Home › Servers & Systems › NVIDIA HGX H100/H200
NVIDIA HGX H100/H200
System
GPU Baseboard
Shipping
The 8-GPU baseboard that defined the standard AI server.
You're looking at a system record: an assembled machine — a server or a whole rack — with what's inside it and where it's deployed.
HGX is the NVLink-switched baseboard OEMs build servers around: eight SXM GPUs, four NVSwitches, ~5–8 kW per board. Nearly every 'H100 server' from Supermicro, Dell, Quanta or Foxconn is an HGX plus CPUs, NICs and power — the unit the 2023–24 build-out was counted in.
Specifications & key facts
Each figure is labeled by basis: vendor spec, vendor claim, reported, estimate, or official. "Not publicly disclosed" means exactly that — we never invent values.
Overview
Configuration
8 × H100/H200 SXM + 4 × NVSwitchVendor spec
GPU-to-GPU
900 GB/s NVLink 4, all-to-allVendor spec
Networking
Typically 8 × 400G ConnectX-7 + BlueFieldVendor spec
Power
Typical server power
~10 kW air-cooledVendor spec
How this connects to the rest of the stack
This record's links at a glance — click a group to jump to it.
Designed by
Interconnect & networking
Cooling
Contains
Competes with
Sources
Record reviewed as of 2026-01-15 — see Methodology.