hccl-test

Benchmark HCCL operations like AllReduce and AllGather across Ascend NPUs with MPI orchestration.

156|53|Updated Feb 14, 2026
One-click install
npx skills add https://github.com/ascend-ai-coding/awesome-ascend-skills --skill hccl-test
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: hccl-test
Source: https://github.com/ascend-ai-coding/awesome-ascend-skills/tree/main/hccl-test
Command: npx skills add https://github.com/ascend-ai-coding/awesome-ascend-skills --skill hccl-test

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve?

HCCL performance testing for Ascend NPU clusters is complex and time-consuming to set up and validate; this Skill provides a repeatable framework to benchmark distributed communication, verify HCCL functionality, and standardize production testing workflows.

Core Features & Use Cases

  • Multi-node benchmarking for HCCL operations (AllReduce, AllGather, AlltoAll, ReduceScatter, etc.) with production-ready workflows.
  • Comprehensive pre-flight checks (SSH access, CANN version consistency, NPU health, and MPI environment) to ensure reliable testing.
  • Automation tooling including hostfile generation, test orchestration, result parsing, and troubleshooting guidance.

Quick Start

Run the pre-test check and initiate a multi-node HCCL test using the provided scripts with your node list and NPUs.

Frequently Asked Questions about hccl-test

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I benchmark HCCL performance across Ascend NPUs in a distributed cluster?▼

To benchmark HCCL performance across Ascend NPUs, use the provided scripts to orchestrate MPI operations like AllReduce and AllGather. The framework handles hostfile creation, MPI invocation, and result parsing for multi-node environments.

What pre-flight checks are needed before running distributed HCCL tests?▼

Required pre-flight checks for distributed HCCL tests include verifying SSH access, CANN version consistency, NPU health, and MPI environment readiness. These checks ensure reliable testing and standardize production workflows.

Can I automate hostfile generation and MPI invocation for multi-node Ascend testing?▼

Yes, you can automate hostfile generation and MPI invocation for multi-node Ascend testing. The included scripts coordinate test orchestration, result parsing, and error handling to satisfy end-to-end production requirements.

What is the best way to validate Huawei HCCL functionality for production readiness?▼

The best way to validate Huawei HCCL functionality for production is using a repeatable framework that benchmarks distributed communication and standardizes testing workflows with comprehensive pre-flight validation and automated result parsing.

Does this framework support testing operations like AllReduce and AllGather on Ascend NPUs?▼

Yes, the framework supports testing HCCL operations including AllReduce, AllGather, AlltoAll, and ReduceScatter on Ascend NPUs. It provides production-ready workflows for benchmarking these distributed communication operations.

Why do I need MPI orchestration for HCCL benchmarking in distributed clusters?▼

MPI orchestration is needed for HCCL benchmarking because it coordinates multi-node environments across Ascend NPUs, enabling synchronized execution of distributed communication tests and reliable result aggregation.