Platform Support (Intel SPR, GNR)

StarlingX OpenStack is validated on current-generation Intel® Xeon® Scalable processor platforms — Sapphire Rapids (SPR) and Granite Rapids (GNR). These microarchitectures deliver the compute density, memory bandwidth, and built-in acceleration that cloud and NFV workloads demand.

The sections below summarize what each platform generation brings to StarlingX OpenStack deployments and the capabilities they share.

Sapphire Rapids

4th Generation Intel® Xeon® Scalable processors (SPR) introduce hardware acceleration and I/O improvements that reduce reliance on discrete adapters and increase per-node throughput for StarlingX OpenStack workloads.

Key capabilities for cloud deployments:

  • DDR5 and optional HBM for higher memory bandwidth and lower latency

  • PCIe 5.0, doubling per-lane bandwidth for high-throughput NICs and NVMe storage

  • Integrated Intel® QAT for hardware-accelerated compression and cryptographic offload, exposed to VMs through SR-IOV

  • Intel® AMX for CPU-native AI inference without a discrete accelerator

  • CXL 1.1 support for memory expansion and coherent device-attached memory

Granite Rapids

5th Generation Intel® Xeon® Scalable processors (GNR) build on the SPR foundation with higher core density and upgraded accelerators, increasing the workload capacity of each node.

Key improvements over SPR:

  • Higher core counts and larger last-level cache for improved VM consolidation and reduced memory latency

  • DDR5 and PCIe 5.0 retained for infrastructure compatibility

  • Enhanced integrated QAT and Intel® DSA engines for higher cryptographic, compression, and data-movement throughput

  • CXL 2.0 support, enabling memory pooling and switching for flexible resource allocation across nodes

Common Platform Capabilities

Both SPR and GNR platforms share a set of capabilities that StarlingX OpenStack leverages across all host roles:

  • Intel® Virtualization Technology (|VT-x| and |VT-d|) — Required for KVM-based Nova compute hosts. VT-d enables SR-IOV and PCI passthrough device assignment to VMs.

  • Hyper-Threading — Simultaneous multithreading is available on both platforms. Disable it for workloads that require deterministic, low-jitter performance.

  • Intel® Turbo Boost — Recommended enabled for maximum single-thread and burst performance on controller and compute hosts.

  • NUMA topology — StarlingX OpenStack exposes the multi-socket NUMA layout to Nova, allowing the scheduler to pin instances and their memory to specific NUMA nodes for predictable memory-access latency.

Note

Configure BIOS settings for virtualization (:abbr`VT-x (Virtualization Technology for IA-32 and Intel 64 Processors)`, VT-d), power profile (maximum performance), and C-state control (disabled) before deploying StarlingX OpenStack on SPR or GNR hardware. For the full list of required settings, see Hardware Requirements.