Deploy: Private Appliance¶
Release binding: v0.1.0-enterprise-ready · 2026-05-19 · Signed release bundle, offline license
HX-SDP is currently available through a Private Appliance or scoped pilot. Contact HolonomiX to agree the target, release bundle, validation criteria, and commercial terms. The Private Appliance packages the signed runtime for customer-controlled infrastructure. Every path below starts from the supplied signed release bundle and ends with a validation report; the release claim is valid only after the target report passes without live patching. Use only the package and target included in your agreed release.
Supported targets¶
| Target | Status |
|---|---|
| VMware OVA on supported NVIDIA GPU hosts | Supported |
| KVM, Proxmox, and OpenStack QCOW2 on supported NVIDIA GPU hosts | Supported (OpenStack QCOW2 may remain gated until OpenStack GPU validation is present in the release dossier) |
| Air-gapped OCI Runtime Kit on supported NVIDIA GPU Kubernetes nodes | Supported |
Not supported: unmanaged consumer Windows hosts, non-NVIDIA GPU paths, public unauthenticated self-service, customer-modified base images, and unvalidated hypervisor or container runtime combinations.
Before any target: verify the bundle¶
Verify the signed release bundle first. See Release Verification.
sha256sum -c manifests/artifacts.sha256
hx-sdp release verify-signature \
--file manifests/artifacts.sha256 \
--public-key manifests/release-pubkey.pem \
--signature manifests/artifacts.sha256.sig
VMware vSphere (OVA)¶
For VMware ESXi or vCenter clusters with NVIDIA vGPU or PCIe passthrough.
- Verify the release bundle.
- Import
images/hx-sdp-vsphere.ovainto vSphere. - Attach the required GPU profile or passthrough device.
- Attach persistent storage for
/var/lib/hx-sdpif the sizing plan requires it. - Boot the VM and read the first-boot console report.
Validate:
KVM and Proxmox (QCOW2)¶
For KVM, Proxmox, or equivalent private virtualization with NVIDIA PCIe passthrough. IOMMU must be enabled on the host and the GPU bound for passthrough.
qemu-img info images/hx-sdp-kvm.qcow2
virt-install \
--name hx-sdp \
--memory 32768 \
--vcpus 16 \
--disk path=images/hx-sdp-kvm.qcow2,format=qcow2,bus=virtio \
--import \
--os-variant ubuntu24.04 \
--network bridge=br0,model=virtio \
--graphics none
Attach the GPU passthrough device through your host policy before workload validation, then run hx-sdp gpu-preflight and hx-sdp validate --mode smoke.
OpenStack (QCOW2)¶
For OpenStack clouds with NVIDIA passthrough or vGPU scheduling already configured.
openstack image create hx-sdp-openstack \
--disk-format qcow2 \
--container-format bare \
--file images/hx-sdp-openstack.qcow2
openstack server create hx-sdp \
--image hx-sdp-openstack \
--flavor <gpu-flavor> \
--network <trusted-network> \
--key-name <operator-key>
First boot does not depend on EC2 metadata. Optional static network material may be injected through cloud-init or config drive according to local OpenStack policy.
Air-gapped OCI Runtime Kit (Kubernetes)¶
For containerized deployments in disconnected Docker, containerd, or Kubernetes environments. The kit contains the engine, gate, and Redis-profile images, a Compose file for single-host GPU deployments, a Helm chart, and its own signed manifest.
Verify the kit:
Load images:
zstd -dc containers/hx-engine.tar.zst | docker load
zstd -dc containers/hx-gate.tar.zst | docker load
zstd -dc containers/redis-profile.tar.zst | docker load
Kubernetes install:
kubectl create namespace hx-sdp
kubectl -n hx-sdp create secret generic hx-sdp-engine-service-key \
--from-literal=value="$HX_GATE_ENGINE_SERVICE_KEY"
kubectl -n hx-sdp create secret generic hx-sdp-tenants \
--from-file=tenants.json=tenants.json
helm upgrade --install hx-sdp ./helm/hx-sdp -n hx-sdp
Validate:
kubectl -n hx-sdp rollout status deploy/hx-sdp-engine
kubectl -n hx-sdp rollout status deploy/hx-sdp-gate
kubectl -n hx-sdp exec deploy/hx-sdp-engine -- hx-sdp gpu-preflight
The complete offline procedure, including ctr -n k8s.io images import and license and TLS secrets, ships in the kit as AIRGAP_LOAD.md. Kubernetes nodes require the NVIDIA Container Toolkit and a GPU device plugin.
After deployment¶
- Install and verify the license: License
- Replace the bootstrap TLS material: Security Model
- Size the corpus against hardware: Hardware and Sizing