DeepSeek, the AI company from China, has just released the open-source Fire-Flyer File System (3FS) – a technology designed to run AI models faster and more resource-efficiently. This move comes as Western organizations express skepticism regarding the company’s transparency. DeepSeek’s “Open Source Week” event has captured the attention of the international AI community, with many experts highly praising the technology released by the company.
DeepSeek Opens Source 3FS – An AI Acceleration File System
During the “Open Source Week” event, DeepSeek open-sourced five key software projects, most notably the Fire-Flyer File System (3FS). This is an AI processing optimization technology that DeepSeek uses internally for training and deploying AI models.
3FS is designed to leverage the performance of modern SSDs and RDMA networks, helping to optimize AI tasks that require processing massive amounts of data. This file system operates without read caching and prioritizes random read requests—a crucial feature when GPUs are constantly accessing data from multiple different servers.
🚀 Day 5 of #OpenSourceWeek: 3FS, Thruster for All DeepSeek Data Access
Fire-Flyer File System (3FS) – a parallel file system that utilizes the full bandwidth of modern SSDs and RDMA networks.
⚡ 6.6 TiB/s aggregate read throughput in a 180-node cluster
⚡ 3.66 TiB/min…— DeepSeek (@deepseek_ai) February 28, 2025
According to Tom’s Hardware, 3FS is capable of combining the bandwidth of thousands of SSDs and the networks of hundreds of storage nodes, helping to simplify application code and leverage standard storage APIs. When running on a 180-node server cluster, the system can achieve read speeds of up to 6.6 TiB/s, while in the GraySort benchmark test, it reached 3.66 TiB/min on 25 nodes.
The AI company Perspective AI has called these specifications a “new milestone” and noted that 3FS could revolutionize big data processing, especially in the fields of AI, research, and data science.
DeepSeek Increases Transparency, Facing Global Competition
Last year, DeepSeek released Fire-Flyer 2, a high-performance AI architecture. Thanks to the combination of 3FS, HaiScale, and other software technologies, DeepSeek claims to achieve 80% of the performance of Nvidia DGX-A100 servers but at only 50% of the cost and using 40% less energy.
The Fire-Flyer 2 system includes:
- 180 storage nodes, each node with 16 16TB SSDs
- 2 NUC 200Gbps
- 10,000 Nvidia A100 GPUs connected via PCIe
The “Open Source Week” event was organized by DeepSeek to increase transparency following criticism from the West regarding the company’s perceived lack of openness and reliability. DeepSeek has released the source code for several important software products such as FlashMLA, DeepEP, and DeepGEMM to drive community innovation and compete with major AI companies worldwide.
This move not only helps DeepSeek demonstrate its capabilities but also positions the company as a significant player in the open-source AI market. By optimizing costs and resources in AI training, DeepSeek could become a formidable competitor to giants like OpenAI, Nvidia, and Google DeepMind.
