A single physical server
machine in the data center. Each host
has its associated compute resources such as accelerators. The number and
configuration of these compute resources depend on the machine family.
Virtual machine (VM) instances are provisioned on top of a physical host.
A collection of blocks
interconnected by a high-speed network fabric. Each cluster is globally unique.
For A4X, A4, and A3 Ultra machines, a cluster provides a common, non-blocking
network fabric for your blocks of accelerator capacity. Within a cluster,
the east to west networking is non-blocking for the entire collection of blocks.
A resource request that
allocates your accelerator
resources physically close to each other to minimize network hops and optimize
for the lowest latency.
A network fabric provides
high-bandwidth, low-latency connectivity across all blocks and Google Cloud
services in a cluster. Jupiter is Google's data center network architecture
that leverages software-defined networking and optical circuit switches to
evolve the network and optimize its performance.
[[["Easy to understand","easyToUnderstand","thumb-up"],["Solved my problem","solvedMyProblem","thumb-up"],["Other","otherUp","thumb-up"]],[["Hard to understand","hardToUnderstand","thumb-down"],["Incorrect information or sample code","incorrectInformationOrSampleCode","thumb-down"],["Missing the information/samples I need","missingTheInformationSamplesINeed","thumb-down"],["Other","otherDown","thumb-down"]],["Last updated 2025-09-03 UTC."],[[["\u003cp\u003eA Node/Host is a single physical server within the data center, equipped with compute resources and on which Virtual Machines are provisioned.\u003c/p\u003e\n"],["\u003cp\u003eA Sub-block is a group of hosts connected by a network fabric (e.g., Jupiter), offering low latency and consistent bandwidth.\u003c/p\u003e\n"],["\u003cp\u003eA Block is comprised of multiple Sub-blocks interconnected through a distributed network fabric layer.\u003c/p\u003e\n"],["\u003cp\u003eA Cluster offers a non-blocking network fabric that covers the Blocks and ensures that the east to west networking is non-blocking across the entire set of blocks.\u003c/p\u003e\n"],["\u003cp\u003eNetwork fabric is the data center network layer that supports high-bandwidth and low latency connectivity across resources.\u003c/p\u003e\n"]]],[],null,["The following terminology is often used when working with\nCluster Director features.\n\n\u003cbr /\u003e\n\nNode or host[link](#node) A single physical server machine in the data center. Each host has its associated compute resources such as accelerators. The number and configuration of these compute resources depend on the machine family. Virtual machine (VM) instances are provisioned on top of a physical host.\n\n\u003cbr /\u003e\n\n\u003cbr /\u003e\n\nSub-block[link](#sub-block) A group of hosts and associated connectivity hardware that are on a single rack.\n\n\u003cbr /\u003e\n\n\u003cbr /\u003e\n\nBlock[link](#block) A collection of sub-blocks.\n\n\u003cbr /\u003e\n\n\u003cbr /\u003e\n\nCluster[link](#cluster) A collection of blocks interconnected by a high-speed network fabric. Each cluster is globally unique. For A4X, A4, and A3 Ultra machines, a cluster provides a common, non-blocking network fabric for your blocks of accelerator capacity. Within a cluster, the east to west networking is non-blocking for the entire collection of blocks.\n\n\u003cbr /\u003e\n\n\u003cbr /\u003e\n\nDense deployment[link](#dense-deployment) A resource request that allocates your accelerator resources physically close to each other to minimize network hops and optimize for the lowest latency.\n\n\u003cbr /\u003e\n\n\u003cbr /\u003e\n\nNetwork fabric[link](#network-fabric) A network fabric provides high-bandwidth, low-latency connectivity across all blocks and Google Cloud services in a cluster. Jupiter is Google's data center network architecture that leverages software-defined networking and optical circuit switches to evolve the network and optimize its performance.\n\n\u003cbr /\u003e"]]