B200 System Topology Explorer

A high-performance server architecture overview.

Connectivity Matrix

This matrix shows the connection type between each component in the system. Refer to the legend below for definitions.

GPU0 GPU1 GPU2 GPU3 GPU4 GPU5 GPU6 GPU7 NIC0 NIC1 NIC2 NIC3
GPU0 X NV18 NV18 NV18 NV18 NV18 NV18 NV18 PIX PIX NODE NODE
GPU1 NV18 X NV18 NV18 NV18 NV18 NV18 NV18 PIX PIX NODE NODE
GPU2 NV18 NV18 X NV18 NV18 NV18 NV18 NV18 NODE NODE PIX PIX
GPU3 NV18 NV18 NV18 X NV18 NV18 NV18 NV18 NODE NODE PIX PIX
GPU4 NV18 NV18 NV18 NV18 X NV18 NV18 NV18 SYS SYS SYS SYS
GPU5 NV18 NV18 NV18 NV18 NV18 X NV18 NV18 SYS SYS SYS SYS
GPU6 NV18 NV18 NV18 NV18 NV18 NV18 X NV18 SYS SYS SYS SYS
GPU7 NV18 NV18 NV18 NV18 NV18 NV18 NV18 X SYS SYS SYS SYS

Connection Legend

NV18 (NVLink)

Highest-speed direct GPU-to-GPU connection.

PIX (PCIe Local)

Fastest GPU-to-NIC path on the same local fabric.

NODE (PCIe NUMA)

Fast path to a NIC on the same NUMA node.

SYS (System Cross-Connect)

Slowest path; data crosses between CPUs.

System Architecture

The server is built around two main CPU complexes, called NUMA nodes. This diagram shows how GPUs and Network Interface Cards (NICs) are grouped within these nodes. All GPUs are interconnected via a high-speed NVLink mesh.

NUMA Node 0

CPU Cores: 0-55, 112-167

GPU0
GPU1
GPU2
GPU3
NIC0
NIC1
NIC2
NIC3

NVLink & System Fabric

NUMA Node 1

CPU Cores: 56-111, 168-223

GPU4
GPU5
GPU6
GPU7
NIC4
NIC5
NIC6
NIC7