+86 400-090-8865
CN
  • Products
  • Solutions
  • Successful Stories
  • Company

Successful Stories

YRCloudFile Empowers iFLYTEK’s AI Cluster Computing

  • IndustryAI, LLM
  • Company ProfileiFLYTEK Co., Ltd. is a listed company in the Asia-Pacific Region focusing on intelligent speech and technology.
  • Use CasesFaster AI Model Training

The Challenge

  • Exponential Data Growth

    Data is growing at an exponential rate, storage systems need to be both efficient and scalable.

  • Sustained High Performance

    The storage system must handle files in different sizes, including both small and large files, while meeting mixed read/write performance requirements. Additionally, it should maintain stable high performance with a large number of files.

  • Variable Requirements

    In each stage of large model training, the requirements for data access performance vary.

The Solution

iFLYTEK uses YRCloudFile to power high-performance data operations for AI model training and inference.YRCloudFile supports both small and large file random read/write operations, featuring high IOPS and low latency read/write characteristics. It perfectly matches GPU computing server performance, supports concurrent access by thousands of high-performance computing nodes, and facilitates rapid iteration of algorithm models, effectively ensuring the data processing efficiency for the business unit.

Outcomes with YRCloudFile

  • Performance Improvements

    Stores nearly 10 billion audio, video, and image files for training, and a single cluster peak bandwidth is over 30+GB/s.

  • Enhance Training Efficiency

    YRCloudFile features high bandwidth and low latency, ensuring GPU and other computing servers operate at full capacity, significantly reducing single training times.

  • Scalable Storage

    Storage scale is flexibly built to accommodate data growth, with total storage nodes exceeding several hundred and an annual growth rate of over 300%.

  • Large-scale Training Platform

    Multiple YRCloudFile storage clusters have been deployed, creating a core training platform with a scale of hundreds of petabytes.

  • Cost Reduction

    The intelligent tiering feature enables more efficient access to hot data and smarter management of cold data, effectively reducing IT Total Cost of Ownership (TCO).

Ready to power your AI & HPC workflows?

Get Started