← Docs Index
Everything needed before running the Intel® AI for Enterprise Solutions installer, whether you're deploying on-premises, air-gapped, or on a single bare-metal box.
Operating system
| OS | Status |
|---|
| Ubuntu 22.04 | ✅ Supported |
| Ubuntu 24.04 | ✅ Supported |
| RHEL 8 / Rocky Linux 8 (x86_64) | ✅ Supported |
Hardware minimums
| Deployment | CPU cores | RAM | Disk |
|---|
| Platform + Observability only | 16 | 32 GB | 200 GB |
| Inference — 3B parameter model | 32 | 64 GB | 300 GB |
| Inference — 8B parameter model | 64 | 128 GB | 400 GB |
Intel® Xeon® processors are recommended for CPU-based inference. NUMA-aware pinning (NRI Balloons) and AMX acceleration are built into the platform for Xeon workloads.
Access
| Requirement | Single-node | Multi-node |
|---|
| Passwordless sudo on installer host | Required | Required |
| Passwordless SSH from installer host to all nodes | — | Required |
| Passwordless sudo on all target nodes | — | Required |
Network
- Internet access is required during install (package downloads, container images, model weights).
- Corporate proxy: set
http_proxy, https_proxy, and no_proxy in env/<name>/global_config.yaml before installing. See Configuration. - Time synchronization (multi-node only): an NTP server (
chrony or systemd-timesyncd) must be running and synced on every node before install. Clock skew causes intermittent etcd failures, TLS errors, and token authentication issues that are hard to diagnose.
Tooling
./es_auto_installer.sh configure installs the following pinned versions into /usr/local/bin. Run it once per machine before anything else. If these tools are already installed at compatible versions, configure skips them.
| Tool | Version | Notes |
|---|
| Python | ≥ 3.11 | Also installs the matching python<ver>-venv package |
| yq | v4.53.2 | mikefarah/yq — YAML processor used by the installer |
| kubectl | v1.34.3 | |
| helm | v3.20.2 | |
git must already be installed before running configure. It is preinstalled on most Ubuntu and RHEL images. The installer uses it to clone external repos on first run.
Credentials
- Hugging Face token — required only for gated models (Llama, Mistral, Gemma, etc.). Get one at huggingface.co/settings/tokens (free account). Export as
HF_TOKEN=hf_... before deploying those models. - Keycloak admin password — auto-generated if not set. Override by exporting
KEYCLOAK_ADMIN_PASSWORD before install. - Grafana admin password — auto-generated if not set. Override by exporting
GRAFANA_ADMIN_PASSWORD before install.
What configure does not do
- Install
git — must already be present - Configure proxy settings — edit
env/<name>/global_config.yaml yourself - Set up SSH keys — you must distribute them to target nodes for multi-node installs
Related Docs