Like the title says. I installed a GPU, everything posts and boots fine. The lights on the Ethernet port are lit up and will stay lit up indefinitely (I assume) if I leave it at the kernel select screen.
But as soon as I load a kernel, the lights go dark. It also is not shown as an active client on my gateway, so it’s not working at all.
I’ve tried lots of commands I’ve found to force it up. It looks to me like the NIC assigned to vmbr0 is correct. Etc. I just can’t get it to work.
If I remove the GPU, it immediately works again. NIC stays up after the kernel loads and I can access the web UI as normal.
rooteprox. *
root@prox:*# ip a
- 10: «LOOPBACK, UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000
link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00
inet 127.0.0.1/8 scope host lo
valid_lft forever preferred_lft forever
inet6 :: 1/128 scope host noprefixroute valid_lft forever preferred_lft forever - enpsso: ‹BROADCAST, MULTICAST> mtu 1500 qdisc noop state DOHN group default qlen 1000 link/ether a8:a1:59:be:f2:33 brd ff:ff:ff:ff:ff:ff
enp0s31f6: «NO-CARRIER, BROADCAST, MULTICAST, UP> mtu 1500 qdisc pfifo_fast master vmbro state DOWN group default qlen 1000 link/ether a8:a1:59:be:f2:32 brd ff:ff:ff:ff:ff:ff
vmbrO: ‹NO-CARRIER, BROADCAST, MULTICAST, UP> mtu 1500 qdisc noqueue state DOWN group default qlen 1000 link/ether a8:a1:59:be:f2:32 brd ff:ff:ff:ff:ff:ff
inet 192.168.1.3/24 scope global vmbro valid_lft forever preferred_lft forever
root@prox: *# cat /etc/network/interfaces
auto lo
iface lo inet loopback
iface enp0s31f6 inet manual
auto vmbro
iface vmbro inet static
address 192.168.1.3/24
gateway 192.168.1.1
bridge-ports enp0s31f6
bridge-stp off bridge-fd o
iface enps0 inet manual
source /etc/network/interfaces.d/*
root@prox: ~# service network restart
Failed to restart network.service: Unit network.service not found.
Check what changes in
lspcicommand between not having the GPU connected vs. Having it connected.I am suspecting that your PCI-E bandwidth is getting exhausted once the kernel activates your GPU.
Edit: Although I could be wrong about this. So makes sense to try passing "nomodeset" to your kernel parameters and see if that changes anything.
Aren't the PCIe lanes directly connected to the CPU? So the connections would be rerouted in hardware to connect to the GPU?
I am not the poster but I am curious if you know what maybe happening on a hardware level.
There's generally one or two slots connected directly to the CPU running in x16 or x8 if there's two and both are connected, 4 lanes linking the CPU to the chipset, and the rest of the slots connect to the chipset and share that same x4 link. If your cpu has 24 lanes (Ryzen do/did a few years ago, Intel might but didn't a few years ago), the remaining 4 lanes usually go to an NVMe slot