Understanding Multi-GPU Topologies Within a Single Host; Architecting AI Infrastructure Series – Part 10
– Frank Denneman
Explains why distributed inference turns GPU communication into part of the critical path and why topology-aware scheduling is required when models span multiple GPUs.
vSphere Express (128 cores, ESX 9 and vCenter 9, no DRS or vSAN) compared with vSphere Standard 8, which EEA customers can still buy until 31 October 2027 with Extended Support until October 2028: features, packaging, price, support window, upgrade path and what a small or mid-size EEA customer should do. As of November 2026.
vdan.cz/tools is a Progressive Web App: add it to the home screen on Android (Chrome, Firefox) or iPhone (Safari) and every tool opens full screen like a native app. Steps, what you get, and how it works.
21 ESXi advanced and kernel settings William Lam recommends for labs and consumer hardware, each with the exact esxcli command, the version it was written for, what changed since, and a deep link to the vDan ESXi settings explorer.
