Intel® Ethernet 800 Series
Linux Performance Tuning Guide
Sub-NUMA Clustering
6th Gen Intel Xeon processors can divide the Last Level Cache (LLC) across a socket into smaller, independent clusters using Sub-NUMA Clustering (SNC). By using memory address checks to group CPU cores with their closest local cache slices, SNC creates distinct sub-NUMA domains within a single physical processor to keep memory access local and reduce latency.
- SNC2 (2-way Sub-NUMA Clustering): Supported on Intel Xeon 6 P-Core processors (XCC/HCC platforms). Divides each physical processor into 2 independent NUMA nodes visible to software.
- SNC3 (3-way Sub-NUMA Clustering): Supported on Intel Xeon 6 P-Core processors (high-density AP platforms). Divides each physical processor into 3 independent NUMA nodes
- Auto: Default setting; automatically matches the detected SNC configuration
- Disable: Disables SNC and defaults to UMA (Uniform Memory Access) mode
- Enable: Enables NUMA interleave mode for the active SNC configuration
Reassigning Network Interfaces Across NUMA Domains in the same socket:
- Set the target NUMA node for each PCIe device interface:
echo <NUMA_NODE> > /sys/bus/pci/devices/<PCI_ADDRESS>/numa_node Example:
echo 1 > /sys/bus/pci/devices/0000:65:00.0/numa_node echo 1 > /sys/bus/pci/devices/0000:65:00.1/numa_node - Verify the change using lspci
lspci -s <PCI_ADDRESS> -vv | grep NUMA - Reload the ICE drivers.