this post was submitted on 10 Jun 2024
54 points (96.6% liked)

Selfhosted

39980 readers
768 users here now

A place to share alternatives to popular online services that can be self-hosted without giving up privacy or locking you into a service you don't control.

Rules:

  1. Be civil: we're here to support and learn from one another. Insults won't be tolerated. Flame wars are frowned upon.

  2. No spam posting.

  3. Posts have to be centered around self-hosting. There are other communities for discussing hardware or home computing. If it's not obvious why your post topic revolves around selfhosting, please include details to make it clear.

  4. Don't duplicate the full text of your blog or github here. Just post the link for folks to click.

  5. Submission headline should match the article title (don’t cherry-pick information from the title to fit your agenda).

  6. No trolling.

Resources:

Any issues on the community? Report it using the report flag.

Questions? DM the mods!

founded 1 year ago
MODERATORS
 

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

  1. 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
  2. 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.

you are viewing a single comment's thread
view the rest of the comments
[–] [email protected] 13 points 5 months ago* (last edited 5 months ago) (3 children)

Check what changes in lspci command 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.

[–] [email protected] 1 points 5 months ago (1 children)

It’s late. I’ll have to pull the card and re run tomorrow. But here’s with the GPU in:

It’s an i7-14700 and an ASRock z690 extreme. I’m actually hoping to put a second GPU in the last PCIe slot so I can let proxmox use the iGPU, pass the 3060 into a Unix moonlight gaming VM, and pass an RX590 into a hackintosh VM.

[–] [email protected] 1 points 5 months ago

I had an issue with an ASrock Tiachi where if I enabled virtualization, the network would disappear entirely. May want to check for FW updates for your board. I had nothing but issues with the shitty BIOS and even had to upgrade my CPU sooner than I wanted to do the update.

Make sure your CPU is still supported by the update.

[–] [email protected] 1 points 5 months ago (1 children)

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.

[–] [email protected] 1 points 5 months ago

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

[–] [email protected] 3 points 5 months ago (2 children)

Ok. I’ll check it out.

Let’s say it is exhausted… what will get me more bandwidth? CPU or mobo..?

Only other pci-e card in at the moment is 16 line HBA seems to be basically 2 cards sandwiched on one board)…

[–] [email protected] 0 points 5 months ago

Try disconnecting everything including the extra board.

[–] [email protected] 4 points 5 months ago* (last edited 5 months ago)

Possible something on your motherboard has PCIe lanes that are dedicated to GPU when it’s slotted, otherwise they can be used for other devices?

For example here’s a post about m.2 slots that, when used, affect the PCI on a particular board. May be worth checking your boards manual to see if there’s something similar.

https://forums.tomshardware.com/threads/questions-about-a-mb-im-looking-at-asrock-z790-pg-riptide.3787003/

The answer not only seemed a HUGE disappointment, but a bit baffling. The pdf manual says if you occupy that 5th m.2 slot, which is the Gen 5 one, the Pci-E 1 slot is automatically downgraded to 8x. This I thought would be unacceptable if running a behemoth like the RTX 4090 I eventually plan to get, as it requires a lot of power and bandwidth.