hardware erratum Intel should document, and that disabling L0s outright is cleaner. Same advice as always: leave PCIe power management off for the NIC.

PXE. The I226 boots EFI only. Irrelevant for OPNsense, occasionally relevant for a lab that netboots old images.

Driver support: the version boundaries

Both parts use igc(4), which the FreeBSD manual says “first appeared in FreeBSD 13.1” and lists as covering I220-V, I221-V, I225-LM, I225-LMvP, I225-V, I225-IT, I225-K, I226-LM, I226-LMvP, I226-V, I226-IT and I226-K. It runs 1000 and 2500 Mbps at full duplex only and defaults hw.igc.max_interrupt_rate to 8000.

The boundaries that caught people, from Netgate’s version table: pfSense CE 2.6.0 and Plus 22.01 (February 2022) are FreeBSD 12.3-STABLE builds, and ServeTheHome’s 2022 advice was that until 2.7 shipped, “the i225-V’s are likely the better option compared to the i226-V as the i225 NICs are supported by the current pfSense 2.6”. CE 2.7.0 (June 2023) moved to a FreeBSD 14-CURRENT base and CE 2.8.0 (May 2025) to 15-CURRENT. OPNsense 22.1 (January 2022) moved to FreeBSD 13-STABLE, and its 22.1.9 changelog already carries an igc fix. Anything current on either platform is fine; the exception is an install image someone burned in 2022.

What you actually need

• New N100 or N150 appliance: I226-V, no discussion. Lower draw, longer support, no IPG history. The N100 build guide covers the rest of the spec and firewallcompare’s mini PC list covers which boxes ship it.
• Existing I225-V box that holds 2.5G: keep it. On v2/B3 silicon there is nothing to fix; on v1 with a switch that triggers the IPG bug, force 1 Gbps or get the NVM 1.38 update from the board vendor.
• Flapping I226-V: not a reason to swap to I225. Update the OS so the L1.2 workaround is in the driver, then work the link-drop checklist.
• Industrial or TSN: I226-IT, the only SKU rated for 2.5G at 85 °C. Both -V parts are client controllers, which is the right size for a 1 to 4 VLAN firewall; pfSense’s hardware requirements cover the CPU and RAM side of that sizing.

Things to test before you call it done

On OPNsense or pfSense:

pciconf -lv | grep -A3 igc
ifconfig igc0 | grep -E 'media|status'
sysctl hw.igc

pciconf gives the device ID (the FreeBSD workaround keys on 8086:125c for the I226-V) and revision. The media line should read 2500Base-T <full-duplex>, and sysctl hw.igc should show eee_setting: 1.

On Linux:

lspci -nn | grep -i ethernet
sudo lspci -vv -s <bus:dev.fn> | grep -E 'ASPM|L1Sub'
sudo ethtool --show-eee <iface>

lspci -vv shows whether L1.2 is enabled on the device itself; a kernel carrying the igc L1.2 commit shows it disabled on an I226 regardless of BIOS. ethtool --show-eee should report no EEE support on an I225 and support on an I226.

Then load it:

iperf3 -c <lan-host> -P 4 -t 60
mtr -n -c 200 <lan-host>

A healthy link holds its rate for the full 60 s with zero loss in mtr. A link that starts at 2.5G and shows a lower speed in ifconfig after the run is the IPG bug on I225 v1 silicon, or an L1.2 stall on an I226 with a driver too old to carry the fix.

For the complete appliance decision, use the N100 throughput and power calculator to explore how NIC choice and enabled services affect its estimated workload envelope.