How to Move Ollama Models to Another Drive (Windows, Mac, Linux, Docker) 2026

Local AI models fill a disk fast. A couple of 30B models and a vision model can take 60 GB or more, and by default Ollama puts them all on your system drive.

The good news is that moving them to a bigger drive takes about five minutes, and you do not need to download anything again.

This guide shows you how to move Ollama models to another drive on Windows, macOS, Linux and Docker, using the app’s Model location setting, the OLLAMA_MODELS variable or a folder link.

It also covers how to check where your models are now, how to see tokens per second after the move, and how to fix the “models disappeared” problem.

Where Ollama Stores Models

Inside models there are two folders:

• blobs: the actual model weights, saved as files named like sha256-.... This is where all the gigabytes are.
• manifests: small files that tell Ollama which blobs make up each model and tag.

Always move both folders together. Moving only blobs makes every model vanish from ollama list.

Which Method to Use

Before You Move

• Pick a fast drive. An NVMe SSD is best. Models load from disk into RAM or VRAM each time they start, so a slow USB hard drive can turn a 5-second load into a minute. Generation speed after loading does not change.
• Use a drive that is always connected. If an external drive is missing when Ollama starts, it cannot find your models.
• Check the space: run ollama list to see each model’s size and make sure the new drive has room for all of them, plus new ones.
• Use NTFS or ext4, not exFAT or FAT32. FAT32 cannot store files over 4 GB, and most model blobs are bigger.
• Quit Ollama completely before moving files (tray icon or menu bar icon > Quit, or stop the service on Linux).

Move Ollama Models on Windows

Option A: The Model location setting (easiest)

1. Quit Ollama: right-click the llama icon in the system tray > Quit Ollama.
2. Create the new folder, for example D:\Ollama\models.
3. Move your models: open %HOMEPATH%\.ollama\models and move the blobs and manifests folders into D:\Ollama\models. Moving across drives copies then deletes, so large libraries take a few minutes.
4. Start Ollama, open the app window and go to Settings.
5. Under Model location, click Browse and choose D:\Ollama\models.
6. Restart Ollama (Quit and open it again), then run ollama list in a new terminal. All your models should be there.

Option B: The OLLAMA_MODELS environment variable

1. Quit Ollama from the tray.
2. Move the models as in Option A.
3. Open Settings (Windows 11) or Control Panel (Windows 10) and search for environment variables. Click Edit environment variables for your account.
4. Click New under User variables. Name: OLLAMA_MODELS. Value: D:\Ollama\models. Click OK.
5. Start Ollama from the Start menu (not an old terminal, which still has the old variables).
6. Open a new terminal and run ollama list.

Prefer the command line? This sets the same user variable in PowerShell:

[Environment]::SetEnvironmentVariable("OLLAMA_MODELS", "D:\Ollama\models", "User")

Option C: A folder junction

A junction makes the old path point to the new drive, so every tool that expects C:\Users\you\.ollama\models keeps working. Quit Ollama, move the whole models folder to D:\Ollama\models, then in Command Prompt:

mklink /J "%USERPROFILE%\.ollama\models" "D:\Ollama\models"

Use either a junction or OLLAMA_MODELS, not both, so you always know where models really are.

Move Ollama Models on macOS

The Mac app has the same Settings > Model location option, which is the easiest way. Quit Ollama from the menu bar, move ~/.ollama/models to the new drive (for example /Volumes/FastSSD/ollama/models), then pick that folder under Model location and restart Ollama.

If you prefer the environment variable, Ollama’s docs say to set it with launchctl because the Mac app does not read your shell profile:

launchctl setenv OLLAMA_MODELS "/Volumes/FastSSD/ollama/models"

Then quit and reopen Ollama. launchctl setenv does not survive a reboot, so the app setting is the better long-term choice. On external drives, give Ollama access if macOS asks (System Settings > Privacy & Security > Files and Folders).

Move Ollama Models on Linux

With the official install script, Ollama runs as a systemd service under its own ollama user, so the new folder must belong to that user.

1. Stop the service:

   sudo systemctl stop ollama

2. Create the new folder and move the models:

   sudo mkdir -p /mnt/data/ollama
   sudo mv /usr/share/ollama/.ollama/models /mnt/data/ollama/

3. Give the ollama user access (from Ollama’s docs):

   sudo chown -R ollama:ollama /mnt/data/ollama

   Every parent folder must also be readable by the ollama user. A drive mounted under your home folder (for example /home/you/... with 700 permissions) will fail.

4. Tell the service where the models are:

   sudo systemctl edit ollama

   Add these lines in the editor and save:

   [Service]
   Environment="OLLAMA_MODELS=/mnt/data/ollama/models"

5. Reload and start:

   sudo systemctl daemon-reload
   sudo systemctl start ollama
   ollama list

Make sure the drive mounts at boot (an entry in /etc/fstab), or the service starts before the drive and sees an empty folder. Adding RequiresMountsFor=/mnt/data under [Unit] in the same override makes systemd wait for the mount.

Ollama in Docker

In Docker the models live wherever the /root/.ollama volume points. To use a folder on another drive, start the container with a bind mount instead of a named volume:

docker run -d --gpus=all -v /mnt/data/ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama

To move an existing named volume, copy it into the new folder first:

docker stop ollama
docker run --rm -v ollama:/from -v /mnt/data/ollama:/to ubuntu cp -a /from/. /to/

Then recreate the container with the new -v. Our Ollama Docker guide has the full setup, including Compose with Open WebUI.

Move the Ollama App Itself (Windows)

The app is small (around 4 GB with GPU libraries) and installs in your home folder without admin rights. If even that is too much for drive C, reinstall it to another drive with the official installer flag:

OllamaSetup.exe /DIR="D:\Programs\Ollama"

This moves the program only. Models still follow the Model location setting or OLLAMA_MODELS. For a fresh install on any system, see how to install Ollama.

Check Tokens per Second After the Move

Moving models changes load time, not generation speed. To confirm everything still runs at full speed, add --verbose:

ollama run gemma4 --verbose "Explain RAID 0 in two lines"

After the answer you get timings like these:

total duration:       4.1s
load duration:        1.9s
prompt eval rate:     812.40 tokens/s
eval count:           58 token(s)
eval rate:            61.25 tokens/s

• eval rate is your generation speed in tokens per second.
• load duration is how long it took to load from disk. This is the number a slow drive makes worse.
• Inside an interactive chat, type /set verbose to turn timings on.

Also run ollama ps and check the PROCESSOR column says 100% GPU. If tokens per second dropped a lot, the cause is GPU offload, not the disk; see our GPU VRAM usage guide.

Free Up More Space

• Delete models you no longer use: ollama rm model:tag. Our delete Ollama models guide covers bulk clean-up.
• Keep one quant per model. You rarely need both a Q4 and a Q8 of the same model.
• Leftover blobs: Ollama removes unused blobs on startup by default. Do not set OLLAMA_NOPRUNE=1 unless you have a reason.
• Pick smaller models that fit your card: see the best Ollama models by VRAM size.

Troubleshooting

ollama list is empty after the move

• Ollama is still running with the old setting. Quit it fully (tray > Quit), then start it again and use a new terminal.
• The path points one level too high or too low. It must point to the folder that contains blobs and manifests.
• Only blobs was moved. Move manifests too.

Ollama downloads the models again

Same cause: it is looking at a different folder. Check the active path in Settings > Model location, or on Linux with systemctl show ollama | grep OLLAMA_MODELS.

“permission denied” on Linux

The ollama user cannot read the folder. Run sudo chown -R ollama:ollama /mnt/data/ollama and check the parent folders with namei -l /mnt/data/ollama/models.

The environment variable does nothing on Windows

You set it as a System variable in a terminal that was already open, or Ollama was not restarted. Set it under your user account, quit Ollama from the tray and start it from the Start menu.

Models load very slowly from the new drive

The drive is slow (USB 2, SMR hard drive or a network share). Move the models you use most to an SSD; generation speed is unaffected once the model is loaded.

Models vanish when the external drive is unplugged

Expected. Reconnect the drive and restart Ollama. Do not pull a model while the drive is missing, or Ollama will download into an empty folder.

Open WebUI or another app still uses the old models

Apps talk to the Ollama server, so they follow the server’s model folder. Restart the app after restarting Ollama. If the app runs its own Ollama in Docker, move that container’s volume instead.

My Take

Move your models the first time drive C gets tight, not when it is full. On Windows and Mac, the Model location setting is the cleanest route; on Linux, a systemd override with OLLAMA_MODELS plus the right owner is all you need.

Put models on an NVMe drive if you can. It keeps load times short, which matters more than people expect when you switch between several models a day.

FAQ

How do I change the Ollama model location?

In the Ollama app open Settings > Model location > Browse and pick a folder, or set the OLLAMA_MODELS environment variable, then restart Ollama.

Where are Ollama models stored on Windows?

In C:\Users\<your name>\.ollama\models, split into blobs (weights) and manifests.

Can I move Ollama models without downloading them again?

Yes. Move the blobs and manifests folders together, point Ollama at the new folder and restart. Nothing is re-downloaded.

Does moving models to an external drive slow Ollama down?

Only the loading step. Once the model is in RAM or VRAM, tokens per second are the same.

How do I see tokens per second in Ollama?

Run ollama run <model> --verbose, or type /set verbose in a chat. The “eval rate” line is your tokens per second.

Does uninstalling Ollama delete models on another drive?

No. Ollama’s docs note the Windows uninstaller does not remove models when you changed OLLAMA_MODELS. Delete that folder by hand if you want the space back.

Can two computers share one Ollama models folder?

Not safely at the same time over a network share. Run one Ollama server and point the other computers at it with OLLAMA_HOST instead.

Sources

• Ollama docs: FAQ (where models are stored, OLLAMA_MODELS, environment variables)
• Ollama docs: Windows (changing model and install location)
• Ollama docs: macOS and Linux
• Ollama app source: Model location setting
• Ollama source: OLLAMA_MODELS and OLLAMA_NOPRUNE
• Ollama source: –verbose timings flag