Disabling Memory Interleaving on AMD EPYC 9474F-based hosts can result in a discrepancy where the reported ESXi memory capacity is less than the actual physical RAM.
While the physical host is populated with 24 x 96GB memory modules, the full capacity may not be exposed to ESXi despite the DIMMs (Dual In-line Memory Modules) being correctly identified via the "localcli hardware ipmi sdr list -p -i -n all" command. (KB 345401).
During the ESXi host boot process, the following message is observed:
boot:
TSC: ######## cpu0:1)WARNING: NUMA: 2385: A significant memory (DRAM) imbalance (more than 30 percent) was detected between NUMA nodes 0(1133051 MB) and 1(786389 MB). This may impact performance
VMware vSphere ESXi 8.0.x
To resolve this issue, contact the hardware vendor to ensure that memory modules are balanced evenly across processor sockets and channels.
Note: Enabling "Memory Interleaving" may allow ESXi to recognize the full memory capacity; however, if Node Interleaving or Memory Interleaving is enabled, NUMA will be disabled, which may result in sub-optimal performance for certain workloads.
On some systems BIOS settings for node interleaving (also known as interleaved memory) determine whether the system behaves like a NUMA system or like a uniform memory accessing (UMA) system. If node interleaving is deactivated, ESXi detects the system as NUMA and applies NUMA optimizations. If node interleaving is activated, ESXi does not detect the system as NUMA. For more information, refer to your server’s documentation. ( Details in Non-Uniform Memory Access (NUMA) )
Recommended Settings for Node Interleaving: Disabled ( Details in Application on Virtual Machine runs slower than expected)
Resource Management in NUMA Architectures
ESX/ESXi host console reports the message: "Significant imbalance between NUMA nodes detected"