dwbufhph6
They would have been better off using a Marvell AQC113-B1-I (6-Speed 10G/5G/2.5G/1G/100M/10M, Industrial Temperature Grade −40°C to 85°C, Unit Price: $32 for 32+ and less for more, lower CPU overhead than RTL8127) for the new revision, which works perfectly in many of my machines, including Linux and the external TB4 10Gbe Adapter. The Sonnet McFiver card (2xM2 PCIe x4, 2xUSB 10G, 1×10G Ethernet on a PCIE x8 card) also uses Marvell AQC113-B1-I, no problems there either, i use two of them in Linux Servers. That Chip supports Gen 4 ×1, Gen 3 ×4, Gen 3 ×2, or Gen 3 ×1, Gen 2 ×2, so dual 10G would be possible with only PCIe Gen 4 ×2 and two chips. Unless of course the source of those errors are somewhere else on the board, like Voltage Fluctuations (40-core iGPU and the NICs share the same 3.3V power rail), PCIe Signal Interference (Heavy GPU or NVMe workloads leading to PCIe handshake timeouts or physical design flaw in the motherboard layout or trace integrity, leading to PCIe handshake timeouts) etc. Then using other controllers won’t fix the problem either.
