The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Run ollama ps while a model is loaded and read the Processor column. 100% GPU means the model is fully on the GPU. 100% CPU means it is running in system memory. A split such as 48%/52% CPU/GPU means partial offload, with part of the work on the GPU and the rest on the CPU. If the result is CPU or a split, the fix depends on which layer is blocking the GPU: the operating system seeing the card, the driver or ROCm stack, device permissions, container passthrough, or Ollama’s own placement.
Native Linux, native Windows, WSL2 and Docker each have a different setup path and a different first test, so this guide treats them separately.
As an Amazon Associate I earn from qualifying purchases.
Ollama using CPU instead of GPU: measure placement first
Load a model before you check, because ollama ps lists only models that are currently loaded:
ollama psJudge the result by the Processor column, not by whether a GPU monitor shows some activity. Ollama’s FAQ documents the placement labels, and the same labels apply to any model:
#1 Best Overall
- Axial-tech fans now feature a smaller fan hub that facilitates longer blades and a barrier ring that increases downward air pressure
- 2.5-slot design allows for greater build compatibility while maintaining cooling performance
- 0dB technology lets you enjoy light gaming in relative silence
- Dual BIOS switch lets you toggle between Quiet and Performance BIOS profiles
- Dual ball fan bearings last up to twice as long as sleeve bearing designs
- 100% GPU: the loaded model is fully on the GPU. There is no CPU fallback to fix.
- 100% CPU: the model is in system memory. Continue with the layer checks below.
- A split such as 48%/52% CPU/GPU: partial offload. Part of the model runs on the GPU and the rest stays on the CPU. Treat this as a placement result to investigate, not as proof of a driver fault by itself.
Before you change drivers or settings, record the following so you can compare results after each change:
- Ollama version, from
ollama --version - GPU model and operating system
- Driver version
- How Ollama was installed: native Linux, native Windows, inside WSL2, or in a container
- The relevant server log (see the logs section below)
Choose your path by environment
The same symptom can come from different layers depending on where Ollama runs. Use this table to pick the first test.
| Environment | First check | Then check | Common failure point |
|---|---|---|---|
| Native Linux, NVIDIA | nvidia-smi on the host |
ollama ps, then journalctl -u ollama |
Driver not returning the GPU, or UVM initialization errors |
| Native Linux, AMD | The Ollama process can open /dev/kfd and /dev/dri |
Server log, plus kernel messages for amdgpu or kfd |
Missing video or render group membership, or a ROCm kernel driver older than the bundled libraries |
| Native Windows | Vendor driver version against Ollama’s stated minimums | server.log under %LOCALAPPDATA%Ollama |
Driver below the minimum, or an AMD GPU without a usable ROCm v7 or Vulkan path |
| WSL2 with NVIDIA | nvidia-smi inside the Linux distribution |
ollama ps in the same distribution |
Windows NVIDIA driver or WSL GPU passthrough not exposing the card |
| Docker on Linux | docker run --gpus all ubuntu nvidia-smi (NVIDIA) |
ollama ps inside the container |
NVIDIA Container Toolkit or Docker’s NVIDIA runtime not configured; AMD device or group access missing |
| Docker inside WSL2 with NVIDIA | nvidia-smi inside the distribution, then a GPU test inside the container |
ollama ps inside the container |
Each boundary must expose the GPU separately; a failure at any one blocks GPU use |
Check each layer in order
The layers stack, and a failure at a lower layer looks like an Ollama problem. Test from the host upward:
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →- Host: the driver utility on the machine that owns the GPU sees the card.
- WSL distribution, if used: the same utility inside that distribution sees the passthrough device. A working
nvidia-smion the Windows host does not prove that Ollama inside WSL has GPU access. - Container, if used: the container runtime exposes the device to the container.
- Ollama:
ollama psreports GPU placement after a model loads.
Read the logs to separate failure types
Logs help distinguish initialization failures from unsupported hardware and from container access problems. Read them when the layer checks pass but ollama ps still reports CPU.
Rank #2
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5070 Ti
- Integrated with 16GB GDDR7 256bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
- Linux (systemd install):
journalctl -u ollama - Windows:
%LOCALAPPDATA%Ollamaserver.log, which holds the most recent server logs. - More discovery detail: set
OLLAMA_DEBUG=1. On Linux, open a service override withsudo systemctl edit ollama.service, add the lineEnvironment="OLLAMA_DEBUG=1"under a[Service]section, then restart the service. On Windows, set it as a user environment variable and fully restart Ollama. - AMD: also set
AMD_LOG_LEVEL=3, and check kernel messages withsudo dmesg | grep -iE 'amdgpu|kfd'.
Ollama not detecting an NVIDIA GPU on Linux
Confirm the driver sees the card
nvidia-smiOllama’s Linux documentation uses this command to confirm that NVIDIA drivers are installed and returning GPU details. Its troubleshooting page recommends current NVIDIA drivers. If nvidia-smi fails, fix the driver before touching Ollama. If it works but the server log shows initialization or device discovery errors, move to the UVM steps below.
UVM initialization errors
Ollama’s troubleshooting page lists these options when the log shows initialization or discovery errors. They change a kernel module, so follow your distribution’s administration practice and make the change outside production hours.
- Check that the UVM driver is loaded:
sudo nvidia-modprobe -u. - Reload the module with
sudo rmmod nvidia_uvm, thensudo modprobe nvidia_uvm. The unload fails while any process still uses the GPU, so stop Ollama and other GPU workloads first. - Reboot, restart Ollama, and recheck
ollama ps.
After suspend or resume
Ollama documents a possible NVIDIA discovery failure after Linux suspend and resume, in which Ollama falls back to the CPU. Its stated workaround is reloading nvidia_uvm as described above. This is one documented case, not the explanation for every NVIDIA CPU fallback.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
NVIDIA GPU in Docker
On a Docker host, a working nvidia-smi on the host is not enough. Test the container runtime directly:
Rank #3
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5060
- Integrated with 8GB GDDR7 128bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
docker run --gpus all ubuntu nvidia-smiIf that command fails, the container cannot see the GPU. Then:
- Install the NVIDIA Container Toolkit.
- Configure Docker’s NVIDIA runtime by following the toolkit’s documentation.
- Restart Docker.
- Launch the Ollama container with
--gpus=all, load a model, and check placement inside the container. For a container namedollama, that isdocker exec -it ollama ollama ps.
Ollama AMD GPU not detected on Linux
Device access and groups
Ollama’s Linux AMD path typically needs the Ollama process to belong to the video and/or render group to reach /dev/kfd. Check the ownership and numeric group IDs of the device nodes, and the groups your user holds:
ls -ln /dev/kfd /dev/dri
idIf the service user is not in the group that owns those device nodes, add it, then restart Ollama. On systemd installs, that is sudo systemctl restart ollama.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchDiscovery stalls on an older ROCm driver
If the log shows AMD discovery timing out, the kernel driver may be older than the ROCm libraries Ollama bundles. Ollama’s GPU documentation says its Linux AMD ROCm path requires ROCm v7. Its troubleshooting page describes an older driver, ROCm 6.x or earlier, stalling discovery and causing CPU fallback when it is incompatible with the bundled ROCm 7 libraries. The documented fix is to update to a compatible ROCm v7 driver with AMD’s amdgpu-install utility, then reboot and restart Ollama.
Rank #4
- Powered by Radeon RX 9070 XT
- WINDFORCE Cooling System
- Hawk Fan
- Server-grade Thermal Conductive Gel
- RGB Lighting
Driver compatibility depends on the GPU and the system. Check AMD’s current supported platform and GPU documentation before you change a driver on a production machine.
AMD GPU in Docker
Ollama’s Docker documentation provides an AMD image, ollama/ollama:rocm, and its example exposes /dev/kfd and /dev/dri to the container:
docker run -d --device /dev/kfd --device /dev/dri -v ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama:rocmIf discovery still fails, pass the numeric group IDs that own those device nodes into the container with --group-add, then recheck the container log.
Ollama GPU not working on native Windows
Check the requirements
Native Windows is separate from WSL. Ollama’s Windows documentation, as checked in October 2026, lists:
Best Value
- Axial-tech fans now feature a smaller fan hub that facilitates longer blades and a barrier ring that increases downward air pressure
- Phase-change GPU thermal pad helps ensure optimal heat transfer, lowering GPU temperatures for enhanced performance and reliability
- 2.5-slot design allows for greater build compatibility while maintaining cooling performance
- Dual-ball fan bearings last up to twice as long as standard conventional sleeve bearings designs
- 0dB technology lets you enjoy light gaming in relative silence
- Windows 10 22H2 or newer, Home or Pro editions.
- NVIDIA: driver 551.61 or newer.
- AMD: an AMD ROCm v7/HIP7-capable driver stack, or a Vulkan-capable AMD driver, for AMD acceleration.
These floors and the supported GPU lists change. Confirm them on Ollama’s current Windows page before you change drivers.
Radeon RX 6000 and RDNA2 systems
Ollama notes that some RDNA2 and Radeon RX 6000 systems may not expose ROCm v7 on current Windows AMD drivers, and it recommends Vulkan as the fallback for those systems. That advice applies to those GPU and driver combinations. It is not a statement about every AMD card.
Restart cleanly
After you change environment variables or drivers, fully quit Ollama, start it again, load a model, and check ollama ps. If you only reopen a window, the previous server process may still hold the old settings, and a fix can appear to fail when it has not been applied.
Ollama GPU not working in WSL
Ollama’s Linux installer notes state that WSL2 GPU support runs through NVIDIA passthrough, and the installer checks for nvidia-smi. Microsoft’s CUDA-on-WSL guidance asks you to install a CUDA-enabled NVIDIA driver on Windows and keep WSL current. NVIDIA’s guide says the Windows driver supplies the GPU interface to WSL, and it warns against installing a Linux NVIDIA display driver inside WSL2.
Quick Recap
Passthrough path, step by step
- On Windows, install a current NVIDIA driver with WSL support.
- Update WSL from PowerShell or Command Prompt with
wsl.exe --update, then open the Linux distribution you plan to use. - Run
nvidia-smiinside that distribution. If the GPU is not visible there, fix the Windows driver or WSL passthrough before you debug Ollama. - Install and run Ollama in the same distribution, load a model, and check
ollama ps. - If you also run Docker in that distribution, verify GPU access inside the container as well. Each boundary must expose the device.
Limits of the WSL2 guidance
- The documented WSL2 path is NVIDIA. Neither Ollama’s WSL2 installer notes nor Microsoft’s CUDA guidance establishes AMD GPU passthrough for Ollama under WSL2, so do not assume an AMD card will work there.
- Microsoft Learn’s CUDA-on-WSL guidance lists Windows 10 21H2 or Windows 11, and a WSL kernel of 5.10.43.3 or higher, as prerequisites. Those are WSL prerequisites. Ollama’s native Windows requirements are newer, so do not use the WSL list to decide whether a Windows machine qualifies for native Ollama.
,
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
