F5F Stay Refreshed Hardware Desktop 3090 retains the phrase "falling off the bus" (XID 79) in its Linux environment.

3090 retains the phrase "falling off the bus" (XID 79) in its Linux environment.

3090 retains the phrase "falling off the bus" (XID 79) in its Linux environment.

M
MCmasterKIDz
Junior Member
44
11-11-2021, 10:55 AM
#1
I have a 3090 connected straight to an ITX motherboard running a 12600K. Operating it in Debian 13 with driver version 550, I'm using it for LLM tasks via llama.cpp. Within hours to a few days after startup, the GPU would halt briefly after each inference and then ramp up fan speed to maximum. System checks reveal the GPU isn't responding to nvidia-smi and displays XID 79. Restarting resolves the issue until it reappears later. Below are the dmesg and lscpci logs following this incident. [144421.064068] port 2 entered disabled state [144421.444945] port 1 entered blocking state [144421.444950] port 1 entered disabled state [144421.444957] veth476fba1: allmulticast mode [144421.444988] veth476fba1: promiscuous mode [144421.462855] eth0 changed to colour frame buffer [149230.182897] i915drmfb frame buffer [151819.691033] pcieport 0000:00:01.0: AER: Multiple correctable error from 0000:00:01 [151819.740572] pcieport 0000:00:01: device [8086:460d] error status [151819.740574] GPU at PCI:0000:01:00: GPU-a84a53a0-b9cb-7360-3b9e-94fd12d74003 has dropped off the bus. [151820.097175] NVRM: GPU has fallen off the bus. [151820.097180] snd_hda_codec_hdmi hdaudioC1D0: HDMI invalid ELD buffer size -1 [151853.328759] GPU-a84a53a0-b9cb-7360-3b9e-94fd12d74003 is not responding.
M
MCmasterKIDz
11-11-2021, 10:55 AM #1

I have a 3090 connected straight to an ITX motherboard running a 12600K. Operating it in Debian 13 with driver version 550, I'm using it for LLM tasks via llama.cpp. Within hours to a few days after startup, the GPU would halt briefly after each inference and then ramp up fan speed to maximum. System checks reveal the GPU isn't responding to nvidia-smi and displays XID 79. Restarting resolves the issue until it reappears later. Below are the dmesg and lscpci logs following this incident. [144421.064068] port 2 entered disabled state [144421.444945] port 1 entered blocking state [144421.444950] port 1 entered disabled state [144421.444957] veth476fba1: allmulticast mode [144421.444988] veth476fba1: promiscuous mode [144421.462855] eth0 changed to colour frame buffer [149230.182897] i915drmfb frame buffer [151819.691033] pcieport 0000:00:01.0: AER: Multiple correctable error from 0000:00:01 [151819.740572] pcieport 0000:00:01: device [8086:460d] error status [151819.740574] GPU at PCI:0000:01:00: GPU-a84a53a0-b9cb-7360-3b9e-94fd12d74003 has dropped off the bus. [151820.097175] NVRM: GPU has fallen off the bus. [151820.097180] snd_hda_codec_hdmi hdaudioC1D0: HDMI invalid ELD buffer size -1 [151853.328759] GPU-a84a53a0-b9cb-7360-3b9e-94fd12d74003 is not responding.

P
Patrions_
Member
216
11-21-2021, 06:30 AM
#2
I encountered a comparable problem with an older driver version. Since you're using Debian on the 550 branch, it might be beneficial to move to NVIDIA's APT repository and install the 590 branch: https://wiki.debian.org/NvidiaGraphicsDr...ata-center
P
Patrions_
11-21-2021, 06:30 AM #2

I encountered a comparable problem with an older driver version. Since you're using Debian on the 550 branch, it might be beneficial to move to NVIDIA's APT repository and install the 590 branch: https://wiki.debian.org/NvidiaGraphicsDr...ata-center

M
Momootje
Junior Member
18
11-22-2021, 04:16 PM
#3
These signs could also point to a GPU malfunction; please confirm if it's due to wear and tear.
M
Momootje
11-22-2021, 04:16 PM #3

These signs could also point to a GPU malfunction; please confirm if it's due to wear and tear.

R
Rodcraft99_PvP
Member
71
11-23-2021, 06:34 PM
#4
It appears there might be some GPU hardware wear, but updating drivers didn’t resolve the issue. Trying the 590-open and 575 proprietary versions didn’t help, though setting the GPU to 1500 has provided some improvement.
R
Rodcraft99_PvP
11-23-2021, 06:34 PM #4

It appears there might be some GPU hardware wear, but updating drivers didn’t resolve the issue. Trying the 590-open and 575 proprietary versions didn’t help, though setting the GPU to 1500 has provided some improvement.

S
shelbywood99
Member
172
11-29-2021, 09:45 PM
#5
Are the rear VRAM chips in 3090s causing overheating? Try directing a fan toward the backplate to observe any improvement.
S
shelbywood99
11-29-2021, 09:45 PM #5

Are the rear VRAM chips in 3090s causing overheating? Try directing a fan toward the backplate to observe any improvement.

D
dianarose32129
Senior Member
570
11-30-2021, 02:01 AM
#6
They added heat sinks to the backplate and swapped out the VRAM thermal pads. Since my tasks aren’t always heavy, I don’t think overheating will be a problem, especially since the heating occurs rapidly even after being idle for a while.
D
dianarose32129
11-30-2021, 02:01 AM #6

They added heat sinks to the backplate and swapped out the VRAM thermal pads. Since my tasks aren’t always heavy, I don’t think overheating will be a problem, especially since the heating occurs rapidly even after being idle for a while.

T
timoty_HD
Junior Member
3
11-30-2021, 06:33 AM
#7
The configuration `nvidia-smi -lgc 800,1600` has provided stable performance over the past four days. It seems there might be some hardware wear, but overall it remains functional for your needs.
T
timoty_HD
11-30-2021, 06:33 AM #7

The configuration `nvidia-smi -lgc 800,1600` has provided stable performance over the past four days. It seems there might be some hardware wear, but overall it remains functional for your needs.