Customer case studies

Operator domestic GPU cluster

Compare ZCube with Clos and validate performance gains in collective communication, model training, and multi-tenant end-to-end inference.

A telecom operatorDomestic GPU clusterIndependent validation

Project information

Project overview

Environment, workload, and validation approach.

Customer type: A telecom operator
Project stage: Independent validation
Hardware ecosystem: Domestic GPU cluster
Workload: Multi-tenant end-to-end inference
Comparison baseline: Clos architecture
Case type: Quantitative validation case

Background and challenges

Project background

Operator scenarios need to host multi-tenant workloads on domestic GPU clusters while balancing collective communication, model-training throughput, and end-to-end inference latency.

Validation or delivery approach

Use a Clos architecture without rail optimization as the comparison baseline
Run end-to-end inference tests in a domestic GPU multi-tenant environment
Record collective communication, model-training throughput, and inference latency separately

Results

The following metrics all use a Clos architecture without rail optimization as the baseline, and apply only to the domestic GPU and multi-tenant test environment described in this case.

119%

Collective communication performance increase

Compared with Clos without rail optimization.

20.1%

Model-training throughput increase

Comparison result for domestic GPU model training.

11.5%

Inference latency reduction

Latency improvement in multi-tenant end-to-end inference.

33%

CapEx reduction

The flat architecture reduced network hardware spend.

Value summary

Case value

In multi-tenant inference, ZCube showed a clear performance advantage over Clos without rail optimization.

Validates performance isolation and end-to-end gains in multi-tenant scenarios
Covers collective communication, large-model training, and inference together
Provides a reference path for operator domestic compute cluster construction

Customer case studies

Other customer case studies

See more ZCube validation and production practice across compute environments, workloads, and project stages.

Domestic large-model companyThousand-GPU NVIDIA inference clusterProduction delivery

Large-model company NVIDIA thousand-GPU inference production cluster

Using a coding inference service and comparing against traditional ROFT, this case validates that ZCube unlocks more effective inference compute through a flat fabric and more balanced network communication.

15%

Average GPU inference throughput increase

40.6%

Time to first token (TTFT) P99 reduction

33%

Network hardware cost reduction

View case study

Training groundDomestic GPU inference clusterIndependent validation

Training-ground domestic GPU inference cluster

In a domestic AI computing ecosystem, compare ZCube with traditional Clos across collective communication, training, and inference.

14.5%

All-to-All throughput increase

22%–30%

TTFT P99 reduction

7%–10%

Inference throughput increase

View case study

Domestic accelerator companyDomestic thousand-GPU clusterProduction delivery

Domestic accelerator company thousand-GPU production cluster

A domestic thousand-GPU cluster has been live and stable for months, continuously carrying real traffic and serving real users.

Thousand-GPU scale

Domestic compute cluster deployed

Months

Stable production operation

Real users

Continuous service

View case study

Start with the network

Make the next AI data center more efficient, starting with the network

Whether you are building a new cluster, upgrading an existing network, tuning performance, or deploying intelligent operations, we can assess how the network can unlock more effective compute.