Ollama Memory Configuration

When running Ollama on xCloud, you may encounter memory errors like:

500: model requires more system memory (4.5 GiB) than is available (2.0 GiB)

This happens because the Ollama Docker container has a memory limit (often 2GB by default), even if your server has more RAM available.

Prerequisites

• SSH access to your xCloud server (see SSH Setup)
• Access to the Server Command Runner in xCloud dashboard

Diagnosing the Issue

Check Container Memory Limit

Run via SSH or the xCloud Command Runner:

docker inspect $(docker ps -q --filter name=ollama) | grep -i memory

Look for the "Memory" value. 2147483648 = 2GB, which is often too low for larger models.

Check Your Model Requirements

Different models require different amounts of RAM:

• llama2-uncensored (7B): ~4.5GB
• deepseek-r1:1.5b: ~2GB
• llama2:13b: ~8GB
• llama2:70b: ~40GB+

Fixing the Memory Limit

Step 1: Get Current Container Configuration

Save the volume mount path before removing the container:

docker inspect ollama-your-domain.com --format '{{range .Mounts}}{{.Source}}:{{.Destination}}{{"\n"}}{{end}}'

Example output: /var/www/your-domain.com/data:/root/.ollama

Step 2: Stop and Remove the Old Container

docker stop ollama-your-domain.com && docker rm ollama-your-domain.com

Step 3: Recreate with More Memory

Replace <VOLUME_PATH> with your path from Step 1:

docker run -d --name ollama-your-domain.com --memory=12g --restart unless-stopped -p 127.0.0.1:18016:11434 -v <VOLUME_PATH>:/root/.ollama ollama/ollama:latest

For example:

docker run -d --name ollama-your-domain.com --memory=12g --restart unless-stopped -p 127.0.0.1:18016:11434 -v /var/www/your-domain.com/data:/root/.ollama ollama/ollama:latest

Step 4: Verify the New Limit

docker inspect ollama-your-domain.com --format '{{.HostConfig.Memory}}'

Should return 12884901888 (12GB).

Reconnecting Open WebUI

After recreating the Ollama container, Open WebUI may lose connection because the container is no longer on the same Docker network.

Step 1: Find the Open WebUI Network

docker inspect $(docker ps -q --filter name=openweb) --format '{{range $k, $v := .NetworkSettings.Networks}}{{$k}}{{end}}'

Step 2: Connect Ollama to the Network

docker network connect --alias ollama <NETWORK_NAME> ollama-your-domain.com

For example:

docker network connect --alias ollama your-openwebui-network-name ollama-your-domain.com

The --alias ollama is important because Open WebUI is configured to connect to http://ollama:11434.

Step 3: Verify Connection

In Open WebUI:

1. Go to Admin Settings > Connections
2. Click the refresh button next to the Ollama API URL
3. You should see “server connection verified”

Memory Limits Explained

The --memory=12g flag sets a ceiling, not a reservation:

• Ollama can use up to 12GB, but only consumes what it actually needs
• Other apps can freely use any RAM that Ollama isn’t actively using
• No memory is pre-allocated or blocked from other containers

Your server’s total RAM should be higher than the sum of all container limits to avoid the OOM (Out of Memory) killer terminating containers.

Troubleshooting

”Failed to fetch models” in Open WebUI

The Ollama container isn’t reachable. Check:

1. Container is running: docker ps | grep ollama
2. Container is on the correct network (see Reconnecting section above)

Model still shows memory error

Verify the container was recreated with the new limit:

docker inspect ollama-your-domain.com | grep '"Memory"'

If it still shows 2GB, the container wasn’t properly recreated.

Container won’t start

Check Docker logs:

docker logs ollama-your-domain.com