Driver 580 GSP firmware crash (Xid 120/154) on RTX 3070 Mobile with HDMI display — 535 works with GSP disabled
System Information
| Component | Detail |
|---|---|
| GPU | NVIDIA GeForce RTX 3070 Mobile / Max-Q (GA104M, rev a1) |
| PCI | 01:00.0 VGA compatible controller |
| Laptop | Lenovo Legion 5 Pro 16ITH6H (82JD) |
| BIOS | H1CN35WW |
| OS | Ubuntu 24.04.4 LTS (noble) |
| Kernel | 6.8.0-106-generic x86_64 |
| PRIME mode | nvidia (dedicated GPU) |
| HDMI Monitor | LG SMARTGAME+ (3840x2160, 700mm x 390mm, EDID serial ecde0c00) |
| HDMI Connector | card2-HDMI-A-2 (via NVIDIA GPU) |
| Internal Display | 2560x1600 (via i915 IGP, not active in current PRIME config) |
EDID (LG SMARTGAME+)
00ffffffffffff001e6ddc77ecde0c00
0122010380462778eaee55ac5240b024
0e5054210900d1c06140454081c00101
01010101010108e80030f2705a80b058
3a00ba892100001a6fc200a0a0a05550
30203500ba892100001a000000fd0030
901eff86000a202020202020000000fc
004c4720534d41525447414d452b032e
Driver Versions Tested
| Driver | Variant | Result |
|---|---|---|
| 580.126.09 | Proprietary (nvidia-driver-580) |
GPU lockup, Xid 120/154 |
| 580.126.09 | Open kernel (nvidia-driver-580-open) |
GPU lockup, Xid 120/154 |
| 535.288.01 | Proprietary + NVreg_EnableGpuFirmware=0 |
Works perfectly |
Description
When an HDMI monitor (LG SMARTGAME+, 4K) is connected, driver 580 triggers a GSP firmware page fault (Xid 120) during HDMI display initialization. This escalates to Xid 154
nvidia-bug-report.log.gz (1.4 MB)
(GPU Reset Required) and enters an infinite error loop that locks the GPU at maximum power draw, causing the system to freeze and the GPU to overheat.
Both the proprietary kernel module (nvidia-driver-580) and the open-kernel module (nvidia-driver-580-open) exhibit identical behavior.
Driver 535 is the last LTS branch where GSP is optional for Ampere GPUs. With NVreg_EnableGpuFirmware=0, the driver bypasses GSP and talks to hardware directly, and the HDMI display works perfectly with no errors.
Steps to Reproduce
-
Install
nvidia-driver-580(580.126.09) on Ubuntu 24.04 -
Set PRIME to
on-demandornvidia -
Connect LG SMARTGAME+ monitor via HDMI
-
Reboot
-
System freezes during display initialization; GPU overheats
-
Observed errors (from prior sessions): Xid 120 (GSP page fault), Xid 154 (GPU Reset Required)
Expected behavior: HDMI display initializes normally, as it does on driver 535 without GSP.
Workaround
Downgrade to nvidia-driver-535 and disable GSP firmware:
sudo apt install nvidia-driver-535
echo "options nvidia NVreg_EnableGpuFirmware=0" | sudo tee /etc/modprobe.d/nvidia-gsp.conf
sudo prime-select nvidia
sudo update-initramfs -u
sudo reboot
Current Working State (Driver 535, collected 2026-03-25)
$ nvidia-smi
Driver Version: 535.288.01 CUDA Version: 12.2
GPU: NVIDIA GeForce RTX 3070 Mobile, 56°C, 17W/115W, 0% utilization
Memory: 2842MiB / 8192MiB
$ xrandr --listmonitors
Monitors: 1
0: +*HDMI-0 3840/700x2160/390+0+0
$ cat /sys/class/drm/card2-HDMI-A-2/status
connected
$ sudo dmesg | grep Xid
(no output — clean, no errors)
$ cat /proc/driver/nvidia/version
NVRM version: NVIDIA UNIX x86_64 Kernel Module 535.288.01 Tue Nov 18 18:26:41 UTC 2025
Kernel boot parameters (working config)
BOOT_IMAGE=/boot/vmlinuz-6.8.0-106-generic root=/dev/mapper/vgkubuntu-root ro quiet splash nvidia.NVreg_EnableGpuFirmware=0 nvidia-drm.modeset=1
dmesg NVIDIA excerpt (clean boot on 535)
[ 3.479] nvidia: loading out-of-tree module taints kernel.
[ 3.591] nvidia-nvlink: Nvlink Core is being initialized, major device number 510
[ 3.592] nvidia 0000:01:00.0: enabling device (0006 -> 0007)
[ 3.644] NVRM: loading NVIDIA UNIX x86_64 Kernel Module 535.288.01
[ 3.657] nvidia-modeset: Loading NVIDIA Kernel Mode Setting Driver for UNIX platforms 535.288.01
[ 3.659] [drm] [nvidia-drm] [GPU ID 0x00000100] Loading driver
[ 4.501] nvidia-modeset: WARNING: GPU:0: Unable to read EDID for display device DP-4
[ 4.511] [drm] Initialized nvidia-drm 0.0.0 20160202 for 0000:01:00.0 on minor 2
[ 4.554] nvidia-uvm: Loaded the UVM driver, major device number 508.
Notes
-
Crash logs from driver 580 were not preserved. The GPU lockup and forced reboot prevented clean log capture. The Xid 120/154 errors were observed live in
dmesgbefore the system became unresponsive. -
The issue is specific to the HDMI output. The internal display (routed through the Intel iGPU) works fine on all driver versions.
-
This appears to be a regression in the GSP firmware’s HDMI initialization path for Ampere (GA104M).
-
Minor note:
nvidia-modesetwarns “Unable to read EDID for display device DP-4” even on the working 535 config, but the HDMI display (HDMI-0 / card2-HDMI-A-2) works correctly regardless. -
The
nvidia-bug-report.log.gzattached was generated on this working 535 configuration with the HDMI monitor connected and functioning.
Request
Please investigate the GSP firmware regression on Ampere (GA104M) when initializing HDMI outputs in the 580 driver branch. Ideally, provide either a fix or a way to opt out of GSP on Ampere in 580+, as was possible in 535.