Home upgrade from 2.5Gb/s to 10Gb/s using MikroTik switches, DAC cables, and SFP+ cards across existing structured cabling, with real iperf3 numbers and thermal monitoring.
Key Takeaways
Existing CAT-6 or CAT-6A runs (under 55m) handled 10GBASE-T; iperf3 confirmed near-line-rate speeds between a Ubiquiti USB dongle and an Asus XG-C100F PCIe card.
USB 10G dongles hit a single-core CPU bottleneck on receive; laptop (ThinkPad) maxed at ~7Gb/s inbound due to ksoftirqd saturation.
10GBASE-T SFP+ modules run very hot; the MikroTik CRS304-4XG-IN in an unventilated sideboard exceeded 70C internal ambient, flagged as a watch item.
Full topology: Protectli VP2440 router (dual SFP+) to MikroTik CRS304 switches at patch panel and living room, then DAC to MikroTik CRS305 in study, then DAC to workstation.
Grafana plus Telegraf over SNMP on MikroTik switches enabled per-module thermal dashboards; SFP+ temps tracked separately from CPU.
Hacker News Comment Review
Key practical split: old-gen 10GBASE-T SFP+ modules (~3W, 30m-rated) cause link flaps from heat; newer 80m/100m-rated modules draw ~1.5W and run significantly cooler – choosing the right module generation matters more than switch brand.
Commenters with software routers (Protectli-style) cautioned that raw throughput looks fine but new-connection latency, connections-per-second, and QoS suffer without hardware offload, which most consumer gigabit gear quietly provides.
Several commenters ran 10G over unknown legacy cabling with no issues, suggesting real-world CAT-5E/6 runs under 20-30m often work despite spec limits; cross-talk is the actual failure mode, not raw attenuation.
Notable Comments
@xxpor: Two SFP+ module generations exist; newer 80m/100m-rated modules at ~1.5W run “much, much cooler” vs old 30m modules at ~3W that cause link flaps.
@zamadatix: Protectli-style software routers pass bandwidth tests but degrade on new-connection latency and connections-per-second vs hardware-offload devices.
@xenadu02: Crosstalk, not wire attenuation, limits 10GBASE-T on older cabling; “dirty paper precoding” degrades under interference from adjacent pairs.
