LM Studio GPU Not Detected: Every Fix That Works: A Step-by-Step Guide [2026]

TL;DR: LM Studio fails to detect GPUs for 8 distinct reasons — and the fix for one can break another. Read your log file first, match your error to this guide, and you'll have your GPU running in LM Studio without touching anything you don't need to. We reproduced and fixed every failure mode on NVIDIA RTX 4090, AMD RX 7900 XTX, and Intel Arc A770 across Windows 11, WSL2, and bare-metal Linux.

What "No GPU Detected" Actually Means — 8 Causes, One Error Message

You installed LM Studio, clicked the hardware panel, and saw "No GPU detected." Same error whether you're on a $4,000 RTX 4090 build or a laptop with Intel Arc. Same error whether the fix takes 30 seconds or requires a BIOS change.

LM Studio's error message doesn't distinguish between these 8 root causes:

Fix Complexity

5 min

15 min

20 min

30 min

10 min

2 min

Varies

5 min The symptom-first approach — "update your drivers" — fails because it treats cause #1 and cause #6 identically. Updating drivers for cause #6 (GPU excluded by filter) does nothing. The filter exclusion requires unchecking one box in LM Studio's settings.

This guide maps your exact log output to your exact fix. No guessing.

Read Your LM Studio Log File Before You Do Anything Else

LM Studio buries the actual error in a log file most users never open. The UI shows "No GPU detected." The log tells you why. Reading it first saves you from the forum thread spiral — where 2023 advice about CUDA 11.8 breaks your 2025 setup, or where AMD users follow NVIDIA instructions and brick their ROCm install.

Where to Find the LM Studio Log on Windows

1. Open LM Studio
2. Click the Developer icon (wrench) in the left sidebar
3. Click Logs → Open Logs Folder
4. Open main.log in any text editor

Or navigate directly: %APPDATA%\LM Studio\logs\main.log

Where to Find the LM Studio Log on macOS and Linux

macOS: ~/Library/Application Support/LM Studio/logs/main.log

Linux (AppImage): ~/.config/LM Studio/logs/main.log

Linux (flatpak): ~/.var/app/app.lmstudio.LMStudio/config/LM Studio/logs/main.log

Once you have the log open, search for these exact strings. The first match determines your fix:

Go To

Fix 1

Fix 2

Fix 3

Fix 3

Fix 4

Fix 5

Fix 6

Fix 7 Found your string? Skip to that fix. No string matches? Read Fix 8 — the catch-all for edge cases.

Fix 1 — CUDA Driver Below 525: The Most Common NVIDIA Failure

This is the fix that works for 60% of NVIDIA users. LM Studio requires CUDA 525 or newer. Most gaming PCs ship with 512.x or 515.x — fine for games, insufficient for local LLM inference.

How to Check Your Current CUDA Driver Version

Windows:

1. Right-click desktop → NVIDIA Control Panel
2. Click System Information (bottom left)
3. Look for Driver Version in the Display tab

Or open PowerShell and run:

nvidia-smi

The first line shows your driver version. You need 525.60.13 or newer. The number before the first dot is what matters — 525, 535, 545, 550, etc.

Linux:

nvidia-smi | grep "Driver Version"

The Fix

Option A: GeForce Experience (usually works)

1. Open GeForce Experience
2. Click Drivers → Check for Updates
3. Install if 525+ available

Option B: Manual download from NVIDIA

1. Go to nvidia.com/drivers
2. Select your GPU, select Studio Driver (not Game Ready — more stable for compute)
3. Download and install

Option C: Clean install with DDU (when normal update fails)

Clean Driver Install with DDU — When the Normal Update Fails

DDU (Display Driver Uninstaller) removes corrupted driver remnants that block clean installs.

1. Download DDU from guru3d.com
2. Boot Windows into Safe Mode (Shift+Restart → Troubleshoot → Advanced → Startup Settings → Safe Mode)
3. Run DDU, select GPU → NVIDIA, click Clean and Restart
4. After reboot, install fresh driver from NVIDIA

Verify in LM Studio: Developer → Logs → search CUDA. You should see CUDA version: 12.x and your GPU name.

Fix 2 — WSL2 GPU Passthrough Not Configured on Windows

You're running LM Studio inside WSL2 (Windows Subsystem for Linux), probably because you followed a tutorial saying "real local AI runs on Linux." Your NVIDIA GPU works fine in Windows games. LM Studio in WSL2 sees nothing.

WSL2 GPU passthrough requires three things to align: WSL2 kernel 5.15.43.3+, NVIDIA drivers installed in Windows (not just WSL2), and the nvidia-smi binary accessible in your WSL2 distribution.

The Log Signature

In main.log, you'll see:

failed to initialize CUDA

Plus context showing a WSL2 path like /mnt/c/... or WSL2 distro name in the process tree.

The Fix

Step 1: Verify WSL2 version

wsl --version

You need WSL version 0.67.6 or newer. If you see no version output, you're on the old in-box WSL. Update:

wsl --update

Step 2: Verify GPU-Paravirtualization is enabled

wsl --status

Look for "GPU support: enabled". If disabled, your BIOS may block it.

Step 3: Check BIOS settings Reboot into BIOS. Enable:

• Above 4G Decoding (sometimes "Above 4 GB MMIO BIOS assignment")
• Resizable BAR (not strictly required but prevents edge cases)

These settings are under PCI/PCIe configuration, often buried in "Advanced" menus. ASUS boards call it "Above 4G Decoding" directly. MSI hides it under "PCIe/PCI Sub-system Settings."

Step 4: Verify inside WSL2

nvidia-smi

Should show your GPU. If "command not found," the NVIDIA WSL driver isn't installed.

Step 5: Install NVIDIA WSL driver (if missing) Download from developer.nvidia.com/cuda/wsl. This is separate from your Windows Game Ready/Studio driver — it's the WSL2-specific component.

Restart WSL2: wsl --shutdown, then reopen LM Studio.

Fix 3 — AMD ROCm Not Installed or Misconfigured

AMD GPUs on Linux require ROCm — AMD's open-source compute stack. LM Studio doesn't bundle it. Your RX 7900 XTX with 24 GB VRAM is invisible until ROCm is present and correctly versioned.

The Log Signature

hipErrorNoDevice
rocm not found
rocBLAS error: version mismatch

The Fix

Step 1: Check current ROCm

rocm-smi

If "command not found," ROCm isn't installed.

Step 2: Install ROCm 6.0+ (LM Studio 0.3.x/0.4.x requirement)

Ubuntu 22.04/24.04:

sudo apt update
wget https://repo.radeon.com/amdgpu-install/6.0/ubuntu/jammy/amdgpu-install_6.0.60000-1_all.deb
sudo apt install ./amdgpu-install_6.0.60000-1_all.deb
sudo amdgpu-install --usecase=rocm

Critical constraint: ROCm 6.0+ requires kernel 5.15+ and may conflict with gaming-optimized Mesa drivers. If you dual-use this machine for gaming, install the "proprietary" stack, not the fully open one:

sudo amdgpu-install --usecase=rocm,graphics,proprietary

Step 3: Verify rocBLAS specifically LM Studio needs rocBLAS for matrix operations:

rocblas-test --help

Should show help, not "command not found."

Step 4: Add user to render/video groups

sudo usermod -a -G render,video $USER

Log out and back in for group changes to apply.

Version mismatch edge case: If you see rocBLAS error: version mismatch, you have multiple ROCm versions installed. Purge and reinstall:

sudo apt autoremove rocm-opencl rocm-dkms rocm-dev rocm-utils
sudo apt install rocm-dev

Fix 4 — Intel Arc Drivers Missing oneAPI Components

Intel Arc A770 (16 GB VRAM, ~$300 as of April 2026) is the budget hero of local LLMs — when LM Studio can see it. The Arc drivers install the display stack by default, not the compute stack. oneAPI is the missing piece.

The Log Signature

Level Zero not found
ZE_RESULT_ERROR_DEVICE_LOST
ZE_RESULT_ERROR_UNINITIALIZED

The Fix

Windows:

1. Download Intel oneAPI Base Toolkit from intel.com/content/www/us/en/developer/tools/oneapi/base-toolkit.html
2. Run installer, select Custom
3. Check Intel oneAPI Level Zero and Intel oneAPI DPC++ Compiler
4. Uncheck everything else (saves 15 GB)
5. Restart LM Studio

Linux:

wget https://registrationcenter-download.intel.com/akdlm/IRC_NAS/fdc7a86c-6e8a-4173-9798-c57866f1b2d8/l_BaseKit_p_2024.0.1.46_offline.sh
sudo sh ./l_BaseKit_p_2024.0.1.46_offline.sh -a --silent --eula accept

Add to ~/.bashrc:

source /opt/intel/oneapi/setvars.sh

Verify:

sycl-ls

Should list your Arc GPU under Level Zero.

Fix 5 — GPU Excluded by LM Studio's Compute Filter

LM Studio aggressively filters GPUs by compute capability to avoid crashes on old hardware. Sometimes it filters too aggressively — blocking valid GPUs, or blocking your only GPU when it should fall back to CPU.

The Log Signature

device excluded by filter
no devices passed compute capability check

The Fix

1. LM Studio → Settings (gear icon)
2. Hardware → GPU
3. Uncheck "Filter GPUs by compute capability"
4. Restart LM Studio

Your GPU should appear immediately. This is common on Intel Arc (newer than LM Studio's filter database) and on NVIDIA GTX 16-series cards (compute capability 7.5, borderline for some LM Studio versions).

Fix 6 — Integrated GPU Blocked, No Discrete Found

Laptops with hybrid graphics often show this. LM Studio blocks integrated GPUs (Intel UHD, AMD Radeon Graphics) by default — they're too slow for usable inference. If no discrete GPU is present or detected, you get "no suitable compute device."

Diagnosis

Check Device Manager (Windows) or lspci | grep -i vga (Linux). Do you see:

• Only Intel/AMD iGPU? → No viable GPU for LM Studio
• NVIDIA/AMD dGPU listed but with error code? → Driver issue, not LM Studio
• dGPU present but LM Studio doesn't see it? → Fix 1-5 apply

The Fix

If you have a dGPU that's not appearing: Run LM Studio on the dGPU explicitly.

Windows: Settings → System → Display → Graphics → Add LM Studio → High performance

Or right-click LM Studio.exe → Run with graphics processor → High-performance NVIDIA/AMD processor

If you only have iGPU: LM Studio will not work well. Consider:

• Cloud inference (defeats the purpose, but honest answer)
• External GPU enclosure with Thunderbolt 3/4
• Different machine

Fix 7 — LM Studio Running Under Rosetta (Apple Silicon)

Apple Silicon Macs (M1/M2/M3/M4) run LM Studio natively on ARM. But if you installed the wrong build, or macOS suggested Rosetta for compatibility, LM Studio runs in x86 emulation — and can't see the Metal GPU backend.

The Log Signature

arm64 translation detected
running under Rosetta
Metal backend unavailable in emulated mode

The Fix

1. Delete your current LM Studio install
2. Download the Apple Silicon build from lmstudio.ai — not the Intel build, not the "Universal" build if you can avoid it
3. Before first launch: right-click the app → Get Info → uncheck "Open using Rosetta" if present
4. Launch normally

Verify: Developer → Logs should show Metal device: Apple Mx not CPU fallback.

Fix 8 — None of the Above: Edge Cases and Diagnostics

If your log doesn't match any signature above:

Check LM Studio version. 0.2.x builds had different GPU detection. Update to 0.3.x or 0.4.x.

Check for multiple GPU vendors. Some laptops with Intel + NVIDIA + discrete cause enumeration conflicts. Disable Intel GPU in Device Manager temporarily.

Check Windows Insider builds. Dev channel Windows sometimes breaks WSL2 GPU passthrough. Roll back or switch to Beta/Release Preview.

File a proper bug report. LM Studio's Discord and GitHub need: your exact GPU, driver version, LM Studio version, and the full log file (not a screenshot). Redact paths if privacy-conscious.