Skip to main content

Root KVM Virtual Machines

NovaPod Root KVM VMs provide hardware-level virtualization with dedicated GPU PCI-passthrough, unshared system kernels, and persistent NVMe boot drives. Unlike containerized pods, a KVM VM gives you the exact freedom of a physical bare-metal server.

When to Choose a VM​

Use a Root KVM VM if your workload requires:

  • Full Root & Kernel Modules: Installing custom Linux kernels, proprietary drivers, eBPF probes, WireGuard, or custom network tunnels.
  • Docker-in-Docker / Kubernetes: Running local docker compose, k3s, Minikube, or MicroK8s clusters directly on the machine.
  • Persistent System Drive: Everything you install on the entire root filesystem (/, /opt, /etc, /usr) is saved permanently across stop/start cycles.
  • Dedicated Public IPv4: A static public IP address with all ingress ports (1–65535) open and configurable via security group firewall rules.
  • Windows OS Support: Running Windows Server GPU instances with Remote Desktop Protocol (RDP) for DirectX, Unreal Engine, or proprietary Windows-only pipelines.

Launching a Root VM​

Via CLI​

np vm create \
  --gpu H100_SXM \
  --count 1 \
  --os ubuntu-24.04 \
  --disk 150GB \
  --public-ip \
  --name production-worker

Via Web Console​

In the Marketplace:

  1. Switch compute mode to Root KVM VPS.
  2. Select your base OS: Ubuntu 24.04 LTS, Debian 12, or Windows Server 2022.
  3. Choose your NVMe disk capacity.
  4. Click Launch VM.

Lifecycle & Storage Billing​

When your computational workload is idle:

  1. Pause Instance:
    • Releases the GPU hardware back to the pool.
    • GPU compute fees drop to $0.00.
    • Your system drive, installed tools, and caches remain 100% frozen on NVMe storage.
    • Standby storage rate: $0.00014 per GB per hour (~$0.10/month for 100GB).
  2. Resume Instance:
    • Re-attaches an identical GPU in seconds.
    • The VM boots straight back up into your saved state.
  3. Destroy Instance:
    • Permanently wipes the VM and disk, stopping all fees.