Massed Compute

Running large language models requires GPU infrastructure that’s both powerful and cost-effective. This guide walks through deploying Ollama on Massed Compute GPU VMs with interactive model selection and proper teardown.

You’ll provision a GPU instance, bootstrap Ollama with NVIDIA drivers, select from popular model options based on your VRAM constraints, and run verification tests before cleanup.

Technology Stack

Requirements

Massed Compute VM Pricing

Step-by-Step Deployment

curl -X POST https://api.massedcompute.com/v1/instances \
  -H "Authorization: Bearer YOUR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "sku": "gpu_1x_l40",
    "imageId": 184,
    "name": "ollama-llm-test"
  }'

ssh ubuntu@YOUR_VM_IP

# Verify GPU is available
nvidia-smi

# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Start Ollama service
sudo systemctl enable ollama
sudo systemctl start ollama

# Verify installation
ollama --version

# Replace MODEL_TAG with your selection
ollama pull llama3.1:8b

# Verify model is available
ollama list

# Run smoke test
ollama run llama3.1:8b "Hello, can you confirm you're working properly?"

# Check GPU utilization
nvidia-smi

# From your local machine
ssh -L 11434:127.0.0.1:11434 ubuntu@YOUR_VM_IP

# Test local API access
curl http://127.0.0.1:11434/api/tags

# Via API
curl -X DELETE https://api.massedcompute.com/v1/instances/INSTANCE_ID \
  -H "Authorization: Bearer YOUR_API_TOKEN"

# Or via dashboard
# Navigate to Instances → Actions → Terminate

Troubleshooting

SSH Connection Issues: If you get host key warnings, refresh your known hosts with ssh-keygen -R YOUR_VM_IP. Confirm the VM is still running in your dashboard.

No GPU Detected: Verify you launched with image 184. Check GPU status with nvidia-smi. If no GPU is visible, the instance may need a restart or driver installation.

Model Pull Out of Memory: Choose a smaller model (3B or 7B) or upgrade to a larger GPU SKU. Check available VRAM with nvidia-smi before pulling large models.

Empty or Hanging Responses: Confirm the model pull completed fully with ollama list (should show non-zero size). Restart Ollama with sudo systemctl restart ollama.

API Binding Issues: Ollama defaults to local-only binding. Check service configuration in /etc/systemd/system/ollama.service.d/ if you need external access (not recommended without authentication).

Skip All of This: Deploy with an AI Agent

This entire workflow is available as a tested, machine-readable recipe in the Massed Compute MCP. Instead of manual setup, configure the MCP server and request deployment with natural language.

Add to your MCP settings:

{
  "mcpServers": {
    "massed-compute": {
      "type": "http",
      "url": "https://vm.massedcompute.com/api/mcp",
      "headers": { "Authorization": "Bearer MC_TOKEN" }
    }
  }
}

Then say:

The agent matches your request against the recipe catalog, provisions the appropriate GPU instance, runs the setup and verification steps above, and reports back with connection details and test results. If any step fails, it stops and reports the specific error rather than continuing with broken infrastructure.

This recipe was last tested on June 10, 2026.

Quick Setup Reference

For experienced users, here’s the condensed deployment sequence:

# 1. Launch GPU VM
curl -X POST https://api.massedcompute.com/v1/instances \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -d '{"sku":"gpu_1x_l40","imageId":184}'

# 2. SSH and install
ssh ubuntu@VM_IP
curl -fsSL https://ollama.com/install.sh | sh
sudo systemctl enable --now ollama

# 3. Pull model and test
ollama pull llama3.1:8b
ollama run llama3.1:8b "Test message"

# 4. Optional: tunnel for local access
ssh -L 11434:127.0.0.1:11434 ubuntu@VM_IP

# 5. Cleanup
curl -X DELETE https://api.massedcompute.com/v1/instances/ID

Frequently Asked Questions