GPU environment setup & SSH
Prepare and check a Linux GPU host before installing a model server.
1. Prerequisites
Use the IP, SSH user and public-key association confirmed for your instance. Install Docker Engine and the NVIDIA driver appropriate to the OS and GPU, then follow the Container Toolkit installation guide. Restarting Docker affects existing containers; schedule that change. These scripts were syntax-checked, not executed on an Arvica GPU.
2. Connect from your computer
Replace the address, username and private-key path. Verify the SSH host fingerprint against the provisioned host before trusting a new host. Keep the private key on your computer.
ssh -i ~/.ssh/arvica_ed25519 ubuntu@INSTANCE_IP3. Check the runtime
Set ARVICA_CUDA_IMAGE to a driver-compatible NVIDIA CUDA image tag. Expected result: host and container both list the allocated GPU. A successful check confirms GPU visibility, not model performance.
#!/usr/bin/env bash
set -euo pipefail
# Run on the confirmed Linux GPU instance, not your laptop.
uname -m
nvidia-smi
nvidia-smi --query-gpu=name,memory.total,driver_version --format=csv
docker version
# Set a CUDA image tag compatible with the installed driver.
: "${ARVICA_CUDA_IMAGE:?Set a compatible nvidia/cuda image tag}"
docker run --rm --gpus all "$ARVICA_CUDA_IMAGE" nvidia-smi
4. Diagnose failures
SSH timeout: check instance state, IP and port access. Permission denied: check SSH user and selected public key. nvidia-smi missing: verify host driver installation. Host works but container fails: check the Container Toolkit/runtime and driver-image compatibility. Never disable SSH host-key checks to bypass a mismatch.