RmInitAdapter failed! [MSI GeForce RTX 3090] [Driver 535.230.02]

Graphics / Linux Linux
, ,
1

Hello everyone,

Trying to install a rather old GPU to a new PC (solely to take advantage of the CUDA toolkit). These are the hardware specs of my system and the software specs currently running, for which the attached nvidia-bug-report.log.gz was generated:

  • Motherboard Model: Gigabyte Β650 GAMING Χ ΑΧ V2
  • GPU Model: MSI GeForce RTX 3090 24GB GDDR6X Suprim X (GA102)
  • Driver Version: 535.230.02
  • Kernel Version: 5.15.0-140-generic
  • OS: Ubuntu 22.04 LTS Server

However, although nvidia driver installation seems to proceed as expected, nvidia-smi command keeps returning No devices were found

Drivers I have tried so far in both Ubuntu Server 24.04 and 22.04:

  • nvidia-driver-570
  • nvidia-driver-570-server
  • nvidia-driver-570-server-open
  • nvidia-driver-560
  • nvidia-driver-550
  • nvidia-driver-535

My BIOS settings are:

  • Updated to the latest version
  • Secure Boot: Disabled
  • Above 4G Decoding: Enabled
  • Re-Size BAR Support: Enabled

I tried the solutions suggested in the following posts (regarding blacklisting nouveau driver and enabling config of unsupported GPUs for open-source drivers), which allow for Ubuntu to boot normally but the problem persists:

Τhe driver is correctly installed. The card is detected on the PCIe bus (Gen 4). Nevertheless, the system reports:

  • nvidia-smi: No devices were found
  • nvidia-debugdump: Found 0 NVIDIA devices
  • Kernel logs: nvidia-drm fails to allocate the KMS interface
  • Device node /dev/nvidia0 is present, indicating partial driver initialization
  • GPU appears in lspci and lshw with the correct vendor ID

Verified Working Components

Component Status
PCIe Detection Detected at 01:00.0
NVIDIA Driver Installed Version 535.230.02, via DKMS
Kernel Headers Match installed kernel
/dev/nvidia* nodes Present
nvidia-persistenced Running normally
lshw Driver Bind Shows driver=nvidia

Failing Components

Component Status Detail
nvidia-smi Reports “No devices were found”
nvidia-debugdump --list Reports 0 devices
KMS Device Init Fails to allocate NvKmsKapiDevice
GPU Compute/Display Engines Not responsive to driver

Key Logs from /var/log/syslog

  • NVRM: kgspExtractVbiosFromRom_TU102: did not find valid rom signature
  • NVRM: kgspInitRm_IMPL: failed to extract VBIOS images from ROM: 0x25
  • NVRM: GPU 0000:01:00.0: RmInitAdapter failed! (0x62:0x25:1859)
  • NVRM: GPU 0000:01:00.0: rm_init_adapter failed, device minor number 0

Key Logs from dmesg

  • [drm:nv_drm_load [nvidia_drm]] ERROR Failed to allocate NvKmsKapiDevice
  • [drm:nv_drm_probe_devices [nvidia_drm]] ERROR Failed to register device
  • nvidia-debugdump --list → Found 0 NVIDIA devices

These logs seem to indicate that driver binding occurs, but the GPU’s internal engines (compute/display) are not responding.

Moreover, I installed Windows 10 OS on this system and tried to install the suggested Nvidia driver through the Nvidia App but still the driver could not communicate with the GPU.

I would appreciate any help to solve this issue! You can find attached my bug report and the output of several diagnostic commands as well.

nvidia_diagnostics_20250522_103045.log (6.1 KB)
nvidia-bug-report.log.gz (140.8 KB)

2

I had somewhat a similar problem on an AMD laptop:

Does your CPU happen to include some iGPU?

3

Hi, yes the CPU is an AMD Ryzen 9 7900Χ with integrated graphics.

Did you find a solution to your issue?

4

Unfortunately not :(
Also I don’t have access to this laptop anymore: I was planning to purchase this model and luckily a friend of a friend allowed me to test his before I made my decision.

Later my other friend speculated that it could be an IRQ conflict: could you please check the output of the below commands:

cat /proc/interrupts | grep -i "gpu"  ## for AMD
cat /proc/interrupts | grep -i "nvidia"

Here are posts from yet another user experiencing problems with AMD CPU/iGPU + Nvidia GPU + Linux, although they managed to get it start at least, but it was unstable:

6

I’ve just received an info from another user of the same Tuxedo laptop model that it works for him with his GTX-1080ti eGPU with driver 575.51.03, kernel 6.11 (Tuxedo OS based on Ubuntu-24.04). Hard to say if it’s because of the newer driver (hopefully) or the different GPU model.
Anyway, you may want to give a try to a similar version mix.

7

I had switched off my iGPU in the UEFI right from the start and only used the RTX 4090 Rog Strix but still again and again had screen freeze problems.

1 Like