hccl-test

Name: hccl-test
Availability: InStock
Author: ascend-ai-coding

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

156|53|Updated Feb 14, 2026

Search Tags:#testing #distributed #benchmark #mpi #hccl #huawei #npus

Ready to run

npx skills add https://github.com/ascend-ai-coding/awesome-ascend-skills --skill hccl-test

💬 Prompt for ChatGPT / Claude Chat▼

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.