Hello,
I’m using a CM5 with 16GB RAM and testing a USB 2.5GbE adapter (RTL8156). The setup uses a short CAT6A cable (2m) with direct connection, link negotiates at 2.5GbE Full Duplex. However, iperf3 with multiple streams (P4–P8) consistently tops out at ~1.7–1.8 Gbit/s. Investigation shows USB is handled by the RP1 I/O controller, xhci-hcd IRQs are hard-pinned to CPU0, and IRQ affinity cannot be changed (Input/output error). RSS is unavailable and RPS provides only marginal improvement.
This doesn’t look like a cable, NIC, or tuning issue, but rather a structural limitation of RP1 + USB xHCI. Is this ~1.8 Gbit/s ceiling expected for USB-based 2.5GbE on CM5 / Raspberry Pi 5? Is a PCIe-based NIC the only reliable way to exceed 2.0 Gbit/s on this platform![:?:]( "Question")
I’m using a CM5 with 16GB RAM and testing a USB 2.5GbE adapter (RTL8156). The setup uses a short CAT6A cable (2m) with direct connection, link negotiates at 2.5GbE Full Duplex. However, iperf3 with multiple streams (P4–P8) consistently tops out at ~1.7–1.8 Gbit/s. Investigation shows USB is handled by the RP1 I/O controller, xhci-hcd IRQs are hard-pinned to CPU0, and IRQ affinity cannot be changed (Input/output error). RSS is unavailable and RPS provides only marginal improvement.
This doesn’t look like a cable, NIC, or tuning issue, but rather a structural limitation of RP1 + USB xHCI. Is this ~1.8 Gbit/s ceiling expected for USB-based 2.5GbE on CM5 / Raspberry Pi 5? Is a PCIe-based NIC the only reliable way to exceed 2.0 Gbit/s on this platform
Statistics: Posted by dogukansahil — Tue Dec 30, 2025 12:03 pm — Replies 0 — Views 30