Issue occurs on Windows but not on Linux.
Issue occurs on Windows but not on Linux.
Hello everyone! I've been facing some challenges for a while now and feel really confused. My NIC was removed from my unraid box and swapped with an Intel X540 in my main system. After the upgrade, speeds are still inconsistent and nowhere near the expected 10Gbps. I updated the drivers to the latest version (2020-11-25), tried various fixes, and eventually decided to boot Linux live to test it. It worked perfectly without needing any driver installations—10Gbps is achievable instantly on Linux but only a tiny bit on Windows. TL;DR: same hardware runs at 10Gbps in Linux but barely on Windows.
The specific version of Windows wasn't noted. Some older Intel network cards don't work with newer OS releases. Generally, I've seen more consistent performance on Linux.
I'm using Windows 10 Pro right now. That specific card previously functioned well before, especially with the older server setup that had a different 1 port 10GbE branded card.
It appears Windows doesn't work well with a server that has an x540 at2. The problem started when I used the old card on my PC and the Intel card in the server, but it resolved completely once they were swapped. However, the issue reappeared when the card failed.
I've been experiencing some variant of this issue for a long time now. Near as I can tell, there is some issue either within the Windows TCP stack or the Linux TCP stack that causes weird behavior when (specifically) sending from Windows to Linux (from Linux to Windows is more YMMV; seems fine for me, but maybe issue scales different with different hardware). I'm slowly going down the rabbit hole of testing different NICs and OSes and configurations as time allows. I've seen reports like this other places in my research, but very little in the way of concrete explanations. I implemented what I thought would be the solution (falling back to VirtIO NICs into my TrueNAS Scale VM instead of hardware passthrough), but what worked in testing failed the same way in "production" (or whatever you call status quo on a homelab...). The only constant I see is that Linux-to-Linux or Windows-to-Windows works perfectly . My last resort may end up being sifting through gigabyte of WireShark logs... My only weird datapoint is that Win11 virtualized with a VirtIO NIC will hit line rate (2.5G in that case; did I mention this problem scales to other speeds too? ). On the TODO list is to install Win11 on bare metal (with various NICs) to see if the issue is actually fixed in Windows 11 (somehow) or if the VirtIO NIC is the reason why it's working. My gut tells me it's either Windows being Windows or an Intel driver bug, but after four dead Mellanox NICs, I'm running out of enterprise-y options... I should start blogging this process because this seems to be a thing.
Here’s a revised version of your update:
I attempted to install Windows 11, but I encountered issues because the latest drivers aren’t compatible. I managed to get full performance, though. Using the same outdated drivers on Windows 10 also resulted in no improvement.