Docs TroubleshootingGPU Not Detected or nvidia-smi Fails

GPU Not Detected or nvidia-smi Fails

Troubleshoot issues where the GPU is not detected, nvidia-smi fails, or CUDA is unavailable.

intermediate

7 min read

LightYear Team

Updated April 24, 2026

gpunvidiacudatroubleshooting

Ready to get started?

Symptoms

nvidia-smi: command not found
NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver
torch.cuda.is_available() returns False
CUDA version mismatch errors

Step 1 — Check If GPU Is Present

>_BASH

$# Check PCI devices
$lspci | grep -i nvidia
$
$# Should show something like:
$# 00:04.0 3D controller: NVIDIA Corporation A100-SXM4-80GB (rev a1)

If no NVIDIA device appears, the GPU may not be attached. Contact support.

Step 2 — Check Driver Installation

>_BASH

$# Check if driver module is loaded
$lsmod | grep nvidia
$
$# Check driver version
$cat /proc/driver/nvidia/version
$
$# If not loaded, try loading manually
$modprobe nvidia

Step 3 — Reinstall NVIDIA Drivers

>_BASH

$# Remove existing drivers
$apt-get purge -y nvidia-* libnvidia-*
$apt-get autoremove -y
$
$# Install fresh
$apt-get install -y nvidia-driver-535
$reboot

After reboot:

>_BASH

$nvidia-smi

Step 4 — Fix CUDA Version Mismatch

>_BASH

$# Check driver version
$nvidia-smi | grep "Driver Version"
$# Driver Version: 535.104.05
$
$# Check CUDA version supported by driver
$nvidia-smi | grep "CUDA Version"
$# CUDA Version: 12.2
$
$# Ensure installed CUDA matches
$nvcc --version

If versions don't match, reinstall CUDA to match the driver's supported version.

Step 5 — Check Secure Boot

Secure Boot can prevent unsigned NVIDIA modules from loading:

>_BASH

$# Check Secure Boot status
$mokutil --sb-state
$
$# If enabled, disable in BIOS or sign the module

Step 6 — PyTorch CUDA Issues

PYTHON

import torch
print(torch.version.cuda)      # Should match installed CUDA
print(torch.cuda.is_available())  # Should be True
print(torch.cuda.device_count())  # Should be >= 1

If CUDA is not available in PyTorch despite drivers working:

>_BASH

$# Reinstall PyTorch with correct CUDA version
$pip3 install --force-reinstall torch --index-url https://download.pytorch.org/whl/cu122

Common Error → Fix Table

Error	Fix
`nvidia-smi not found`	Install nvidia-driver-535
`Driver/library version mismatch`	Reboot after driver install
`CUDA driver version insufficient`	Update driver to match CUDA version
`torch.cuda.is_available() = False`	Reinstall PyTorch with correct CUDA build

Was this article helpful?