Combining SGX and TDX on the OpenMetal Large v5

Q: Can I combine Intel SGX enclaves with TDX guest VMs on the OpenMetal Large v5?

Yes — SGX and TDX work in parallel on the Xeon 6517P; the typical pattern is to run a workload as a TDX-protected guest VM and then run high-value secrets (keys, signing operations) inside SGX enclaves nested within that guest.

Explore bare metal dedicated servers

The two technologies solve different isolation problems. TDX isolates entire guest VMs from the host: the OpenMetal-managed hypervisor and host OS cannot read TD memory, so a compromised host operator or a malicious privileged process on the host cannot exfiltrate guest state. SGX isolates specific regions of application memory inside a process: even another application running as the same OS user, or a kernel module on the same OS, cannot read SGX enclave memory. They are complementary boundaries, not competing ones.

Most regulated production deployments combine both. The TD provides the host-level boundary for the whole workload — a tenant VM running banking analytics, a healthcare claims-processing server, a sovereign cloud workload — and SGX enclaves inside the TD provide a second boundary for the narrow operations that warrant it: key derivation, certificate signing, software-HSM emulation, payment HMAC generation, and protocol-level secret handling. The combined pattern is appropriate when both the infrastructure operator and other in-VM processes are part of the threat model.

On the Large v5, every server ships with Intel SGX enabled by default and up to 128 GB of EPC (Enclave Page Cache) per Xeon 6517P, for 256 GB EPC per dual-socket server. TDX activation requires the 1 TB RAM upgrade per OpenMetal policy; once active, TDX and SGX coexist with no special configuration. Guest VMs running TDX-aware kernels and SGX-aware runtime libraries (Open Enclave SDK, Intel SGX SDK, Gramine, EGo) can launch enclaves from inside TDs without losing TD isolation. See the Enabling Intel SGX and TDX guide page for implementation details.