EXAScaler Lustre File System

The EXAScaler Lustre File System is designed to enhance the performance of AI, HPC, and data-intensive applications by providing a high-performance, parallel file system. It is known for its ability to deliver extreme throughput and minimal latency, making it ideal for large-scale operations such as AI training and HPC simulations. This system is exclusively used by NVIDIA for their internal clusters, ensuring that it unlocks the full potential of compute investments.

Key features include:

• Performance at Scale: Achieves up to 30 times the performance for single-threaded workloads and supports multi-petabyte scale.
• Efficient Data Management: Reduces power and cooling requirements by 15 times.
• Enterprise-Class Simplicity: Offers client-side data compression without degrading performance, supports online upgrades, and provides enhanced monitoring and health reporting.
• Secure and Reliable: Features native multi-tenancy to support multiple worldwide applications and users seamlessly.

EXAScaler maximizes the performance of DGX SuperPODs, as highlighted by NVIDIA's endorsement. Scaleway, a leading NVIDIA Cloud Provider in Europe, utilizes EXAScaler to deliver rapid, secure, and eco-friendly AI cloud services, powering innovations like the Nabu 2023 DGX H100 cluster.

By eliminating delays with terabytes per second throughput, ultra-fast checkpointing, and real-time AI access, EXAScaler ensures that GPUs are fully utilized, preventing idle time and enhancing return on investment. This results in faster AI breakthroughs and more efficient training processes.

For those looking to optimize their AI and HPC infrastructure, EXAScaler represents the end of hardware-centric storage design, unleashing the full potential of QLC for enterprise AI and providing maximum performance with TLC flash when needed.